Advanced Technologies and Communication Solutions for Internet of Things

Transcription

1 International Journal of Distributed Sensor Networks Advanced Technologies and Communication Solutions for Internet of Things Guest Editors: Young-Sik Jeong, Naveen Chilamkurti, and Luis Javier García Villalba

2 Advanced Technologies and Communication Solutions for Internet of Things

3 International Journal of Distributed Sensor Networks Advanced Technologies and Communication Solutions for Internet of Things Guest Editors: Young-Sik Jeong, Naveen Chilamkurti, andluisjaviergarcía Villalba

4 Copyright 2014 Hindawi Publishing Corporation. All rights reserved. This is a special issue published in International Journal of Distributed Sensor Networks. All articles are open access articles distributed under the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.

5 Editorial Board Miguel Acevedo, USA Sanghyun Ahn, Korea Ana Alejos, Spain Mohammod Ali, USA Jamal N. Al-Karaki, Jordan Habib M. Ammari, USA C. Anagnostopoulos, Greece Masoud Ardakani, Canada Muhammad Asim, UK Stefano Avallone, Italy Javier Bajo, Spain N. Balakrishnan, Canada Prabir Barooah, USA Paolo Bellavista, Italy Roc Berenguer, Spain Juan A. Besada, Spain Alessandro Bogliolo, Italy Richard R. Brooks, USA James Brusey, UK Erik Buchmann, Germany Carlos T. Calafate, Spain Tiziana Calamoneri, Italy Juan C. Cano, Spain Xianghui Cao, USA Jian-Nong Cao, Hong Kong J. P. Carmo, Portugal Jesús Carretero, Spain Luca Catarinucci, Italy Henry Chan, Hong Kong Chih-Yung Chang, Taiwan Yao-Jen Chang, Taiwan Periklis Chatzimisios, Greece Ai Chen, China Hanhua Chen, China Peng Cheng, China Jinsung Cho, Korea Thomas Wook Choi, Korea Hyunseung Choo, Korea Kim-Kwang R. Choo, Australia Chengfu Chou, Taiwan Chi-Yin Chow, Hong Kong W.-Y.Chung,RepublicofKorea Tae-Sun Chung, Korea Mauro Conti, Italy Xunxue Cui, China Iñigo Cuiñas, Spain Alfredo Cuzzocrea, Italy Dinesh Datla, USA Amitava Datta, Australia Danilo De Donno, Italy Luca De Nardis, Italy Ilker Demirkol, Spain Der-Jiunn Deng, Taiwan Longjun Dong, China Chyi-Ren Dow, Taiwan G. P. Efthymoglou, Greece Frank Ehlers, Italy M. Erol-Kantarci, Canada Michael Farmer, USA Gianluigi Ferrari, Italy Silvia Ferrari, USA Giancarlo Fortino, Italy Luca Foschini, Italy David Galindo, France Deyun Gao, China Weihua Gao, USA Quanbo Ge, China Athanasios Gkelias, UK Iqbal Gondal, Australia Nikos Grammalidis, Greece J. Gubbi, Australia Cagri Gungor, Turkey Song Guo, Japan Andrei Gurtov, Finland Mohamed A. Haleem, USA K. Han, Republic of Korea Qi Han, USA Z. Hanzalek, Czech Republic Wenbo He, Canada Tian He, USA Junyoung Heo, Korea Feng Hong, Japan Zujun Hou, Singapore Jiangping Hu, China Haiping Huang, China Jiun-Long Huang, Taiwan Yung-Fa Huang, Taiwan Xinming Huang, USA Chin-Tser Huang, USA Wei Huangfu, China Mohamed Ibnkahla, Canada Lillykutty Jacob, India Won-Suk Jang, Korea Yingtao Jiang, USA Haifeng Jiang, China Hong-Bo Jiang, China Shengming Jiang, China Ning Jin, China Raja Jurdak, Australia I. Kamel, United Arab Emirates Li-Wei Kang, Taiwan Rajgopal Kannan, USA Gour C. Karmakar, Australia Jamil Y. Khan, Australia Sherif Khattab, Egypt S. Kim, Republic of Korea H. Kim, Republic of Korea Lisimachos Kondi, Greece Marwan Krunz, USA Gurhan Kucuk, Turkey S. S. Kumar, The Netherlands Kun-Chan Lan, Taiwan Yee W. Law, Australia Y.-K. Lee, Republic of Korea Yong Lee, USA S. Lee, Republic of Korea Seokcheon Lee, USA Joo-Ho Lee, Japan Kyung-Chang Lee, Korea JongHyup Lee, Korea Zan Li, China Shuai Li, USA Shijian Li, China Shancang Li, UK Zhen Li, China Ye Li, China Jing Liang, China Weifa Liang, Australia Yao Liang, USA Qilian Liang, USA I-En Liao, Taiwan Wen-Hwa Liao, Taiwan Jiun-Jian Liaw, Taiwan Alvin S. Lim, USA Kai Lin, China

6 Yaping Lin, China Zhigang Liu, China Wenyu Liu, China Ming Liu, China Donggang Liu, USA Yonghe Liu, USA Zhong Liu, China Hai Liu, Hong Kong Chuan-Ming Liu, Taiwan Leonardo Lizzi, France KennethJ.Loh,USA Jonathan Loo, UK J. López Riquelme, Spain Pascal Lorenz, France Chun-Shien Lu, Taiwan King-Shan Lui, Hong Kong Jun Luo, Singapore Juan Luo, China Yingchi Mao, China Yuxin Mao, China Álvaro Marco, Spain J. Martinez-de Dios, Spain N. Meratnia, The Netherlands Shabbir N. Merchant, India Lyudmila Mihaylova, UK Mihael Mohorcic, Slovenia José Molina, Spain Jose I. Moreno, Spain V. Muthukkumarasamy, Australia Kshirasagar Naik, Canada Kameswara Rao Namuduri, USA G. Nikolakopoulos, Sweden Alessandro Nordio, Italy Michael O Grady, Ireland Gregory O Hare, Ireland Giacomo Oliveri, Italy Saeed Olyaee, Iran Suat Ozdemir, Turkey S. Pack, Republic of Korea M. Palaniswami, Australia Meng-Shiuan Pan, Taiwan Ai-Chun Pang, Taiwan Seung-Jong J. Park, USA Soo-Hyun Park, Korea Miguel A. Patricio, Spain Wen-Chih Peng, Taiwan Janez Per, Slovenia Dirk Pesch, Ireland Shashi Phoha, USA Robert Plana, France C. Pomalaza-Ráez,Finland Antonio Puliafito, Italy Hairong Qi, USA Shaojie Qiao, China Meikang Qiu, USA Nageswara S.V. Rao, USA Md. Abdur Razzaque, Bangladesh Luca Reggiani, Italy Pedro P. Rodrigues, Portugal Joel J. Rodrigues, Portugal M. Saad, United Arab Emirates Sanat Sarangi, India Stefano Savazzi, Italy Marco Scarpa, Italy Arunabha Sen, USA Olivier Sentieys, France Salvatore Serrano, Italy Xiaojing Shen, China Zhong Shen, China Xingfa Shen, China Chin-Shiuh Shieh, Taiwan Minho Shin, Korea Louis Shue, Singapore Hichem Snoussi, France Guangming Song, China Antonino Staiano, Italy M. A. Tahir, Pakistan Tan-Hsu Tan, Taiwan Guozhen Tan, China Jindong Tan, USA Shaojie Tang, USA Bulent Tavli, Turkey Sameer S. Tilak, USA Chuan-Kang Ting, Taiwan Anthony Tzes, Greece F. Vasques, Portugal A. B. Waluyo, Australia Jianxin Wang, China Ju Wang, USA Honggang Wang, USA Yu Wang, USA Zhi Wang, China T. Wettergren, USA Ran Wolff, Israel Yuanming Wu, China Chase Qishi Wu, USA Wen-Jong Wu, Taiwan Jianshe Wu, China Na Xia, China Feng Xia, China Bin Xiao, Hong Kong Qin Xin, Faroe Islands Jianliang Xu, Hong Kong Yuan Xue, USA Chun J. Xue, Hong Kong Geng Yang, China Ting Yang, China Hong-Hsu Yen, Taiwan Li-Hsing Yen, Taiwan Seong-eun Yoo, Korea Ning Yu, China Changyuan Yu, Singapore T. Zahariadis, Greece Hongke Zhang, China Tianle Zhang, China Jiliang Zhou, China Yi-hua Zhu, China Xiaojun Zhu, China Yifeng Zhu, USA Yanmin Zhu, China T. L. Zhu, USA Qingxin Zhu, China Li Zhuo, China Shihong Zou, China

7 Contents Advanced Technologies and Communication Solutions for Internet of Things,Young-SikJeong, Naveen Chilamkurti, and Luis Javier García Villalba Volume 2014, Article ID , 3 pages A Carrier Class IoT Service Architecture Integrating IMS with SWE,DongliangXie,ChaoShang, Jinchao Chen, Yongfang Lai, and Chuanxiao Pang Volume 2014, Article ID , 11 pages SDN: Evolution and Opportunities in the Development IoT Applications, Ángel Leonardo Valdivieso Caraguay, Alberto Benito Peral, Lorena Isabel Barona López, and Luis Javier García Villalba Volume2014,ArticleID735142,10pages An Upper-Ontology-Based Approach for Automatic Construction of IOT Ontology,YuanXu, Chunhong Zhang, and Yang Ji Volume2014,ArticleID594782,17pages An Effective Routing Protocol with Guaranteed Route Preference for Mobile Ad Hoc Networks, Feng-Tsun Chien, Kuo-Guan Wu, Yu-Wei Chan, Min-Kuan Chang, and Yi-Sheng Su Volume 2014, Article ID , 18 pages Self-Organized Cognitive Sensor Networks: Distributed Channel Assignment for Pervasive Sensing, Li-Chuan Tseng, Feng-Tsun Chien, Abdelwaheb Marzouki, Ronald Y. Chang, Wei-Ho Chung, and ChingYao Huang Volume2014,ArticleID183090,10pages Broadcast Aggregation to Improve Quality of Service in Wireless Sensor Networks,EvyTroubleyn, Jeroen Hoebeke, Ingrid Moerman, and Piet Demeester Volume2014,ArticleID383678,12pages A Zone-Based Media Independent Information Service for IEEE Networks,Fábio Buiati, Luis Javier Garcia Villalba, Delfín Rupérez Cañas, Ana Lucila Sandoval Orozco, and Tai-hoon Kim Volume2014,ArticleID737218,6pages Implementation of Intelligent Electronic Acupuncture System Using Sensor Module,You-SikHong, Baek-Ki Kim, and Bong-Hwa Hong Volume2014,ArticleID238502,7pages Node Placement Analysis for Overlay Networks in IoT Applications,YuxinWan,JunweiCao,KangHe, Huaying Zhang, Peng Yu, Senjing Yao, and Keqin Li Volume 2014, Article ID , 12 pages Constructing the Green Campus within the Internet of Things Architecture, Hsing-I Wang Volume 2014, Article ID , 8 pages Ensuring Healthcare Services Provision: An Integrated Approach of Resident Contexts Extraction and Analysis via Smart Objects, Nan-Chen Hsieh, Lun-Ping Hung, Jong Hyuk Park, and Neil Y. Yen Volume2014,ArticleID481952,12pages Home Appliance Management System for Monitoring Digitized Devices Using Cloud Computing Technology in Ubiquitous Sensor Network Environment, Yun Cui, Myoungjin Kim, Yi Gu, Jong-jin Jung, and Hanku Lee Volume 2014, Article ID , 10 pages

8 GTrust: Group Extension for Trust Models in Distributed Systems, Robson de Oliveira Albuquerque, Luis Javier García Villalba, and Tai-Hoon Kim Volume2014,ArticleID872842,9pages Webit&NEU: An Embedded Device for the Internet of Things, Jialiang Wang, Hai Zhao, Jiuqiang Xu, and Yuanguo Bi Volume 2014, Article ID , 10 pages iotsilo: The Agent Service Platform Supporting Dynamic Behavior Assembly for Resolving the Heterogeneity of IoT, Euihyun Jung, Ilkwon Cho, and Sun Moo Kang Volume 2014, Article ID , 11 pages Distributed Risk Aversion Parameter Estimation for First-Price Auction in Sensor Networks,XinAn, Shuo Xu, Jiancheng Chen, and Yuan Zhang Volume 2013, Article ID , 9 pages An Efficient Adaptive Anticollision Algorithm Based on 4-Ary Pruning Query Tree, Wei Zhang, Yajun Guo, Xueming Tang, Guohua Cui, Longkai Wu, and Ying Mei Volume 2013, Article ID , 7 pages Improving Performance through REST Open API Grouping for Wireless Sensor Network,MinChoi, Young-Sik Jeong, and Jong Hyuk Park Volume 2013, Article ID , 13 pages

9 Hindawi Publishing Corporation International Journal of Distributed Sensor Networks Volume 2014, Article ID , 3 pages Editorial Advanced Technologies and Communication Solutions for Internet of Things Young-Sik Jeong, 1 Naveen Chilamkurti, 2 and Luis Javier García Villalba 3 1 Department of Multimedia Engineering, Dongguk University, Seoul , Republic of Korea 2 Department of Computer Science and Engineering, La Trobe University, Melbourne 3086, Australia 3 Department of Software Engineering and Artificial Intelligence (DISIA), School of Computer Science, Universidad Complutense de Madrid (UCM), Ciudad Universitaria, Madrid, Spain Correspondence should be addressed to Young-Sik Jeong; ysjeong1964@gmail.com and Luis Javier García Villalba; javiergv@fdi.ucm.es Received 19 March 2014; Accepted 19 March 2014; Published 28 May 2014 Copyright 2014 Young-Sik Jeong et al. This is an open access article distributed under the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited. 1. Introduction Internet of Things (IoT) is one of the important issues to describe several technologies and research disciplines that enable the IoT to reach out into the real world of physical objects. IoT is also a novel paradigm that is rapidly gaining in the scenario of Wireless Sensor Networks (WSN) and Wireless Telecommunications. The basic idea of this concept is the pervasive presence around our lifestyle of a variety of things or objects [1 3]. Tremendous advances in processing, wireless sensor networks, mobile communication, and systems/middleware technologies are leading to new paradigms and platforms for computing environment. There might be many issues to realize and provide intelligent services and much effort and enormous attention have been focused on the IoT. The research area poses challenges such as the advanced technologies for sensors and actuators, identifications with objects in IoT, interoperable serviceoriented technologies to share real-world data among heterogeneous devices, interoperable middleware, networking technologies for wired and wireless networking to interconnect things, application services that store, integrate, and process in real-time variable data streams from devices, infrastructure for storage and computing capabilities for IoT application services and for processing big data, quality of service assurance for efficient resource management to allocate, track, and resource utilization, scalable management of network, computing, and storage capacity across multiple objects, advanced security, privacy, authentication, trust and verification with the IoT applications, and numerical analysis and simulation technologies for IoT application with wireless sensor networks. The topics have been more aggressively covered by journals in the advanced technologies and application of the related wireless sensor networks and wireless telecommunications with the IoT [1 3]. This special issue discusses the following: advanced technologies for sensors and actuators; interoperable service-oriented technologies; interoperable middleware; networking technologies for wired and wireless networking to interconnect things; application services to store, integrate, and process real-time information; infrastructure for IoT application and services and big data processing; quality of service for efficient resource management; scalable management of network, computing, and storagecapacityacrossmultipleobjects;advancedsecurity, privacy, authentication, trust, and verification with the IoT applications; software defined networking and the opportunities in the development of IoT applications; numerical analysis and simulation technologies for IoT application with wireless sensor networks. 2. Related Works L.-C. Tseng et al. discussed the problem of distributed channel assignment in self-organized cognitive sensor networks

10 2 International Journal of Distributed Sensor Networks withunknownchannelandunknownnumberofclustersin the paper entitled Self-organized cognitive sensor networks: distributed channel assignment for pervasive sensing. The proposed method outperforms the random selection scheme in terms of average capacity, while the performance loss compared to the exhaustive search is limited. In addition, its fairness level is comparable to that of the random selection and surpasses the exhaustive search scheme. E. Troubleyn et al. proposed to use broadcast aggregation as a solution to overcome these drawbacks. Their paper has shown that broadcast aggregation reduces the average queue occupation with 2 (of the 15 available) places, which leads to fewer packet drops and it has been entitled Broadcast aggregation to improve quality of service in wireless sensor networks. Thisleadsonitsturntoathroughputandreliability increase up to 23% compared with no aggregation and up to 15% compared with unicast aggregation. Moreover, this paper has shown that packets become less dependent on the individual timeouts per destination, which reduces the drawbacks of partial aggregation. F. Buiati et al. presented a zone-based MIIS architecture, in which the access networks are grouped into mobility zones, managed by different MIIS servers in the paper entitled A zone-based media independent information service for IEEE networks. The decentralized MIIS deployment provides higher resilience and scalability with regard to the mobility information distribution. The results show that the proposed scheme outperforms the standard MIIS in terms of discovery delay and signaling overhead. Future work includes the study of security mechanisms and interoperator service agreement models. Y.-S. Hong et al. implemented intelligent electronic acupuncture system using sensor modules in the paper entitled Implementation of intelligent electronic acupuncture system using sensor module. This paper used the sensor modules to obtain a patient s diagnosis signals. These sensor modules consist of 5 parts. These sensor modules detect and analyze the abnormal signals from human body. The authors analyzed the signals to make instructions for the treatment. And then, the researchers designed the sensing pads for electronic acupuncture and also developed adaptive wireless acupuncture system to adjust strength and time of acupuncture and several acupuncture points of patients by using fuzzy technology. Y. Wan et al. proposed a local search algorithm, and a theoretical approximation ratio bound has been provided in the paper entitled Node placement analysis for overlay networks in IoT applications. The IoT-based overlay node placement problem is formulized and analyzed. The major contributions of the paper include providing the time complexity of multihop k-onpp (overlay node placement problem) and its theoretical limit boundary of approximation ratio and proposing a local search algorithm. Furthermore, thetimecomplexityandapproximationratioboundaryofthe local search algorithm are given. The proposed local search algorithm is evaluated by both time and efficiency where efficiency refers to the degree of approximation of algorithm results with optimal solutions. Another algorithm, TAG, is used for comparison. Finally, a simulation experiment based on network simulator EstiNet is provided. The experimental results show network delay benefits from the proposed method. H.-I.Wangproposedtheconceptofthe Internetof Things to construct a green campus environment which will realize the idea of energy saving in the paper entitled Constructing the green campus within the Internet of Things architecture. The architecture of the construction of green campus is established and three application systems have been developed as well. The efforts of this work allow the campus to manage the computer labs and the air conditioners more efficiently. The sensor network will save more energy since data are reported periodically and the analysis will be carriedoutintimetolocatetheproblems. A. L. Valdivieso discussed the advantages of the innovative concept of software defined networking (SDN) in the development of Internet of Things in the paper entitled SDN: evolution and opportunities in the development IoT applications. Software defined networking (SDN) appears as aviablealternativenetworkarchitecturethatallowsprogrammingthenetworkandopeningthepossibilityofcreating new services and more efficient applications to cover the actual requirements. SDN proposes the separation between data and control planes and a centralized control of the network. Moreover, SDN establishes open interfaces between the control and data plane. This paper describes this new technology and analyzes its opportunities in the development of IoT applications. It also presents the first applications and projectsbasedonthistechnology,suchashomenetworking, security, virtualization, multimedia, and mobile networks, among others. Finally, the issues and challenges around the topic are discussed. N.-C. Hsieh et al. implemented a system to deliver appropriate services according to individual needs based on its preprocessingofclassificationandtofurtherreducethecosts of manpower and loading of care staff through the analysis of assessment logs in the paper entitled Ensuring healthcare services provision: an integrated approach of resident contexts extraction and analysis via smart objects. With this systematic appraisal, this system can not only help care staff determine the needs of residents but also produce personalized health plans (i.e., weekly schedule towards comprehensive assessment and personalized care services). Results of implemented (and in use as well) system have demonstrated the feasibility that it can enhance the quality of care services to residents, working load of care staff, and efficiency of care-related information management for medical institution. Y.Cuietal.proposedasystemthatusesUPnPtocollect metadata from home appliances and cloud computing technology to store and process the metadata collected from ubiquitous sensor network environments in the paper entitled Home appliance management system for monitoring digitized devices using cloud computing technology in ubiquitous sensor network environment. This system utilizes a home gateway and is designed and implemented using UPnP technology to search for and collect device features and service information. It also provides a function for transmitting the metadata from thehomeappliancestoacloud-baseddataserverthatuses

11 International Journal of Distributed Sensor Networks 3 Hadoop-based technology to store and process the metadata collected by a home appliance monitoring service. R. de O. Albuquerque et al. proposed and described a trust model for distributed systems based on groups of peers in the paper entitled GTrust: group extension for trust models in distributed systems. A group is defined as a collection of entities with particular affinities and capabilities. All entities may have a trust and a reputation value of each other in the system. In many cases, it may be necessary to trust the whole system instead of one particular entity. In such cases, group trust represents the trust of their particular members. To achieve this, this paper presented a group trust calculation model. This paper implemented the proposed model in a P2P simulation tool and presented main results for group trust calculation. According to J. Wang et al., a new embedded device, Webit&NEU, and its reduced embedded real-time operating system used for IoT are implemented by their China Liaoning Province Embedded Technique Key Laboratory in the paper entitled Webit&NEU: an embedded device for the Internet of Things. Besides, related modules in terms of RFID technique, wireless communication, and network protocol are also provided in this paper. Compared with several current solutions of connecting devices and Internet, it has the advantages of good real-time performance, light weight, and low cost. E. Jung et al. suggested the agent service platform named iotsilo in which agents can communicate and cooperate on behalf of the heterogeneity devices in the paper entitled iot- Silo: the agent service platform supporting dynamic behavior assembly for resolving the heterogeneity of IoT. With this delegation approach, the iotsilo can support diverse devices without worrying about their differences. In designing an agent, several software design patterns are adopted to enable the agent to assemble behaviors for hiding the heterogeneity of devices. To investigate the effectiveness of the iotsilo, the authors developed eleven different types of the IoT devices to emulate real-world things with Arduino, deployed the devicesinbothkoreaandjapan,andthenconductedthree experiments. X. An et al. proposed a generalized nonparametric structural estimation procedure for the first-price auctions in the distributed sensor networks in the paper entitled Distributed risk aversion parameter estimation for first-price auction in sensor networks. To evaluate the performance of the aggregated parameter estimators, extensive Monte Carlo simulation experiments are conducted for ten different values of risk aversion parameters including the risk neutrality case in multiple classic scenes. According to W. Zhang et al., to reduce these excessive idle timeslots which the 4-ary tree anticollision algorithm brings, an anticollision algorithm based on adaptive 4-ary pruning query tree (A4PQT) is proposed in the paper entitled An efficient adaptive anticollision algorithm based on 4- ary pruning query tree. On the basis of the information of collision bits, some idle timeslots can be eliminated through pruning the 4-ary tree. Both theoretical analysis and simulation results support that A4PQT algorithm can significantly reduce recognition time and improve throughput of the RFID system. M. Choi et al. designed Internet of Things architecture, especially for wireless sensor networks in the paper entitled Improving performance through REST Open API Grouping for wireless sensor network. The architecture consists of wireless sensor networks with a microcontroller at the very bottom level. They are connected to smart devices at the next level. However, the computing capability of the smart devices is generally less powerful than that of the conventional devices. Thus, it is necessary to offload the computation-intensive part by careful partitioning of application functions. This research focused on designing the concept of MapReduce like approach through the web service grouping of several web services into one. Also, this paper proposed two methods: REST API grouping and REST API caching. First, the web service composition results in reducing energy consumption and communication latency by composing two or more REST web services into one. Second, the web service caching technique provides fast access that is recently accessed or frequently accessed. References Young-Sik Jeong Naveen Chilamkurti Luis Javier García Villalba [1] L. Atzori, A. Iera, and G. Morabito, The Internet of Things: a survey, Computer Networks,vol.54, no.15,pp ,2010. [2] Gartner Inc., Gartner s Hype Cycle Special Report for 2011, 2012, [3] H. S. Ning and Z. O. Wang, Future internet of things architecture: like mankind neural system or social organization framework? IEEE Communications Letters, vol.15,no.4,pp , 2011.

12 Hindawi Publishing Corporation International Journal of Distributed Sensor Networks Volume 2014, Article ID , 11 pages Research Article A Carrier Class IoT Service Architecture Integrating IMS with SWE Dongliang Xie, Chao Shang, Jinchao Chen, Yongfang Lai, and Chuanxiao Pang State Key Laboratory of Networking and Switching Technology, Beijing University of Posts and Telecommunications, Beijing , China Correspondence should be addressed to Dongliang Xie; xiedl@bupt.edu.cn Received 4 September 2013; Revised 22 January 2014; Accepted 22 January 2014; Published 22 May 2014 Academic Editor: Luis Javier Garcia Villalba Copyright 2014 Dongliang Xie et al. This is an open access article distributed under the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited. Integrating the sensing capabilities of wireless sensor network (WSN) into the traditional telecom network is an important stage to realize future ubiquitous intelligence in the Internet of Things. Driven by the vision of service oriented architecture (SOA), this paper proposed a carrier class Internet of Things (IoT) service architecture named as MUSE. MUSE integrates WSN with IMS OSE framework to enable the WSN services to be operable and manageable. Also sensor web enablement (SWE) framework is adopted to shield the heterogeneity of different WSNs. MUSE consists of two key entities MUSE Enabler and MUSE Gateway. On the one hand, the architecture promotes the node manageability and enriches the diversity of high level task planning flexibility. On the other hand, the architecture extends the telecom context-aware service and realizes service operability and network scalability. Moreover, the key components of the architecture and the detailed service procedure were introduced in the paper. Besides, an intelligent building prototype with 20 nodes was illustrated and the feasibility and performance of MUSE were verified at last. 1. Introduction With the ongoing development towards future internet of things (IoT), we are standing at the beginning of the age of 50 to 100 billion intelligent devices to be connected [1]. As the information acquisition engine and perception extension of IoT, the core value of wireless sensor network (WSN) is to collect massive information in a multiangle and multiparameter way, which has been applied in many fields, such as environment, transportation, industry, health care, and intelligent building [2]. With the huge number of things/objects and sensors/actuators connected to the Internet, how to access these heterogeneous and globally distributed sensor networks in a unified way and how to operate and manage these different kinds of sensors and actuatorsefficientlyareurgenttobesolved.inthispaper, from the view of telecom operators, we argue that integration of WSN and IP multimedia subsystem (IMS) is a feasible and cost-efficient way to address these challenges. Driven by the vision of service oriented architecture, information-oriented service, rather than connection-oriented service, is gradually becoming the intrinsic feature of ubiquitous information society. The carriers emphasize much on providing information service instead of simple network access. From the view of carriers, the operability of ubiquitous information-oriented service of WSN/IoT should be incarnated as follows. Service Operability. Current sensing capability is application specific, localized, and isolated rather than service-oriented. For carriers, they just treat sensing capability as a part of information acquisition approach but not as basic service ability, for example, SMS, MMS, or voice. In order to catalyze novel context-aware applications in the future telecom field, it is essential for carriers to bring the variety of sensing capabilities of WSN into the traditional service. Correspondingly, the carriers existing operation capabilities, such as authentication, authorization, accounting (AAA), quality of service (QoS), and service level agreement (SLA) mechanism, would enable ubiquitous perception services to be operable and manageable. However, in the current carrier operation platform, there is lack of unified information models and service procedures designed for WSN/IoT context-aware services.

13 2 International Journal of Distributed Sensor Networks Network Scalability. In the traditional vertical design model, WSN is highly customized and coupled with specific applications. Once the WSN has been deployed, it is hard to make flexible expansion and it may even need to be redeployed under some circumstances. Whereas future IoT needs to enable public and uncertain users participatory information acquisition, local knowledge share, and data mining, this will help to generate a consolidated view of the physical world. Connected by the carriers existing infrastructure, decentralized and disrupted sensor networks could cooperate via the heterogeneous access network. Decoupling the service and infrastructure is a feasible approach to break information silo and greatly improve the scalability of WSN. Node Manageability. The future IoT service should reflect the semantic service into sensors/actuators operations. The carriers need to provide the unified and centralized management for sensors/actuators registration, update, and cancellation. Thus a variety of sensors, from simple thermometer, humidity, noise, and complex camera to triaxial accelerometer, could be discovered, accessed, and utilized on a global level. Task Plan Flexibility. From the view of carriers service, WSN could be considered as a black box and the sensing process is executed by the task parameters transferred to it. To enable flexible IoT service, task planning and adjustment mechanism are needed. Ubiquitous perception service should support the parameterized and metadata-based WSN task assets.thefunctionofwsncanbesubjectedtoafeasibletask updating. The four features mentioned above comprise the prerequisites of the future carrier class IoT service. In this paper, we propose a service-oriented IoT architecture named as MUSE. Referring to unified nodes definition and data model of sensor web enablement (SWE) [3], MUSE promotes thenodemanageabilityandenrichesthediversityofhigh level task plan flexibility. Moreover, MUSE integrates WSN with IMS OSE framework, which extends the traditional telecom service and realizes service operability and network scalability with IMS. Besides, there are two key components that are illustrated. The first one is a newly introduced service enabler called MUSE Enabler, which decouples services and theinfrastructureinserviceabilitylayer.thesecondoneis MUSE Gateway, which serves as a medium between the IMS enabler and the WSN. The significance of MUSE is that, on the one hand, it achieves effective transmission of WSN information with the wide coverage of IMS and makes the service of WSN operational and scalable with the third-party management of IMS. On the other hand, it enables carriers to provide richer services for end users by utilizing the sensing capability and context-aware information. The rest of the paper is organized as follows: in Section 2, we introduce the related works of representative standardization organizations, research institutions, and academia. Section 3 proposes the MUSE architecture and analyses two key components and supporting technologies in detail. The four specific procedures of carrier s operation are discussed in Section 4. InSection 5, thefeasibilityofourframeworkis illustrated by a conceptual application. The prototype proves that the architecture is appropriate for the future carrier class IoT service provision and the network load could be reduced effectively. 2. Related Work To understand and deploy the ubiquitous service of IoT is a challenge and an unsolved problem. Many existing researches of standardization organizations, carriers, and academia are introduced as follows. SWE, proposed by Open Geospatial Consortium (OGC), is a relative comprehensive framework aiming at achieving a collaborative, coherent, consistent, and consolidated sensor data collection, fusion, and distribution system. SWE makes sensors discoverable and accessible and realizes the object of accessing heterogeneous WSN and sharing WSN resources. Moreover, it provides task planning for sensors to acquire observations of interest for flexible WSN services [4]. Yet for all, from carriers vision, SWE needs to be integrated into a telecom framework to operate the perception information service. IMS, proposed by 3rd Generation Partnership Project (3GPP), is the de facto standard of the 3G/4G core networks, offering the mechanism of multimedia service provision and the flexible session management [5]. IMS is intended to deliver next generation interactive and interoperable services, cost effectively, over an architecture providing the flexibility of the Internet. It is done by having a horizontal control layer that isolates the access network from the service layer. Except for the original services in the service layer, Open Mobile Alliance (OMA) defines the open service environment (OSE) and the standardized mobile service enabler specifications like Presence and XDMS, which support the creation of interoperable end-to-end mobile services. IMS functions well with the basic multimedia services based on streaming media, but for the fragmented and redundant sensing data services like WSN, related standards and mechanisms are being challenged. E-SENSE is an IoT research project proposed in European FP7 [6]. The goal of E-SENSE is to make ambient intelligence available in 3G/4G networks. To fulfill the requirements of different scenarios, this project proposes an architecture that integrates WSN into the IMS service platform and introduces the design of its gateway and network protocol stack. E- SENSE makes WSN manageable and controllable to some degree, whereas the WSN service is limited to providing the data observation service and without task planning. For the future ubiquitous perception service, we should introduce a more complete service procedure. European ISP Telefonica proposes an experimental platform named as ubiquitous sensor networks (USN) [7, 8]. The project composes and expands the present service enablers such as Presence, XDM, and Billing in OMA-defined OSE environment. It also decouples services from WSN by introducing the SWE information model. USN makes it possible for applications to be developed and deployed independently. Whereas this project concentrating on the USN Enabler, for

14 International Journal of Distributed Sensor Networks 3 Telecom operation framework AAA BSS OSS Application layer Service ability layer IMS/third-party enabler App Network access layer Security GW GW GW Perception extension layer App Maps IM PoC Signaling trigger control layer IMS (HSS, CSCFs) WSN enabler Service trigger GW WSN enabler SWE services Storage Catalogue Service logic Gateway SIP interface SIP extension IMS interface SensorML O&M Service logic Storage SWE client WSN interface Figure 1: MUSE Architecture. future IoT, description of service procedures, for example, gateway registration, services capabilities publishing, and network remote management, needs to be emphasized further. Researchers from Concordia University propose an architecture for the integration of the sensing capabilities of WSN in IMS to provide perception information services to end users [9, 10].BasedontheIMSPresenceservice,the system adopts the publish/subscribe mode to realize WSN data access. In the architecture, sensor data are stored in the message body of SIP message in the form of extended PIDF [11]. It contributes to realizing the publish/subscribe mechanism by presence of service enabler which has little influence on current architecture. We prefer to design a newly enabler instead of modifying Presence enabler in that the future IoT service needs a more flexible service provision mechanism to transmit massive and redundant data of WSN. The above studies promote the current sensing capability to service-oriented pattern to some extent, respectively. However, these studies have not yet proposed a systematic solution that satisfied the four carriers features and requirements mentioned in Section 1. This paper proposes a standardized ubiquitous IoT service architecture in IMS-switched network which accords with the standard service specifications of carriers. 3. Architecture Description MUSE satisfies the intrinsic requirements of the ubiquitous IoT service, and the key feature lies in supporting the decentralized and heterogeneous WSN accessing telecom network in unified and flexible approaches. In this section, we first propose the architecture of the IMS/WSN convergence system which follows the technical deployment specification in IMS. Further, we design two core network components: MUSE Enabler and MUSE Gateway. The MUSE Enabler realizes the functions of sensor networks tracking, data acquisition, and task planning by referring to services in the SWE framework and the catalogue service in OGC. The MUSE Gateway manipulates adaptation of protocols between WSN and telecom networks and format conversion of data including metadata and observation data. Finally, we describe several core technologies in this architecture Overall Structure. Based on the typical layered IoT framework, Mari Carmen Domingo tried to introduce IMS into the framework, which makes a beneficial tentative for the future IoT ubiquitous services [12]. In this paper, MUSE is also based on the typical layered framework shown in Figure 1, from bottom to top as follows: perception extension layer, network access layer, signaling control layer, service ability layer, and application layer. The functions of the five layers are defined as follows. (1) Perception extension layer. It realizes the effective information acquisition and transmission of the physical world. (2) Network access layer. It is responsible for interacting with the IMS network, as well as both data collection and network management of WSN. (3) Signaling control layer. It consists of IMS entities, for example, HSS and CSCF. And it realizes the functions of signaling route, access authentication, session control, service trigger, and so forth. (4) Service ability layer. It is responsible for integrating and managing the WSN resources functions such as service discovery and data acquisition. (5) Application layer. It supplies perceptive applications to the end users by utilizing the I/O interface provided by the service ability layer. For the mobile communication network operators, this layered architecture makes the WSN seamlessly embedded into the telecommunication networks operation and

15 4 International Journal of Distributed Sensor Networks management frameworks, and the value is added to the information by using context awareness which comes from detecting, storing, processing, and integrating situational and environmental information gathered by sensor nodes. In the MUSE architecture, we also introduce the information model and the service procedure borrowing from SWE framework and make modifications to accord with the service specifications of carriers. The main technological improvements of this architecture are as follows. Information repository Observation data User profile SWE services SOS SAS SPS WNS (1) Data format standardization. We utilize the SensorML as the standard model and use XML Schema to describe sensors systems and processes. What is more, O&M is used for encoding observations and measurements from a sensor [13, 14]. (2) Service trigger. In order to realize the service trigger in IMS, we propose the rule of service point trigger (SPT) for WSN services. (3) Protocol extension. SIP messages are extended by modifying the header and the message body to accommodate the control signaling. (4) Catalogue service. Catalogue service from OGC is introduced to discover the WSN and services. Specifically, the two key components designed in the architecture are as follows. (1) The MUSE Enabler is a centralized service engine. All requests from the upper applications are handled by it and all the metadata and observation data published by the WSN gateway are managed in it. (2) The MUSE Gateway is proposed in the network access layer, which connects different types of WSN with the mobile communication networks. To introduce the two components into this architecture, several core technologies such as the service triggering rules, the extension of SIP messages, and the catalogue service are proposed. The advantage of this architecture is that it enables WSN as a unified, flexible, and manageable service in a standard way Network Entity MUSE Enabler. MUSE Enabler is the most crucial entity in the IMS-switched architecture we proposed. MUSE Enabler in the service ability layer is designed according to the OMA OSE standard. It interacts with entities such as Presence, PoC, and XDM through a standardized way. XDM enabler specifies user-specific service-related informationdefined in well-structured XML documents. In this paper, we adopt the XCAP protocol as the communication protocol betweenthewsnenablerandthexdmserver.thegoalof MUSEEnableristointroducetheIoTserviceintothetelecom operation platform, integrate WSN resources, and provide ubiquitous perception services to end users. MUSE Enabler involves three functions that are (1) registration and tracking of WSN that realize the management of WSN resources, (2) decomposition of the access request, data acquirement from relatedwsn,andfusiondatapushing,and(3)wsntask planning that provides functions of WSN management and sensor observation according to the service request. WSN info Service logic function Catalogue service Publisher Register Notifier Subscriber SIP interface Figure 2: MUSE Enabler. Shown in Figure 2, the reference structure of MUSE Enabler consists of two layers. The lower layer is SIP interface and the upper layer includes three modules which are information repository, service logic, and SWE functions. (1) SIP interface module. SIP interface module provides standardized SIP interface through which other entities could interact with the enabler. It receives and resolves SIP messages from other network entities and forwards the message body according to the message type. (2) Service logic module. Service logic module takes charge of service logic such as the data query of upper layer s applications and the verification of sensor task feasibility. This module is also responsible for the service logic of MUSE Gateway registration and data publishing. The workflow of the module is as follows: resolve the SIP message of SIP interface module, generate a standardized description interfaced to SWE function module, and invoke the SWE function module to execute specific operations of the WSN. (3) SWE functions module. Referring to the standard SWE service model, SWE functions module realizes WSN services of data observation, task planning, alerting, and notification [15]. It executes specific sensor query or operating parameter configuration according to the parameters that transmitted from the service logic module. In this module, the catalogue function executes registration and service discovery of WSN according to the OGC catalogue service standard. (4) Information repository module. Information repository module takes charge of storing WSN metadata, observation data, user profile, and subscription rules. Through the data access interface, SWE function

16 International Journal of Distributed Sensor Networks 5 module is capable of retrieving and updating data in the module. MUSE Enabler is designed according to IMS specifications, which achieves the seamless migration among different carriers. MUSE Enabler integrates sensor networks that with distributed, heterogeneous characteristics into a unified enabler and all the manipulations of the upper layer are gathered and processed by MUSE Enabler. That shields the differences of the underlying networks. Meanwhile, the enabler supplies an encapsulation interface for the upper layer, which provides convenience for third-party application developers and hides the internal network of carriers. Moreover, it manages user profile and billing information within the hierarchical architecture in IMS, avoids the function duplication in the traditional vertical network structure, provides perfect QoS and security mechanism, and realizes the operability of ubiquitous perception services MUSE Gateway. As a bridge that associates WSN with IMS, MUSE Gateway plays an important role in the IoT service architecture [16]. The goal of Gateway is to enable WSN to access the telecom network and respond to the service request of MUSE Enabler. Gateway contains three functions: (1) communication protocol adaptation. As a multimode device, gateway could communicate with WSN that adopts short-range wireless communication protocols such as ZigBee and 6LoWPAN [17] and access the IP-based IMS system via xdsl, GPRS, and so forth. (2) Format conversion of metadata and observation data: gateway adopts SensorML as the unified description for sensors and O&M as the sensor data information model, which shields the heterogeneity of underlying networks. (3) Support of SWE: gateway initiates the progress of WSN registration and responds to the upper layer request, making the service progress accord with the standard service procedure that SWE defines. As shown in Figure 3, MUSE Gateway could be divided into three layers. As the interface to WSN, the lower layer takes charge of the WSN management and WSN data collection. The upper layer that interacts with IMS is responsible for receiving and sending SIP messages. The middle layer, including the information repository module, the service logic module, and the SWE client module, is responsible for storing and processing sensory data. (1) Information repository module. This module contains three repositories. The sensor data repository stores observation data that WSN collects. The information model repository contains the sensor description model named SensorML and the observation data model named O&M. The subscription rules repository contains the data publishing trigger that MUSE Enabler defines. (2)Servicelogicmodule.Thismodulecarriesoutthe service logic of WSN registration and observation datapublishing.italsorespondstotherequestofdata querying and sensor task planning that the MUSE Enabler sends. This module generating and resolving SIP interface Service logic function Register Publisher Subscribe Notifier Information repository Info. model Subscribe rules WSN interface Observation data SWE client SAS client SOS client Info. acquisition Figure 3: MUSE Gateway. WNS client SPS client WSN management thesipmessageintheserviceproceduremainly servesforthesweclientmodule. (3) SWE client module. This module realizes sensor observation and sensor task planning and notification services according to the standard SWE service model. It interacts with the information repository module and generates perceptual information to send to MUSE Enabler. It also calls the WSN interface and realizes the WSN task planning function of service platform. Modules cooperating with each other in Gateway standardize the metadata and observation data of WSN and send them to MUSE Enabler through the SIP message, which eliminates the heterogeneity of WSN. Moreover, it strengthens the WSN management of the service platform by introducing the WSN registration and task planning Supporting Method. Several key challenges exist in the MUSE architecture as follows: (1) how the IMS service is deployed, (2) how to use a standardized communication protocol and data description protocol to complete the service process,and(3)howtoimplementthewsnregistration anddiscovery.inthissection,wedesignasetofsupporting technologies to solve the challenges Service Trigger. Service trigger is the basis of deploying WSNservicesinIMS.TheIMSservicetriggercanbesummed up as service requests forwarding among the network element entities according to service rules [18]. In order that the WSN service process can be carried out smoothly, the initial filtering rule is defined to correctly route the SIP message in the service process of WSN network registration and sensing information publishing.

17 6 International Journal of Distributed Sensor Networks When a subscriber signs a service contract with the internet service provider (ISP), the ISP establishes the IMS user configuration information and stores it in HSS. Then S- CSCF will generate a third-party REGISTER request to the MUSE Enabler. Thus MUSE Enabler knows the existence of the gateway and the WSN attached to it. SPT for the WSN gateway registration in this architecture is shown in Algorithm 1. This SPT means that SIP message, whose method is REGISTER and header is WSN INFO, will be forwarded to wsn enabler@open-ims.com server. By utilizing the service trigger mechanism, SIP messages in the service process can be forwarded to the MUSE Enabler correctly SIP Extension. SIP message is the control protocol of the IMS network service, which is used to create, modify, and release sessions of one or more participants. In order to realize the basic function of network registration, data publishing, and data query, at the same time, is compatible with the existing IMS signaling system, the original SIP message is extended in this scheme. The REGISTER message is responsible for registration and status update of gateway in this scheme. The REGISTER message carries SensorML describing WSN network through extending its message body [13], so as to satisfy the requirements of WSN gateway registration. Specifically, SensorML is an Extensible Markup Language based on XML encoding in the SWE framework, which provides the standard sensor model and observation process. The PUBLISH message in the IMS domain is mainly responsible for event status update. Because the sensor message publishing in WSN is similar to the original event status update in IMS, the WSN gateway utilizes the PUBLISH messagetopublishdata.gatewayneedstoconstructthedocument of SensorML and O&M in the body of the PUBLISH message to complete the publishing of the WSN metadata and observation data. Specifically, O&M describes sensing observation data in a unified standard utilizing XML format in the SWE framework [14] and thus shields the difference of sensing observation data derived from the heterogeneous WSN. SIMPLE protocol in IMS is an event notification framework, which is based on the SIP message and is extended for IM and Presence services, and it mainly consists of the SUBSCRIBE message and the NOTIFY message [19, 20]. In this architecture, the upper layer application utilizes the SUBSCRIBE message in which filtering rules of the sensor network are carried out to complete the subscription and query of observation data. When sensing data meets the user s subscription conditions, MUSE Enabler uses the NOTIFY messages to notify the upper layer application that it carries the relevant data description. The above extension of SIP messages makes different network elements communicate with each other in a standardized message format, meets the functional requirements of MUSE Enabler, realizes information fusion among heterogeneous sensor networks, and provides a unified interface to upper layer applications. <InitialFilterCriteria> <TriggerPoint> <SPT> <Method>REGISTER</Method> <SIPHeader> <header>wsn INFO</header> <content> </content> </SIPHeader> </SPT> </TriggerPoint> <ApplicationServer> <ServerName>sip:wsn enabler@open-ims.com </ServerName> </ApplicationServer> </InitialFilterCriteria> Algorithm 1: SPT for gateway registration Catalogue Service. Catalogue service is a featured service prototype in MUSE. The integration of WSN and IMS system is optimized by the SWE standardized model. Furthermore,thescalablecatalogueserviceisdesignedto promote the interoperability of the sensing systems and make effective integration of spatial information resources. The integration of catalogue service and the basic SWE service canbeusedtostoreandmanageinformationsuchasservice metadata, sensor metadata, and sensing observation data. What is more, the catalogue service is used as the entrance service for external calls. The records of catalogue service are mainly composed of service capability document, SensorML describing perception description, task template, and a part of the O&M describing observation data. For the data request and response operations, the original perception information services interact directly with a related database. These operations procedures are slightly adjusted after the integration of catalogue service to ensure that the metadata database is consolidated and managed by the catalogue service and thus are transparent to the SWE services and users. Some adjusted operations are shown in Table 1. Catalogue service provides a uniform management for metadata and a more effective WSN resources integration mechanism, which is expected to be one of the vital factors in MUSE. 4. Service Procedure We define a set of service procedures following the corresponding SWE service model to standardize the services and support the operation of MUSE. The procedures provide functions of discovering, accessing, and utilizing WSN resources aiming at fulfilling the requirements of service operations. Specifically, the main interactive procedures are WSN gateway register, service discovery, observation data management, and sensor task planning.

18 International Journal of Distributed Sensor Networks 7 Table 1: Catalogue operations. Operation GetCapabilities DescribeSensor RegisterSensor Operation procedure Function: the operation provides service metadata to the user. Description: service metadata is stored in the metadata repository, which is managed by the catalogue service. Thus the user needs to send requests to the catalogue service rather than the SWE service instance to acquire service metadata. Function: users acquire sensor metadata through this operation. Description: the sensor metadata is also stored in the metadata repository and the user can only interact with the catalogue service rather than SOS itself. Function: the operation enables WSN to register itself into a SOS instance. Description: the sensor metadata should be forwarded to the catalogue service for further processing Gateway Register Procedure. The gateway register procedure allows MUSE Gateway to register in the corresponding serviceinstance.basedonthebasicimsthird-partyregister mechanism, the register message is transferred to the MUSE Enabler; then the WSN gateway can be registered to the corresponding SWE service instance and it will activate an update in the catalogue service. The register request carries basic identification information of gateway in its message head and a SensorML as its message body to describe the sensing capability of the WSN. When the S-CSCF receives the message, it generates a third-party REGISTER request to the MUSE Enabler due to the trigger point downloaded from the HSS. Then the MUSE Enabler triggers a SOS operation called RegisterSensor and meanwhile requests the catalogue service to update the service ability and other related information. So far MUSE Enabler discovers the existence of gateway and attaches it. When the status of the gatewayischanged,anupdateprocedureisperformedto refresh the state information in the metadata repository. By providing the gateway register procedure, MUSE then hastheabilitytomanageitswsnandnodeseffectively,which is the premise of WSN resource integration Service Discovery Procedure. In MUSE, the service discovery procedure provides standardized service capabilities query and acquisition mechanism. The user first sends the GetRecords request to the catalogue service in MUSE Enabler, and then a record list would be obtained as a feedback in case the request is valid. Different from the original work flow of SWE services, this procedure does not need to post requests to the specific service instance, in that the service metadata requested by users is managed through the catalogue service uniformly. The user s request is actually encapsulated within a SUBSCRIBE message in SIP; the SIP message will then be sent to the catalogue service. For the catalogue service, it queries the metadata repository using the given conditions and will return a service metadata document if the query is valid. By providing the service discovery procedure, MUSE can manage its services in a more standardized and effective way, which would support basic operations in the architecture and provide a unified service interface for third-party applications Sensor Observation Procedure. The sensor observation procedure in MUSE is designed to enable sensor data consumers, such as terminal users, applications, and other service instances, to acquire sensor observation data. The procedure is based on the SOS specification and redefined to adapt the uniform management through the catalogue service. There are two typical situations of the sensor observation procedure. The first one is the observation data acquirement of data consumers. A SUBSCRIBE message with the description of the interested perception information is sent to MUSE Enabler from the data consumer. This message carries semantic requests for perception information and the message body should fit the rules of GetObservation operation defined in SOS. The message is then forwarded to the corresponding service instance. The instance queries the data repository to get the requested observation data and returns an O&M document encapsulated in a NOTIFY message. For the second situation, MUSE Gateway, namely, data producer, publishes observation data actively. To make the procedure more flexible, two mechanisms are proposed as follows. (1) Regular uploading. The sampling parameters such as upload intervals and data types are predefined in thecapabilitydocumentofthecorrespondingsos instance. The only way to change these sampling parameters is to adopt sensor task planning operations. Due to its more efficient transmission, this mechanism is the main approach in MUSE. (2) Initiative uploading. Similar to the InsertObservation operation defined in SOS, the mechanism enables MUSE Gateway to publish observation data initiatively and thus makes the sensor data observation flexible. The sensor observation procedures standardize the process of requests and publishing, which supplies the data foundation to upper-layer perception information services Sensor Planning Procedure. The sensor planning procedure in MUSE is used for WSN task scheduling. We refer to the SPS standard task description encapsulated in SWE common data model, which enhances the task description through parameterization.

19 8 International Journal of Distributed Sensor Networks Operation Reserve Update Submit Table 2: Suboperations of sensor planning procedure. Description Freeze allocated task-related resources Update task parameters when in reserving or execution state Notify the service instance that the task is to be executed Gateway Gas monitor Sensor nodes Air conditioner Task scheduled Update Humidifier Light Submit Execution Complete Figure 5: Intelligent building scenario. Reserve Submit Reserved Reserve expired Figure 4: Task state transition diagram. Cancelled Failed Every sensor planning request must verify its feasibility by invoking the GetFeasibility operation [21]. The sensor planning procedure is a transaction operation. To ensure the WSN resources are managed effectively, we propose some suboperations which are shown in Table 2. Basedonourdefinition,astatetransitiondiagramis shown in Figure 4. In the diagram,a Reserved task shall not change to the state of Execution unless the client confirms it. If the client does not confirm a Reserved task in time, the task will expire. Meanwhile an Execution task can be updated before the task is executed successfully and transited into the final state, whereas, in some exceptional situations, for example, when the client cancels the scheduled task or the serverfailstocompletetheplannedtask,thetaskwillreach the final state. The transaction operation guarantees the instantaneity and the accuracy of sensor planning, which improves the utilizing efficiency of scheduled resources in certain degree. The sensor planning procedure enables the MUSE to support the integrated task scheduling service, which fulfills the requirements of the sensor observation planning and the WSN management planning for the service platform and thus brings benefits to the service diversity and service efficiency of MUSE. 5. Prototype Implementation and Proof of Concept Application In order to verify MUSE s feasibility and to evaluate its performance, we deployed an intelligent building prototype with 20 Micaz WSN nodes. Furthermore, a MUSE Gateway basedonmb510isdeployed.inthescenario,endusersobtain the perception information from WSN through the IMS. By the comparison of the sensor observation procedure between the traditional mode and the MUSE mode, the MUSE can reduce the network load effectively Conceptual Application Scenario. A typical indoor structure is shown in Figure 5. Many sensors are deployed in the house, which fulfills different functions including the house security monitoring, the family comfortable adjustment, and the smart home energy management. All types of interested observation data include the temperature, humidity, smoke, and light. In the security monitoring function, the temperatureandsmokeareusedtodetectthefireemergency;inthe family comfortable adjustment function, the temperature and humidity are used to control the conditioner and humidifier; and in the home energy management function, the temperature, humidity, and light are used to control the lights and conditioners in the room. In traditional WSN applications, these sensing data are independent, though the sensor observation data type is similar. In the MUSE mode, all observation data of WSN couldbegatheredbymusegatewayandthenconverged to the MUSE Enabler. Moreover, the task requests from upper layer applications could be adjusted flexibly through the MUSE Enabler Prototype Architecture and SIP Process. As shown in Figure 6, 20 Micaz sensor nodes are scattered in the building and a MB510 sink node is connected to a laptop running the UCT IMS Client. The sink node and the laptop perform MUSE Gateway s functions such as data collection and WSN registration altogether. The IMS core entity, including HSS andcscfs,isimplementedbasedontheopenimscore.and then a monitoring and surveillance client is implemented in an Android phone with an IMS soft terminal called IMS Droid. The gateway supports the multimode access such as Wi- Fi, IEEE , and Bluetooth. It also supports the plug and play feature and adopts basic SIP to adapt different carriers. Because of the lack of multimode devices, we adopt a tradeoff method that utilizes a MIB 510 sink node and a laptop running IMS client to carry out the function of MUSE

20 International Journal of Distributed Sensor Networks 9 WSN Open IMS core FHoSS CSCFs hss@open-ims.com cscf@open-ims.com (1) Register (3) Publish Sink Gateway gate@open-ims.com MUSE enabler wsn enabler@openims.com WLAN access point Figure 6: Prototype architecture. (2) Subscribe (4) Notify IMS client consumer@open-ims.com Gateway. The MB510 node is responsible for the sensing data collection, and the laptop realizes the data process function and the data publish function. The full signaling process of the service procedure is shown in Figure 7. To simplify the signaling process, we replace all CSCF entities in IMS with xcscf in the figure. Gateway register: a WSN is registered at first and the catalogue service updates the corresponding service information. InsertObservation: the registered WSN publishes its observation data. GetRecords and GetCapabilities: consumers query records and get interested service instance capabilities. DescribeSensor: the users choose to get sensor descriptions. GetObservation: consumers request for sensor observation data and expected to get a return encoded by O&M. InthesignalingprocessasshowninFigure 7, MUSE Enabler maintains the observation data which is published by MUSE Gateway periodically and end users get the data from MUSE Enabler when needed. This kind of proxy mode can reduce the number of SIP messages when end users request the data, which improves the efficiency of data access. On the other hand, as the WSN catalogue service is added to the scheme, end users have to access catalogue service to get the information of WSN and sensors before they request data. This may increase the network load for the WSN whose structureisalwayschanged Network Load Evaluation. In the prototype, we deploy five perception applications to analyze the performance of network load. The five applications gather temperature and humidity data inside the room. Undoubtedly each application has its own data sampling requirements like sampling time and data type for needs of each application. In traditional mode, every sensor network provides the temperature and humidity observation, and every node has a data package including sampling time, node identity, and sampling value, whose total size is about 40 bytes. As shown in Table 3, for the intelligent humidifier application, the sampling time is 2 minutes and there are 800-byte temperature and humidity data totally published at one time. SotheWSNpublishes24KBdataperhourintheintelligent humidifier application and 216 KB data per hour for all the five applications totally. In MUSE mode, we set the sampling time to the minimum value (30 seconds), and the original data size published once is 800 bytes. Since the data are encapsulated into O&M format, with about 400-byte description data added, the total packagesizeisexpectedtoincreaseto1.2kb.asthefive applications can share sensor data produced by the WSN in this area, the network load decreases to 144 KB per hour. The MUSE architecture successfully reduces the overall data size by reusing observation data, though single time data size is increased from 0.8 KB to 1.2 KB. As shown in Figure 8, in the traditional mode, the data size increases as the increasing of application number and the amplification is related to the requirements of the new added application. In the MUSE mode, the data size is related to the integral requirements, which has no significant change after it increases to a certain extent. When the number of applications is 1 or 2, MUSE mode shows no obvious advantage in network load, due to that the observation data of MUSE involves extra data descriptions, whereas, as the number of applications increases to some degree, such as 3, 4, or 5, MUSE mode reduces the data size obviously by data multiplexing. This mechanism is adaptive in comprehensive applications to reduce their data loads, especially for carriers to operate widely deployed application scenarios. 6. Conclusion In this paper, we proposed an ubiquitous IoT service architecture named as MUSE aiming to integrate WSN over IMS telecom framework in a flexible and unified way. MUSE takes advantage of the SWE framework to standardize the perception information model and the WSN service procedure, which shields the heterogeneity of different WSNs. By utilizing the OGC catalogue service, MUSE realizes the unified management of WSN and services. To realize the integration of WSN and IMS, we defined the service trigger rule of WSN services and extended the SIP protocol. And then the reference model of two main components in MUSE architecture is given. At last we illustrated four basic service procedures which provide the service reference mode for carriers. The advantages of integrating WSN over IMS telecom framework remain as follows. On the one hand, the carrier s strong operability enables ubiquitous perception services to be operable and manageable. On the other hand, the services which carriers provide can be flexibly expanded and deployed. The standardized integration of regional deployed WSN and wide area deployed mobile communication networks is an evolution stage to realize future ubiquitous intelligence intheinternetofthings.inthefuturewewillcontinuethe research and try to solve the key questions in the MUSE architecture, such as the more grained sensor data acquisition

21 10 International Journal of Distributed Sensor Networks MUSE enabler Gateway Consumer xcscf Catalogue service WSN services Gateway register Insert observation GetRecords GetCapabilities DescribeSensor (1) Register (4) OK (ID) (5) Publish (O&M) (8) OK (9) Subscribe (12) Notify (13) Subscribe (18) Notify (19) Subscribe (2) Register (3) OK (ID) (6) Publish (O&M) (7) OK (10) Subscribe (11) Notify (14) Subscribe (15) Query (16) Service capabilities (17) Notify (20) Subscribe (21) Notify GetObservation (22) Notify (23) Subscribe (26) Notify (24) Subscribe (25) Notify Figure 7: Signaling process of the service procedure. Table 3: Network load under different model. Mode Application Sampling data type Sampling period (s) Original mode MUSE mode Data size single time (KB) Data size per hour (KB/h) Self-adaptive humidifier T and H Patch board overtemperature alarm T Indoor temperature conditioner T and H Fire alarm T and H Plant nurseries T Total 216 Synthetic of the 5 applications TandH

22 International Journal of Distributed Sensor Networks 11 Network load (KB/h) Traditional mode MUSE mode Number of applications Figure 8: Network load changing curve in different mode. and task planning services, security, and privacy in catalogue service, mobility support for sensors and devices, and the highly effective stream data transmission. Moreover, we will attempt to carry the standardization works of future IoT service based on our work mentioned before. Conflict of Interests The authors declare that there is no conflict of interests regarding the publication of this paper. Acknowledgments This research is supported by National Natural Science Foundation of China (no and no ) and the Fundamental Research Funds for the Central Universities. References [1] M. Zorzi, A. Gluhak, S. Lange, and A. Bassi, From today s INTRAnet of things to a future INTERnet of things: a wirelessandmobility-relatedview, IEEE Wireless Communications,vol. 17, no. 6, pp , [2] I. F. Akyildiz, W. Su, Y. Sankarasubramaniam, and E. Cayirci, Wireless sensor networks: a survey, Computer Networks, vol. 38, no. 4, pp , [3] OGC sensor web enablement: overview and high level architecture, OGC , Open Geospatial Consortium, [4] Z.Chen,N.Chen,L.Di,andJ.Gong, Aflexibledataandsensor planning service for virtual sensors based on web service, IEEE Sensors Journal, vol. 11, no. 6, pp , [5] IP Multimedia Subsystem (IMS), Stage 2, 3GPP TS [6] A. Gluhak, M. Presser, D. Babb, L. Herault, and R. Tafazolli, e-sense reference model for sensor network in B3G mobile communications systems, in Proceedings of the 15th IST Mobile & Wireless Communications Summit, Myconos, Greece, June [7] M. Strohbach, J. Vercher, and M. Bauer, A case for IMS, IEEE Vehicular Technology Magazine,vol.4,no.1,pp.57 64,2009. [8] J. Bernat, Ubiquitous sensor networks in IMS: an ambient intelligence telco platformintelligence telco platform, in Proceedings of the ICT-MobileSummit 2008 Conference,Estocolmo, Sweden, [9] M. El Barachi, A. Kadiwal, R. Glitho, F. Khendek, and R. Dssouli, A presence-based architecture for the integration of the sensing capabilities of wireless sensor networks in the IP multimedia subsystem, in Proceedings of the IEEE Wireless Communications and Networking Conference (WCNC 08), pp ,LasVegas,Nev,USA,March2008. [10] H. Ru Cheng, F. Belqasmi, R. H. Glitho, and F. Khendek, The design and implementation of architectural components for the integration of the IP multimedia subsystem and wireless actuator networks, IEEE Communications Magazine, vol. 49, no.12,pp ,2011. [11] H. Sugano, S. Fujimoto, G. Klyne, A. Bateman, W. Carr, and J. Peterson, Presence information data format (PIDF), RFC 3863, [12] M. Domingo, A context-aware service architecture for the integration of body sensor networks and social networks through the IP multimedia subsystem, IEEE Communications Magazine,vol.49,no.1,pp ,2011. [13] M. Botts and A. Robin, OpenGIS sensor model language (sensorml) implementation specification, OGC Document OGC , Open Geospatial Consortium (OGC), Wayland, Mass, USA, [14] S. Cox, Observations and measurements part1 observation schema, OGC Document OGC r1, Open Geospatial Consortium (OGC), Wayland, Mass, USA, [15] X. Chu, T. Kobialka, B. Durnota, and R. Buyya, Open sensor web architecture: core services, in Proceedings of the 4th International Conference on Intelligent Sensing and Information Processing (ICISIP 06), pp , IEEE, Bangalore, India, December [16] M. El Barachi, A. Kadiwal, R. Glitho, F. Khendek, and R. Dssouli, The design and implementation of a gateway for IP multimedia subsystem/wireless sensor networks interworking, in Proceedings of the 69th IEEE Vehicular Technology Conference (VTC 09), pp. 1 5, Barcelona, Spain, April [17] IETF 6lowpan Working Group Homepage, [18] A. Gouya and N. Crespi, Service orchestration in IMS, in IMS Handbook, pp , CRC Press. [19] SIMPLE WG, SIP for instant messaging and presence leveraging extensions, 2006, simple-charter.html. [20] J. Rosenberg, H. Schulzrinne, B. Campbell, C. Huitema, and D. Gurle, Session initiation protocol (SIP) extension for instant messaging, RFC 3428, IETF, [21] I. Simonis and P. Dibner, OpenGIS sensor planning service implementation specification, OGC Document OGC r3, Open Geospatial Consortium (OGC), Wayland, Mass, USA, 2007.

23 Hindawi Publishing Corporation International Journal of Distributed Sensor Networks Volume 2014, Article ID , 10 pages Review Article SDN: Evolution and Opportunities in the Development IoT Applications Ángel Leonardo Valdivieso Caraguay, Alberto Benito Peral, Lorena Isabel Barona López, and Luis Javier García Villalba Group of Analysis, Security and Systems (GASS), Department of Software Engineering and Artificial Intelligence (DISIA), School of Computer Science, Office 431, Universidad Complutense de Madrid (UCM), Calle Profesor José García Santesmasess/n, Ciudad Universitaria, Madrid, Spain Correspondence should be addressed to Luis Javier García Villalba; Received 9 December 2013; Accepted 27 December 2013; Published 4 May 2014 Academic Editor: Young-Sik Jeong Copyright 2014 Ángel Leonardo Valdivieso Caraguay et al. This is an open access article distributed under the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited. The exponential growth of devices connected to the network has resulted in the development of new IoT applications and on-line services. However, these advances are limited by the rigidity of the current network infrastructure, in which the administrator has to implement high-level network policies adapting and configuring protocols manually and usually through a command line interface (CLI). At this point, Software-Defined Networking (SDN) appears as a viable alternative network architecture that allows for programming the network and opening the possibility of creating new services and more efficient applications to cover the actual requirements. In this paper, we describe this new technology and analyze its opportunities in the development of IoT applications. Similarly, we present the first applications and projects based on this technology. Finally, we discuss the issues and challenges in its implementation. 1. Introduction The emergence of new services and applications on-line, both in fixed terminals and mobile devices, has made the communication networks a strategic point in companies, institutions, and homes. The continued evolution of these services and the growth of the information circulating the Internet, bring unanticipated challenges to developers andcompanies.theadvancesinmicro-electro-mechanical systems (MEMS) increase the development of devices that automatically record, process and send information through the network. This kind of device, mainly consisting of sensors and actuators (RFID, Bluetooth Devices, Wireless Sensor Networks (WSN), Embedded Systems, and Near Field Communication (NFC)), has led to the origin of new ideas, concepts, and paradigms such as the Internet of Things (IoT). This device uses different ways to connect to the network including the traditional network infrastructure. At present, there are 9 billion connected devices and a number of 24 billon is expected for 2020 [1]. This device uses different ways to connect to the network including the traditional network infrastructure. However, the traditional equipment and network protocols are not designed to support the high level of scalability, high amount of traffic and mobility. The current architectures are inefficient and have significant limitations to satisfy these new requirements. The infrastructure responsible for transmitting the information coming of IoT devices (routers, switches, 3G and 4G networks, and access points) should be adapted to the new post-pc services (VoIP, sensor virtualization, QoS, cloud computing, and IoT applications) while providing security, stability, high rate, and availability amongst others. Some efforts such as the European Union Projects of SENSEI [2], Internet of Things-Architecture (IoT-A) [3], or Cognitive Management Framework for IoT [4] as well as the new protocols for Wireless Sensor Networks including the Data Driving Routing Protocol [5] have tried to get smarter connectivity between network elements (fully integrated Future Internet). However, these may not be the best options for a particular device or application domains (Smart Grid,

24 2 International Journal of Distributed Sensor Networks Intelligent Transportation, Smart Home, Health Care, and Environmental Monitoring and others). For this reason, in the last years, the idea of customizing the network behavior has emerged and gives users the flexibility to use the network resources according to their needs. Additionally, the development of new technologies to take decisions on IoT networks uses different calculation algorithms (genetic algorithms, neural networks, evolutionary algorithms, and other artificial intelligence techniques). It is desirable that these algorithms can be easily and dynamically implemented in the network equipment without waiting to be published in aprotocol. Software Defined Networking (SDN) is a network architecture that eliminates the rigidity present in traditional networks. Its structure allows the behavior of the network to be more flexible and adaptable to the needs of each organization, campus, or group of users. Besides, its centralized design allows important information to be collected from thenetworkandusedtoimproveandadapttheirpolicies dynamically. The development in recent years has impulsed new concepts, such as the network operating system (NOS). NOS tries to emulate the progress in computer systems. In this paper, the evolution of the main NOS is also analyzed. With this tool, it is possible to test the SDN concept in multiple projects (Home Networking, Data centers, Security, Virtualization, and Multimedia among others). Similarly, SDN has led to the design of models that integrate and finally achieve convergence of commonly separate architectures (Wifi-4G-LTE). However, these opportunities are still far from being implemented globally in production. Important issues such as convergence with existing networks, scalability, performance, and security are the challenges that should be overcome to be positioned in the market. Inthispieceofwork,wedescribeSDNanditsevolutionin the last years and analyze the opportunities and challenges in the future for this technology. Specially, the development of new generations IoT application (Smart Environment). The work is structured as follows: in Section 2 the limitations of traditional networks is presented; next, Section 3 defines the Software-Defined Networking concept; Section 4 presents the evolution of Network Operating Systems (NOS); then, in Section5, the first SDN applications are reviewed; in Section 6, the challenges of SDN technology are discussed; finally, Section7 presents the conclusions. 2. Limitations of Traditional Architectures The idea of transmitting information between two points through a network led to the design of communication protocols (TCP/IP, HTTPS, and DNS) and the creation of specialized devices in the transmission of information. These devices have evolved resulting in a variety of equipment (hub, switch, router, firewall, IDS, middlebox, and filters). This development has produced an exponential increase in the number of connected devices, in addition to the increase of the transmission rate and the emergence of online services (e-banking, e-commerce, , VoIP, etc.). All the devices responsible for transmitting information have similar features in their design and manufacture. First, there is a specialized hardware in the packet processing (data plane), and over the hardware works an operating system (usually Linux) that receives information from the hardware and runs a software application (control plane). The software contains thousands of lines of code for determining the next hop that a packet should be taken in order to reach its destination. The program follows the rules defined by a specific protocol (there are currently about 7000 RFCs) or some proprietary technology of the vendor. Modern equipment also analyzes information packets to search malicious information or intrusions (firewalls and IDS). However, all technology or software used in the manufacturing of these devices is rigid or closed to the network administrator. The administrator is limited only to configure some parameters, usually through low level commands using a command line interface (CLI). Moreover, each node is an autonomous system which finds the next hop to be taken by a packet to reach its destination. Some protocols (OSPF, BGP) allow the nodes to share control information between them, but only with its immediate neighbors and in a limited way in order to avoid tradditional load on network. This means that there is not a global view of the network as a whole. If the users need to control and modify a particular path, the administrator has to test with parameters, priorities, or uses gadgets to achieve the expected behavior in the network. Each change in the network policy requires individual configuration directly or remotely from each of the devices. This rigidity makes the implementation of high-level network policies difficult. Moreover, the policies require to be adaptive and dynamically react according to the network conditions. As Operating Systems (OS) evolve and adapt to the new needs and technological trends (support multi-cpu, multi-gpu, 3D, touch screen support, etc.), the network adaptability to new requirements (VLAN, IPv6, QoS, and VoIP) is implemented through protocols or RFCs. However, in the operating system the separation between hardware and software allows the continuous update of application, or even the reinstallation of a new version of an OS. In the area of networks, the design and implementation period of anewideacouldtakeseveralyearsuntilitispublishedin a protocol and incorporated in new devices. Some services are proprietary of the vendors and require that all network infrastructurebelongtothesamevendortoworkproperly. This limitation brings on the dependence on a specific technology or vendor. 3. Software-Defined Networking SDN The concept of Software-Defined Networking is not new and completely revolutionary; rather it arises as the result of contributions, ideas, and developments in research networking. In [6], three important states are determined in the evolution of SDN: Active Networks (mid-90s to early 2000), separation of data and control planes ( ), and the OpenFlow API and NOS ( ). All these aspects are discussed below. The difficulty for researchers to test new ideas in a real infrastructure and the time, effort and resources needed to standardize these ideas on the Internet Engineering

25 International Journal of Distributed Sensor Networks 3 Task Force (IETF) necessarily give some programmability to network devices. Active networks offer a programmable network interface or API that opens the individual resources of each node for the users, such as processing, memory resources, and packet processing and includes personalized features for the packets that circulate through the node. The need to use different programming models in the nodes was the first step for research in network virtualization, as well as the development of frameworks or platforms for thedevelopmentofapplicationonthenode.thearchitectural Framework for Active Networks v1.0 [6, 7] contains a shared Node Operating System (NodeOS), a set of execution environments (Execution Environments (EEs)), and active applications (Active Applications (AAs)). The NodeOS manages the shared resources, while the EE defines a virtual machine for the packet operations. The AA operates within an EE and provides the end-to-end service. The separation of packets to each EE depends on a pattern in the header of incoming packets to the node. This model was used in the PlanetLab [8] platform, where researchers conducted experiments in virtual execution environments and packets were demultiplexed to each virtual environment based on its header. These developments were important, especially in the investigation of architectures, platforms, and programming models in networks. However, their applicability in industry was limited and mainly criticized for its limitations in performance and safety. The work presented in [9]is an effort to provide the best performance to the active networks, and the Secure Active Network Environment Architecture [10] tried to improve their security. The exponential growth in the volume of traffic over the network produces the necessity to improve the supervision process and uses best management functions such as the management of paths or links circulating the network (traffic engineering), prediction traffic, reaction, and fast recovery if there are network problems, among others. However, the development of these technologies has been strongly limited by the close connection between the hardware and software of networking devices. Besides, the continuous increase in link rates (backbones) means that the whole transmission mechanism of packets (packet forwarding) is focused on the hardware, separating control, or network management to an application of software. These applications work best on a server, because it has higher processing and memory resources compared with a single network device. In this sense, the project ForCES (Forwarding and Control Element Separation) [11] standardized by the IETF (RFC 3746) established an interface between data and control plane in the network nodes. The SoftRouter [12] usedthis software interface to install forwarding tables in the data plane of routers. Additionally, the Routing Control Platform (RCP) [13] project proposed logical centralized control of the network, thus facilitating the management, the innovation capacity, and programming of network. RCP had an immediate applicability because it uses an existing control protocol BGP (Border Gateway Protocol) to install entries in the routing tables of the routers. The separation of data plane and the control plane allows the development of clean-slate architectures, such as the 4D project [14] and Ethanet [15]. 4D architecture proposes architectureoffourlayersbasedonfunctionality:dataplane, discovery plane, dissemination plane, and decision plane. Moreover, the Ethanet project [15] proposes a centralized control system of links to business networks. However, the need for custom switches based on Linux, OpenWrt, NetFPGA with support for Ethane protocol made it difficult the applicability of this project. At the present time, the Open- Flow protocol [16] is the most widely used in the research community and it has been the basis of different projects. Companies like Cisco have also submitted a proposal for a new architecture called Cisco Open Network Environment (Cisco ONE). Simplifying the previous analysis, the term Software- Defined Networking proposes some changes to the networks of today. First, the separation or decoupling of the data planeandcontrolplane,allowingevolutionanddevelopment independently. Secondly, it proposes a centralized control plane, thus having a global view of the network. Finally, SDN establishes open interfaces between the control plane and data plane. The differences between these architectures are shown in Figure 1. The programmability of the network provided by SDN can be compared with the mobile applications running on an Operating System (Android and Windows Mobile). These applications use the resources of the mobile (GPS, accelerometer, and memory) through the API provided by the OS. Likewise, the network administrator can manage and program resources in the network, according to user needs, through available APIs (proprietary or open) on the controller OpenFlow. OpenFlow [16] was originally proposed as an alternative for the development of experimental protocols on university campus, where it is possible to test new algorithms without disrupt or interfere with the normal operation of traffic of other users. Nowadays, the Open Networking Foundation (ONF) [6] is the organization responsible for the publication of the OpenFlow protocol and other protocols for SDN, such as OF-Config [17]. The advantage of OpenFlow, compared with previous SDN protocols, is the use of elements and features of hardware available in most network devices. These elements are the routing tables and the common functions are as follows: read the header, send the packet to a port, and drop a packet, among others. OpenFlow opens up these elements and functions; so these can be controlled externally. This implies that, with a firmware update, the actual hardware could potentially support OpenFlow. The companies do not need a complete change of their hardware to implement SDN in their products and services. The OpenFlow architecture proposes the existence of acontroller,aswitchopenflow,andasecureprotocolof communication. These elements are shown in Figure 2. Each OpenFlow switch consists of flow tables that are managed by the controller. Each flow table has three elements: packet header, actions, and statistics. The packet header is like a mask that select the packets which will be processed by theswitch.thefieldsusedforcomparisoncanbefrom

26 4 International Journal of Distributed Sensor Networks Applications Control plane Data plane OpenFlow protocol API API Control plane Data plane Control plane Control plane Secure channel Data plane Data plane Data plane Data plane Figure 1: Comparison between traditional and SDN architectures. Applications (QoS, firewall, routing...) OpenFlow protocol Secure channel API Controller API Packet in Flowtable Packet header Packet header Action Action Statistics Statistics Packet out Switch OpenFlow Figure 2: Elements of the OpenFlow architecture. layer2,3,or4ofthetcp/iparchitecture.thatmeans that there is not a separation between layers as in current architectures. All packets processed by the switch are filtered through this method. The number of fields that the switch can process depends on the version of the OpenFlow protocol. In OpenFlow v1.0 [18] (the most used version), there are 12 fields, while the latest version OpenFlow v1.3 defines the existence of 40 fields including support for IPv6. Once the header of an incoming packet matches the packet header of the flow table, the corresponding actions for that mask are performed by the switch. There are main and optional actions. The main actions are as follows: forward the packet to a particular port, encapsulate the packet and send it to the controller, and drop the packet. Some optional actions are as follows: forward a packet through a queue attached to a port (enqueue action) or 802.1D processing capabilities. If the header of an incoming packet does not match with the packet header of the flow table, the switch (according to its configuration) sends the packet to the controller for its analysis and treatment. Finally, the statistics field uses counters to collect statistic information for administration purposes. The OpenFlow protocol defines the following types of messages between the switch and the controller: controller to switch, symmetric, and asynchronous. The messages type controller to switch manage the state of the switch. Symmetric messages are sent by the controller or switch to initiate the connection or interchange of messages. The asynchronous messages update the control of the network events and the changes of state switch. Similarly, OpenFlow establishes two types of switches: OpenFlow-only and OpenFlow-enabled. OpenFlow-only switches use only OpenFlow protocol to process packets. On the other side, OpenFlow-enabled switches can additionally process the packet using traditional algorithms of switching or routing. The controller receives the information from the various switches and remotely configures the flow tables of the switch. Here, the user can literally program the behavior of

27 International Journal of Distributed Sensor Networks 5 Applications QoS Routing Firewall Controller Procera Northbound API Infrastructure OpenFlow protocol Switch, virtual switch, router Southbound API Figure 3: NOS, southbound, and northbound interfaces. the network. Unlike active networks, which proposed a Node Operating System. OpenFlow opens the notion of a Network Operating System (NOS). In this respect, in [19], the NOSisdefinedasthesoftwarethatabstractstheinstallation of the state in the switches of network of the logic and applications that control the behavior of the network. In recent years, the NOS has evolved according to the needs and applications for researchers and network administrators. 4. Network Operating Systems Evolution (NOS) The concept of Network Operating Systems (NOS) is based on the function of an operating system in computing. That is, the Operating System allows user to create applications using high-level abstraction of information, resources, and hardware. In SDN, some authors [20 22] have classified the abstractions of network resources as southbound and northbound interfaces (Figure3). The function of the southbound interfaces is to abstract the functionality of the programmable switch and connect it to the controller software. A clear example of southbound interface is OpenFlow. On the southbound interfaces, you run a Network Operating Systems. An example of NOS is NOX [23 25], among others. On the other hand, the northbound interfaces allow applications or highlevel network policies to be easily created and they transmit these tasks to Network Operating System (NOS). Examples of these interfaces are Frenetic [26, 27], Procera [21, 28], Netcore [29], and McNettle [30]. Then, they are analyzed in the main NOS and northbound interfaces. The NOX software [23]isthefirstNOSforOpenFlowand consists of 2 elements: processes of controller (controller) and a global view of the network (network view). Depending on the current state of the network, the user can make decisions and set the network behavior through these processes. In NOX, traffic is handled at the level of flows (flow-based granularity); that is, all packets with the same header are treatedsimilarly.thecontrollerinserts,deletesentries,and reads the counters found in the flow tables of the switches. Furthermore, due to the dynamic nature of traffic, NOX uses events (event handlers) that are registered with different priorities to be executed when a specific event occurs in thenetwork.themostusedeventsareswitchjoin,switch leave, packet received, and switch statistics received. Additionally, NOX includes system libraries implementations and common network services. Finally, NOX is implemented in C++ providing high performance. Moreover, there is an implementation entirely in Python denominated POX, which provides a more friendly developed language. Beacon [24] is a Java-based OpenFlow controller. Its interface is simple and unrestricted; that is, the user can freely use the constructors available in Java (threads, timers, sockets, etc.). Furthermore, Beacon is a NOS based on events; that is, the user sets the events that the controller listens to. The interaction with OpenFlow messages of the switch is done by the library OpenFlowJ, an implementation of the OpenFlow 1.0 [18] protocol and IBeaconProvider interface that contains the following listeners: IOFSwitchListener, IOFInitializerListener, and IOFMessageListener. Additionally, Beacon has multithreading support and provides important APIs implementations (Device Manager, Topology, Routing, and Web UI) as well as the ability to start, add, and complete applications without completely terminating a process in Beacon (runtime modularity). Although a NOS can handle the flow tables of the switches,therearesomeproblemsthatcancausemalfunction of the network [20, 22, 31]. For example, the controller receivesthefirstpacketthatarrivesattheswitchandhas not matched a header in the flow table. Then the controller analyzes it, assigns actions, and forwards these instructions to the switch so that the other similar packages follow the same route. However, during this time, the second, third, orfourthsimilarpacketscanbereceivedbythecontroller and cause an erratic operation. In other words, there are

28 6 International Journal of Distributed Sensor Networks virtuallytwoprocessesrunning,oneonthecontrollerand anotherontheswitch,andtheseprocessesarenotfully synchronized. Another limitation is the composition; that is, if the user wants to configure two different services on the same switch (e.g., routing and monitoring), it is necessary to manually combine the two actions on the switch, prioritize, and keep the semantics of each element of the network. This makes the design, coordination, and reuse of the libraries very difficult. Additionally, the switch has to handle two types of messages simultaneously: packets and control messages. Any mismatch cancauseapackettobeprocessedwithaninvalidpolicy andtherebycausingmajorsecurityproblemonthenetwork. For example, if there are two entries in a flow table with the same priority, the switch behavior might be nondeterministic, because the execution would depend on the design of the switch hardware. For this reason, the research communityhasworkedonsecureinterfacesthatautomatically interact and coordinate the correct behavior of the switch (northbound). Procera [21, 28] is a framework that allows politics or high-level network configurations to be expressed. This architecture provides different actions and control domains to program the behavior of the network. The main domains of control are as follows: time, data usage, flow, and status. With these domains, the user can determine a behavior depending, for example, on the time of day, amount of data transmitted, privileges or groups of users, type of transmitted traffic, and so forth. Actions can be temporal or reactive and are expressed on a high-level language based on Functional Reactive Programming (FRP) and Haskell. In [21] are the details of this language as well as examples of using Procera in monitoring applications and users control on a college campus. Frenetic [26, 27] is a high-level language dedicated to SDNnetworksdevelopedinPython.Itisstructuredby2 sublanguages: a Network Query Language and a Reactive Network Policy Management Library. The Network Query Language allows the user to read the status of the network. This task is performed by installing rules (low-levels rules) on the switch which does not affect the normal operation of the network. In addition, the Network Policy Management Library is designed based on a language for robots, Yampa, [32] and web programming libraries in Flapjax [33]. The actions use a constructor type rule containing a pattern or filters and action list as arguments. The main actions are as follows: sending to a particular port, sending packet to the controller, modification the packet header, and blank action that is interpreted as discard the packet. The installation of these policies is performed by generating policy events (queries), primitive events (Seconds, SwitchJoin SwitchExit, and PortChange), and listener (Print and Register). The results of experiments [26] show that Frenetic provides simplicity and a significant savings in code and lower consumption of network resources compared to NOX. One of the additional advantages of this language is the composition; that is, independent functional modules can be written and the runtime system coordinates its proper function in the controller and the switch. There are 2 types of composition: sequential and parallel. In sequential composition, the output of one module is the input of the next, for example, a load balancer that first modifies the IP destination of a packet and then searches the output port according to the new IP header. In parallel composition, both modules are executed virtually simultaneously in the controller; for example, if the balancer sends a packet with destination IP A to port 1, and packet B IP destined to port 2, this composition would result in a function that sends incoming packets for ports 1 and 2. McNettle [30] is a controller specially designed to offer high scalability at the SDN network. This is achieved using a set of message handlers (one for each switch) having a function that handles the switch-local and network-state variables and manages the supply actions from the network flows. The idea is that the messages from the same switch are handled sequentially, while messages from different switches are handled concurrently. Similarly, each message is processed in a single core CPU to minimize the number of connections and synchronizations inter-cores among other performance improvements. The tests performed in [30] show that McNettle have a higher multicore performance compared to NOX or Beacon. The controller proposed in [31] is based on the verification of the established politics, instead of searching bugs monitoring the controller operation. To perform the verification, the first step is to make use of the high-level language Netcore [29] to describe only the network behavior. Then, the Netcore Compiler translates the politics to network configurations as flow table entries. The flow tables information is analyzed by the Verifier Run-time System which transforms the network configuration into a lower abstraction level named Featherweght OpenFlow. Featherweght Openflow is a model that use synchronization primitives to guarantee the coherent behavior of the flowtables. Additionally, the Kinetic tool is described in [22]; this tool allows performing consistent updates in the network using two mechanisms: per-packet consistency and per-flow consistency. The per-packet consistency mechanism ensures that a packet that is transmitted across the network is processed with the same configuration when an update occurs. The per-flow consistency mechanism ensures that every packet that belongs to the same flow (e.g., a TCPconnection)willbeprocessedinthesamewaybyevery switch in the network. 5. First SDN Applications Software-Defined Networking provides the ability to modify the network behavior according to user needs. In other words, SDN itself doesnt solve any particular problem, but provides a more flexible tool to improve the network management. In order to test the advantages of this architecture, the research community has presented multiple projects of interest. Next, some of these applications are described Home Networking. In the emerging topic of Internet of Things (IoT), the management of devices and network

29 International Journal of Distributed Sensor Networks 7 resourcesinhomenetworksisabigchallengeduetothe number of users and devices connected to the same point (usually an access point). In [21, 34], the authors present an implementation of an OpenFlow-based system that allows the monitoring and management of user and control of theinternetaccessbasedon usagecaps oralimiteddata capacityforeachuserordevice.thesystemprovidesvisibility of the network resources and management of access based on user, group, device, application, or time of day and even enables the ability to exchange data capacity with another user. The system of control and network monitoring uses the friendly interface Kermit. The capacity management andnetworkpoliciesarebasedontheresonancelanguage [35] Security. The global vision of the network can improve the security of the systems. This security cannot be based only in the host-security, because such defenses are ineffective when the host is compromised. In [36],thePedigreesystem is presented as an alternative to provide security in the traffic moving in an enterprise network. This OpenFlow-based system allows to the controller the analysis and the approval of connections and traffic flows in the network. The host has a security module in the kernel (tagger) that is not under users control. This module labels the connections request to send information through the network (processes, files, etc.). This label is sent to the controller (arbiter) in the start of the communication. The controller analyzes the tagger and accepts or rejects the connection according to its policies. Once the connection is authorized, the corresponding flow tables are installed in the switch. Pedigree increases the tolerance to a variety of attacks, such as polymorphic worms. The systems increase the load in the network traffic and the host.however,thisloadisnothigherthancommonantivirus software Virtualization. The concept of virtualization in networks is similar to OS-virtualization, where different Operating Systems can share hardware resources. That is, in network virtualization, it is intended that multiple virtual networks can operate on the same infrastructure, each with its own topology and routing logic. Initially, VLAN technologies and private virtual networks allow the different users to share network resources. However, the separation is controlled only by the network administration and with limited parameters (port number) and just work with known network protocols. With the SDN data-control separation, the possibilities to create new advanced virtual networks are promising. For example, Flowvisor [37, 38] is an OpenFlow-based project that allows creating slices based on multiple parameters, such as bandwidth, flowspace (src/dst MAC, src/dst IP, and src/dst TCP ports), or CPU switch load. Each slice is independent; that means that it does not affect the traffic of the others slices. Additionally, it is possible to subdivide slices in order to create hierarchical models. A network service that takes advantage of virtualization of network resources is the migration of virtualnetworks.in [39], a system that enables the migration of the switch configuration to another device into the network is proposed, but without disrupting the active network traffic. The controller copies the flow tables configuration from the old to the new switch and modifies the paths automatically. This service enables the possibility to replace a network device avoiding the disruption or packet loss. This advantage can be usedtodynamicallymodifytheresourcesusedinthenetwork (green networks). In other words, the network can turn off or disable the unnecessary devices (nights or weekends) and automatically enable them in function of the traffic demand (peak hours) Mobile Networks. The devices in the infrastructure on mobile carrier networks share similar limitations as computer networks. Likewise, the carrier networks execute standards and protocols, for example, the Third Generation Partnership Project (3GPP) as well as the private vendors implementations. At this point, the SDN paradigm and its flow-based model can be applied on this kind of infrastructure offering better tools. Software- Defined Mobile Network (SDMN) [40] isanarchitecture that enables openness, innovation, and programmability to operators, without depending on exclusive vendors or over the top (OTT) service providers. This model consists of two elements: MobileFlow Forwarding Engine (MFFE) and the MobileFlow Controller (MFC). MFFE is a simple and stable data plane and with high performance. It has a more complex structure than an OpenFlow switch, because it must support additional carrier functions, such as layer 3 tunneling (i.e., GTP-U and GRE), access network nodes functions, and flexible charging. The MFC is the high performance control plane, where the mobile networks applications can be developed. Additionally, MFC has 3GPP interfaces to interconnect with different Mobile Management Entities (MMEs), Serving Gateways (SGWs), or Packet Data Network Gateways (PGWs) Multimedia. The multiple online multimedia services, for example, the real time transmissions, require high levels of efficiency and availability of the network infrastructure. According to studies presented by CISCO, the IP video traffic will grow from 60% in 2012 to 73% by 2017 [41]. Moreover, inthelastyears,theconceptofqualityofexperience(qoe) [42] gained particular strength, which attempts to redefine the Quality of Service (QoS) considering the level of user acceptance to a particular service or multimedia application. Therefore, SDN allows the optimization of the multimedia management tasks. For example, in [43]is improved theqoe experience through the path optimization. This architecture consists of two elements: the QoS Matching and Optimization Function (QMOF) that reads the different multimedia parameters and establishes the appropriate configuration for this path, and the Path Assignment Function (PAF) that regularly updates the network topology. In case of degradation of the quality on the links, the system automatically modifies the path parameters taking in count the priorities of the users. Similarly, the project OpenFlow-assisted QoE Fairness Framework (QFF) [44]analyzesthetrafficinthenetworkand identifies the multimedia transmissions in order to optimize

30 8 International Journal of Distributed Sensor Networks them in function of the terminal devices and the network requirements Reliability and Recovery. One of the most common problems in the traditional networks is the hardness to recover a link failure. The convergence time is affected by the limited information of the node to recalculate the route. In some cases, it is necessary the intervention of the network administrator to reestablish the network datapath. At this point, the global vision of SDN enables the customizing of recovery algorithms. [45] proposed an OpenFlow-based system that uses the mechanism of restoration and protection tocalculateanalternativepath.inrestorationmechanism,the controller looks for an alternative path when the fail signal is received. Meanwhile, in protection the system anticipates a failure and previously calculates an alternative path. Similar to a failure on switch or routers, the malfunction of the SDN controller (NOS failure, DDoS attack, and application error) can cause a collapse of the whole network. Therefore, the reliability of the network can be ensured through backup controllers. However, it is necessary to coordinate and update the information of control and configuration between principal and backup controllers. The CPRecovery [46] component is a primary backup mechanism that enables the replication of information between primary and backup controller. The system uses the replication phase to maintain the updated backup controller and the phase of recovery that starts the controller backup at the moment it detects a failure of the principal controller. 6. Challenges of SDN Technology The SDN advantages as applied technology in production networks are still close but not immediate. Furthermore, there are some challenges in terms of security, scalability, and reliability, among other aspects, which must be overcome in order to be considered acceptable for commercial users. Next, theseaspectsareanalyzed. As it was previously explained, the separation between data and control plane enables their independent development and evolution. In data plane, the rate of packet processing depends on the used hardware technology, such as Application-Specific Integrated Circuits (ASIC), Application- Specific Standard Products (ASSP), Field Programmable Gate Array (FPGA), or multicore CPU/GPP. Meanwhile, in control plane the performance also depends principally on the hardware and the NOS (Beacon, POX, Floodlight). However, a poor performance of one of the two levels can cause significant problems, such as packet loss or delay and incorrect behavior of the network of denial of service DDoS. For this reason, a balance in performance, cost, and facility of development for the hardware and software of SDN components is necessary. Moreover, OpenFlow uses the common hardware resources of actual networks, such as the flow tables. However, SDN can be extended beyond flow tables and use additional resources offered by actual hardware [17]. The integration and research of new features between control and data plane is arecentlyopentopic.applicationslikeencryption,analysis, and traffic classification and devices such as middleboxes and custom packet processors can be integrated and efficiently usedbythesdntechnology.ontheotherhand,thenumber and the position of the controllers in a network are open questions. The analysis presented in [47] exposes that the determining factors for the selection of the number and position are the topology and the expected performance of the network. The security is another fundamental aspect that must also be taken into account. For example, not all network applications should have the same access privileges [20]. The assignment of profiles, authentication, and authorization to access the network resources are necessary. In addition, OpenFlow establishes the optional use of TLS (Transport Layer Security) as authentication tool between switch and controller. However, there are not clear specifications that provide security for multiple controller systems that interchange information among them and with the switches. Additionally, Openflow establishes that an unknown packet could be send completely (or its packet header) to the controller, it can easily be affected by DDoS attacks by the sending of multiple unknown packets to the switch. The transition between actual network architectures to SDN-based architectures is also an open issue. Despite the emergence of network devices with OpenFlow support (IBM and NEC) in the market, it is impossible to replace the network infrastructure completely. The transition period requires mechanism, protocols, and interfaces allowing coexistence between both architectures. Currently, there are important efforts to achieve this objective; the Open Networking Foundation (ONF) published the IF- Config Protocol [48] as a first step to the configuration of OpenFlow devices. Similarly, the European Telecommunications Standards Institute (ETSI) as well as the el IETFs Forwarding and Control Element Separation Working Group (ForCES) works on the standardization of interfaces for the appropriated development of this technology. 7. Conclusion The exponential growth of devices and online services that exchange information over the network consolidated the concept of Internet of Things (IoT). In this new approach, the rigidity of traditional architectures is inefficient suggesting rethinking new ways to use the infrastructure and technology communications. Software-Defined Networking has emerged as an alternative to the current problems of traditional networks. It allows administrators to have a global view of the network, as well as the opportunity to control the network according to the needs of each organization. This work presents the basis of this technology and the development of Network Operating Systems NOS, as well as some interesting projects based on this paradigm. Additionally, the problems and challenges for the implementation of SDN in production networks are analyzed. It isnoteworthythatsdnprovidesthetoolstoimprovethe management of the network behavior. The use of this tool and the development of new SDN applications are new fields of study. In the future, this paradigm will bring new ways

31 International Journal of Distributed Sensor Networks 9 of viewing and using the communication networks as well as new business models focused on offering services and network applications. Conflict of Interests The authors declare that there is no conflict of interests regarding the publication of this paper. Acknowledgments Part of the computations of this work were performed in EOLO,theHPCofClimateChangeoftheInternational Campus of Excellence of Moncloa, funded by MECD and MICINN. This is a contribution to CEI Moncloa. Ángel Leonardo Valdivieso Caraguay and Lorena Isabel Barona López aresupportedbythesecretaría Nacional de Educación Superior,Ciencia,Tecnología e Innovación SENESCYT (Quito, Ecuador) under Convocatoria Abierta 2012 Scholarship Programno Theauthorswouldalsoliketothank AnaLucilaSandovalOrozcoforhervaluablecommentsand suggestions to improve the quality of the paper. References [1] J. Gubbi, R. Buyya, S. Marusic, and M. Palaniswami, Internet of Things (IoT): a vision, architectural elements, and future directions, Future Generation Computer Systems,vol.29,no.7, pp ,2013. [2] SENSEI, Integrated EU Project, [3] European Lighthouse Integrated Project, Internet of Things Architecture, [4] P. Vlacheas, R. Giaffreda, V. Stavroulaki et al., Enabling smart cities through a cognitive management framework for the internet of things, in IEEE Communications Magazine, vol.51, pp , [5] L.Shi,B.Zhang,H.T.Mouftah,andJ.Ma, DDRP:anefficient data-driven routing protocol for wireless sensor networks with mobile sinks, International Journal of Communication Systems, vol. 26, no. 10, pp , [6] Open Networking Foundation, [7] K. Calvert, Architectural Framework for Active Networks Version 1.0, [8] PlanetLab, [9] T. Wolf and J. S. Turner, Design issues for high-performance active routers, IEEE Journal on Selected Areas in Communications,vol.19,no.3,pp ,2001. [10] D. Alexander, W. Arbaugh, A. Keromytis, and J. Smith, A secure active network environment architecture: realization in SwitchWare, IEEE Network,vol.12,no.3,pp.37 45,1998. [11] L. Yang, R. Dantu, T. Anderson, and R. Gopal, Forwarding and Control Element Separation (ForCES) framework, RFC 3746, [12]T.Lakshman,T.Nandagopal,R.Ramjee,K.Sabnani,andT. Woo, The soft- router architecture, in Proceedings of the ACM SIGCOMM Workshop on Hot Topics in Networking, November [13] M.Caesar,D.Caldwell,N.Feamster,J.Rexford,A.Shaikh,and J. Van der Merwe, Design and implementation of a routing control platform, in Proceedings of the 2nd Conference on Symposium on Networked Systems Design and Implementation (NSDI 05),vol.2,pp.15 28,USENIXAssociation,May2005. [14] A. Greenberg, G. Hjaimtysson, D. Maltz et al., A clean slate 4D approach to network control and management, ACM SIGCOMM Computer Communication Review,vol.35,no.5,pp , [15]M.Casado,M.J.Freedman,J.Pettit,J.Luo,N.Mckeown, and S. Shenker, Ethane: taking control of the enterprise, in Proceedings of the ACM SIGCOMM 2007: Conference on Computer Communications,vol.37,pp.1 12,August2007. [16] N. McKeown, T. Anderson, H. Balakrishnan et al., OpenFlow: enabling innovation in campus networks, ACM SIGCOMM Computer Communication Review, vol. 38, pp , [17] A. Valdivieso, L. Barona, and L. Villalba, Evolution and challenges of software defined networking, in Proceedings of the 2013 Workshop on Software Defined Networks for Future Networks and Services, pp , IEEE, November [18] OpenFlow Switch Specification v.1.0.0, December [19] N. Feamster, J. Rexford, and E. Zegura, The road to SDN: an intellectual history of programmable networks, Technical Report, Princeton University, Princeton, NJ, USA, [20] S. Sezer, S. Scott-Hayward, P. K. Chouhan et al., Are we ready for SDN? Implementation challenges for software-defined networks, in IEEE Communications Magazine, vol.51,pp.36 43, [21] H. Kim and N. Feamster, Improving network management with software defined networking, in IEEE Communications Magazine,vol.51,pp ,2013. [22] M. Reitblatt, N. Foster, J. Rexford, C. Schlesinger, and D. Walker, Abstractions for network update, in ACM SIGCOMM Computer Communication Review,vol.42,pp ,2012. [23] N. Gude, T. Koponen, J. Pettit et al., NOX: towards an operating system for networks, ACM SIGCOMM Computer Communication Review,vol.38,no.3,pp ,2008. [24] D. Erickson, The beacon OpenFlow controller, in Proceedings of the 2nd ACM SIGCOMM Workshop on Hot topics in Software Defined Networking, pp , ACM, August [25] Floodlight, [26]N.Foster,R.Harrison,M.J.Freedmanetal., Frenetic:a network programming language, ACM SIGPLAN Notices, vol. 46,no.9,pp ,2011. [27] N.Foster,A.Guha,M.Reitblattetal., Languagesforsoftware defined networks, in IEEE Communications Magazine, vol.51, pp ,2013. [28] A. Voellmy, H. Kim, and N. Feamster, Procera: a language for high-level reactive network control, in Proceedings of the 1st Workshop on Hot topics in Software Defined Networks, pp.43 48, ACM, August [29] C.Monsanto,N.Foster,R.Harrison,andD.Walker, Acompiler and run-time system for network programming languages, ACM SIGPLAN Notices,vol.47,no.1,pp ,2012. [30] A. Voellmy and J. Wang, Scalable software defined network controllers, in ACM SIGCOMM Computer Communication Review,vol.42,pp ,ACM,2012. [31] A. Guha, M. Reitblatt, and N. Foster, Machine-verified network controllers, in ACM SIGPLAN NOTICES, vol. 48, pp , ACM, 2013.

32 10 International Journal of Distributed Sensor Networks [32] A.Courtney,H.Nilsson,andJ.Peterson, TheYampaarcade, in Proceedings of the ACM SIGPLAN 2003 Haskell Workshop,pp. 7 18, ACM, August [33] L. A. Meyerovich, A. Guha, J. Baskin et al., Flapjax: a programming language for ajax applications, ACM SIGPLAN Notices, vol.44,no.10,pp.1 20,2009. [34] H. Kim, S. Sundaresan, M. Chetty, N. Feamster, and W. K. Edwards, Communicating with caps: managing usage caps in home networks, in Proceedings of the ACM SIGCOMM 2011 Conference (SIGCOMM 11), vol. 41, pp , August [35] A. Nayak, A. Reimers, N. Feamster, and R. Clark, Resonance: dynamic access control for enterprise networks, in Proceedings ofthe1stacmworkshoponresearchonenterprisenetworking (WREN 09), pp , ACM, August [36] A. Ramachandran, Y. Mundada, M. B. Tariq, and N. Feamster, Securing Enterprise Networks Using Traffic Tainting,2009. [37]R.Sherwood,G.Gibb,andM.Kobayashi, Carvingresearch slices out of your production networks with OpenFlow, in ACM SIGCOMM Computer Communication Review,vol.40,pp , ACM, [38] R.Sherwood,G.Gibb,K.K.Yapetal., Flowvisor:anetwork virtualization layer, in OpenFlow Switch Consortium,2009. [39] P.S.Pisa,N.C.Fernandes,H.E.Carvalhoetal., OpenFlowand Xen-based virtual network migration, in Wireless in Developing Countries and Networks of the Future, vol.327,pp , Springer, Berlin, Germany, [40]K.Pentikousis,Y.Wang,andW.Hu, MobileFlow:toward software-defined mobile networks, in IEEE Communications Magazine,vol.51,pp.44 53,2013. [41] The Zettabyte EraTrends and Analysis, May [42] S. M. Patrick Le Callet and A. Perkis, Qualinet white paper on definitions of quality of experience, in Proceedings of the European Network on Quality of Experience in Multimedia SystemsandServices,COSTActionIC,1003,March2012. [43] A. Kassler, L. Skorin-Kapov, O. Dobrijevic, M. Matijasevic, and P. Dely, Towards QoE-driven multimedia service negotiation and path optimization with software defined networking, in Proceedings of the 20th International Conference on Software, Telecommunications and Computer Networks, vol.1,pp.1 5, IEEE, September [44] P. Georgopoulos, Y. Elkhatib, M. Broadbent, M. Mu, and N. Race, Towards network-wide QoE fairness using OpenFlowassisted adaptive video streaming, in Proceedings of the 2013 ACM SIGCOMM Workshop on Future Humancentric Multimedia Networking (FhMN 13), pp , ACM, August [45] S.Sharma,D.Staessens,D.Colle,M.Pickavet,andP.Demeester, A demonstration of fast failure recovery in software defined networking, in Testbeds and Research Infrastructure. Development of Networks and Communities, vol.44,pp , Springer, Berlin, Germany, [46] P. Fonseca, R. Bennesby, E. Mota, and E. Passito, A replication component for resilient OpenFlow-based networking, in Proceedings of the 2012 IEEE Network Operations and Management Symposium, pp , IEEE, April [47] B. Heller, R. Sherwood, and N. McKeown, The controller placement problem, in ACM SIGCOMM Computer Communication Review,vol.42,pp ,ACM,2012. [48] OpenFlow Management and Configuration Protocol (OF- Config) v.1.1.1, March 2013.

33 Hindawi Publishing Corporation International Journal of Distributed Sensor Networks Volume 2014, Article ID , 17 pages Research Article An Upper-Ontology-Based Approach for Automatic Construction of IOT Ontology Yuan Xu, Chunhong Zhang, and Yang Ji Mobile Life and New Media Laboratory, Beijing University of Posts and Telecommunications, Beijing , China Correspondence should be addressed to Yuan Xu; Received 7 November 2013; Accepted 4 February 2014; Published 3 April 2014 Academic Editor: Luis Javier Garcia Villalba Copyright 2014 Yuan Xu et al. This is an open access article distributed under the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited. Ontology gives us a reliable group of concepts and the relations between concepts in an IOT system. It does not only save words of format but also accurately transfers semantic data between human users and the computers. Hence, the usefulness of resources in IOT system depends on whether the domain ontology can be constructed effectively and correctly. In this paper we propose an automated method to construct the IOT ontology. First, we explain the necessity of introducing ontology automatic construction in IOT system and summarize the major challenges in existing approaches. Secondly, we introduced the existing ontology construction methods and summarize their issues. Thirdly, we give a framework of our ontology construction and research the key algorithms in detail: (1) knowledge-tuple extraction algorithm which contains contextual information; (2) concept semantic similarity algorithm which is based on the structure of tuple; (3) knowledge-tuple extraction model which is based on the structured information. Then we build a prototype and evaluate the ontology. Finally, we make conclusions and suggest directions for future research. 1. Introduction Ontology is a normalized knowledge base as a set of concepts within a domain and the relationships between pairs of concepts [1]. Ontologies provide the structural frameworks for organizing information and are used in artificial intelligence, semantic web, systems engineering, software engineering, biomedical informatics, library science, and enterprise bookmarking. In Internet Of Thing (IOT) domain, experts extracted concepts from system architecture to construct the machine-understandable ontology and now there are more than ten IOT domain ontologies. Through the normalized representation for domain concepts and relationships, ontology provides a support service for semantic-based heterogeneous resource search and service development in IOT. The current IOT domain ontologies are mostly based on artificial construct [2] by domain experts; the authority one is SSN ontology which has been built by W3C Semantic Sensor Network Incubator; in addition there are more than ten IOT domain ontologies include CSIRO, MMI, CESN, A3ME, and OntoSensor [3, 4]. These ontologies concepts are derived from the general knowledge of IOT domain such as platform, gateway, device, and sensor; they do not combine any actual IOT system and only contain a small set of concepts (tens to hundreds of concepts). At the same time, these ontologies have a slow improvement. But IOT system needs an ontology which combined system s features andprovidedarichsetofconcepts(tensofthousandsof concepts); in addition the ontology needs to improve itself with the development of IOT system constantly. Therefore, the existing IOT domain ontologies are not suitable for using in IOT systems directly. There are many researches on ontology automatic construction in recent years, include TextOntoEx [5], TF-IDF- ART [6], and GRAONTO [7]. But these researches are still in primary stage. The automatic construction models typically include the following sections. In order to obtain the domain knowledge, constructers need to collect concerned texts and web pages as information source first. Then they use the word segmentation and remove stop words to wipe off the irrelevant vocabularies from texts and web pages. The knowledge tuple (which consists of concepts and relationships between concepts, such as platform-hasgateway and gateway-has-device ) is the basic unit of ontology and an ontology is combined by a lot of knowledge

34 2 International Journal of Distributed Sensor Networks tuples; therefore, constructers need to extract knowledge tuples (mostly they extract the subject-predicate-object triple tuples from sentences as the knowledge tuples, subject and object are concepts, predicate is relationship). Taking into account the knowledge tuples are independent in existence, the constructers need to calculate the similarity between concepts in knowledge tuples and merge the similar concepts and constitute a connected ontology. Existing tuple extraction mainly use unstructured documents [5 7] (refer to information that either does not have a predefined data model or is not organized in a predefined manner, such as text and web page) as information source, and commonly uses the natural language processing algorithms to extract knowledge tuples from each sentence of the information s source; this approach makes each tuple lose context information in paragraph. In addition, IOT environment includes various resource description files (a type of structured information source, refers to information with embedded coding, such as markup, is used to give the whole and parts of the document with various structural meanings according to a schema) and this helps obtain a more comprehensive knowledge, but the existing models lack research on resource-description-based automatic construction. In the research on concepts similarity, existing approaches use linguistic similarity of vocabularies to merge synonyms; this neglects the scene that the same word in different contexts has different meanings and represents different concepts. Moreover, [8] put forward that selecting an existing domain ontology as the upper ontology is the first step in classic five steps for ontology construction, but existingapproachesmostlyskipthisstep.welistthefour major challenges and problems in this paper as follows. (1) Upper ontology (also known as a top-level ontology or foundation ontology) is an ontology which describes very general concepts that are the same across one domain; it mostly is used to support a broad semantic interoperability between a large numberofontologiesinadomain[9]. Selecting an existing domain ontology as the upper ontology can make full use of the most general concepts in this domain and support the interoperability with other ontologies, but existing construction models mostly neglect the interoperability between ontologies and skip this step. Therefore we will research the IOT upper-ontology evaluationand selection in Section 3.4. (2) Knowledge tuples use concepts and relationships to describe the domain knowledge (e.g., platformhas-gateway, platform and gateway are concepts; has is relationship); they are the basic unit in ontology construction. Existing construction models commonly use the natural language processing algorithms to extract knowledge tuples from each sentence of information source; this approach neglects the context information in paragraph when extracting thetuples(e.g.,wehaveasentence Commondata formats include xml and json. in an IOT platform description text, we can extract tuples formatinclude-xml and format-include-json. But without the context information, we will never know whether thesedataformatsareusediniotplatform).based on the discussion above, existing tuple extraction approaches lost the context information in paragraph. Therefore, we propose a context-informationbased (CIB) knowledge-tuple extraction algorithm in Section 3.2. (3) Existing knowledge-tuple extraction algorithms mostly rely on unstructured documents and neglect the structured documents. But IOT environment includes various resource descriptions, which are a type of structured documents. Resource description has been used to describe various IOT resources such as device and data. We use the Xively (previously Pachube, one of the most famous IOT platform) as an example; it has more than twenty thousand resource descriptions and exceeds the number of characters in other pages in this website. This helps to obtain a more comprehensive knowledge. Therefore we propose a resource-description-based (RDB) knowledge-tuple extraction model in Section 3.2. (4) Existing concept similarity algorithms mostly use the linguistic similarity of vocabularies to merge synonyms, this neglects the scene that the same word in different contexts has different meanings and represents different concepts. This issue will lead to the concepts in ontology only that reflect the linguistic characteristics of words without any context information in specific scene (we use a couple of tuples as example, device-has-name-is-sensor and location-has-name-is-china; they have the common concept name, but they represent different context information and we cannot merge two name in ontology. If we merge them, we can get that china is also a name of device and it is obviously wrong). In linguistics, polyseme is a word or phrase with different senses. Majority of polyseme is verb and adjective; they express one meaning in a certain context. According to statistics, each word has more than three meanings on average, so polysemy is a common phenomenon in human language. In IOT environment, there are inheritance relationships between concepts. Because of inheriting different characters from parent concepts, the vocabularies with same meaning in linguistics should reflect different concepts in ontology. Taking into account that the context information of a word in a tuple comes from the tuple structure, we propose a tuple-structure-based (TSB) similarityalgorithmin Section 3.5. The remainder of this paper is organized as follows. Section2 introduces the related work on ontology automatic construction. The details of the proposed ontology construction method are explained in Section 3. Theprototypeand evaluation are shown in Section 4. Finally, conclusions and future work are shown in Section 5.

35 International Journal of Distributed Sensor Networks 3 2. Related Work Due to the critical importance of ontology in the various fields and applications, constructing domain ontology efficiently and effectively has been an important research area recently.thus,differentmechanismsandmethodologiesfor designing and building ontology have been proposed. At present, most of the researches on ontology automatic construction are in accordance with the four steps in Section 1: (1) preprocessing the information sources; (2) extracting knowledge tuples from information sources; (3) calculating the semantic similarity between concepts and relationships in knowledge tuples, merging the similar concepts and relationships, constituting an initialization ontology; (4) calculating the importance and Boolean relations between concepts in initialization ontology, and constituting a hierarchical ontology. In the research of extracting knowledge tuples, existing construction models lack an in-depth study for combining knowledge tuples with context information; this issue will affect the accuracy of tuple extraction directly. References [5, 10 13] put forward their own knowledge-tuple extraction method, respectively. Reference [10] extracts adjacent vocabularies as binary knowledge tuples in the text; this approach ignores the potential semantic relationships between nonadjacent vocabularies in papers. Reference [11] extracts knowledge tuples which are less than triple in the sentence based on pattern extraction algorithm. TextOntoEx [5] defines four modes of semantic pattern and divides the basic elements in semantic pattern into four categories: class, verbs, constants, and modifiers; then extracts knowledge tuples by pattern matching. Reference [12] proposed that the existing tuple extraction methods are generally based on shallow natural language process (NLP). The shallow NLP which is based on regular expressions will lose a lot of information in extraction; therefore, they use the Stanford depth NLP which is based on statistical parsing to extract knowledge tuples. References [5, 11, 12] all used the pattern extraction method which isbasedonnlp,butthelengthofthepatternanalysisis confined to a single sentence and ignores the impact from context of the paragraph. Reference [13] proposed that the existing patterns are extracted in a single sentence, ignoring the relationship between sentence and paragraph. First, they usethepaper stitleandotherstructuredinformationto build a kernel ontology with hierarchical relationship, then supplement the ontology with unstructured information and get the integrity ontology; this approach considers the context information in tuple extraction preliminary. But this method is limited by the accuracy of the papers titles and processing the sentence regardless of the structured information still needs more in-depth search. In the research of calculating the semantic similarity between concepts in ontology construction, existing construction models lack an in-depth study for structural similarity; this issue will lead to the concepts of ontology to only reflect the linguistic characteristics of words rather than the semantic features in the specific scene. References [11, 14 17] put forward their own semantic similarity measure. Reference [11] use the Chinese information source; they calculate the semantic similarity between concepts based on the Chinese semantic similarity tree from CKIP [18]. References [15, 16] measure the semantic similarity between concepts according to the words distance in WordNet dictionary. Reference [16] iterative improve the existing ontology through ontology merging technique; the relationship between concepts in initialization ontology and extended ontology also can be obtained by WordNet dictionary. The literature [11, 14 16] all used the linguistic similarity of individual word to measure the semantic similarity between words, ignoring the affect from knowledge tuples structure. Reference [17] introduced a set-similarity comparison in concepts similarity calculation. It achieved the adjacent concepts of objective concept and constituted an unordered set to measure the semantic similarity between concepts. This method considers the structural similarity in the condition of the adjacent nodes, but do not consider the similarity comparison in the view of mathematical model and still need more in-depth search. The above references do not consider the introduction of existing domain upper ontology into automatic construction; also they do not considered the research for structured-information-based knowledge-tuple extraction model. Therefore, we will research the IOT upperontology evaluation and selection in this paper. At the same time, we propose a context-information-based knowledgetuple extraction algorithm, a tuple-structure-based semantic similarity algorithm and a structured-information-based knowledge-tuple extraction model. 3. Upper-Ontology-Based Ontology Construction Model 3.1. Architecture of Ontology Construction System. In this section, we provide an overview of our architecture of ontology automatic construction. We describe the construction modules and data flows among different modules in Figure 1.We first research the construction models; ontology construction canbedividedintotwomodes:globalconstructionand incremental construction. In global construction, it needs to useallexistingdomainknowledgetorebuildtheontology when adding or changing knowledge. Global construction appies to a fully developed field; the knowledge of the field is complete and relatively static, without frequently adding or changing new knowledge. In incremental construction, when adding new knowledge, construction module will supplement and repair relevant parts in existing ontology according to the new knowledge. Incremental construction applies to an emerging field and the knowledge of the field is not complete. The field will put forward new knowledge frequently. Thus it requires a frequent iteration and updating of existing ontology. Compared with the global construction, incremental construction does not need to rebuild the ontology with all existing knowledge when adding new knowledge into the field; it only needs to update the relevant part with thenewknowledgeinontology.becausetheiotdomain has a mass of knowledge and is growing rapidly, incremental

36 4 International Journal of Distributed Sensor Networks Resources Tuple extraction Normal Vector group Unstructured Text Context + NLP Tuple with lab Extract rules Tuple Tuple Word stem Synonym Tuple Similarity LCS IC SSN New concept, property Compared vector Ontology Resource description Xml Tuple with lab WordNet Dic Upper ontology IOT ontology Figure 1: Architecture of ontology construction. construction will bring the advantages in computing and timeliness. Our ontology construction model is an incremental construction model. We iterate the ontology through adding new knowledge tuple to existing ontology constantly. In order to guarantee the connectivity of ontology and reduce redundancy concepts, we need to calculate the similarity betweenconceptsinnewtupleandexistingontologywhen adding a new tuple. Through merging similar concepts, we will achieve the connection between new tuple and existing ontology. The architecture contains four modules. In order to obtain the normalized domain knowledge from information source (IS), we use the tuple extraction module to extract tuples from unstructured and structured files. Then normal module will use word stem and synonym to normalize the vocabularies in tuples. In order to make full use of the most general concepts in IOT domain and support the interoperability with other ontologies, we use upper ontology module to evaluate existing IOT domain ontologies and select SSNasourupperontology.Thentakingintoaccountthatthe tuples are independent in their existence, we use the similarity module to calculate the structure similarity between concepts in new tuple and existing ontology. Then we merge the similar concepts and constitute a connected IOT ontology. (1) In the tuple extraction module, we extract knowledge tuples with context information from unstructured IS and resource description files. Unstructured IS is a conventional form to represent the knowledge in a domain, such as thesis and related web pages. Therefore, we can use the natural language processing (NLP) algorithms to extract tuples and get the normalized knowledge in unstructured IS. Existing ontology automatic construction models extract knowledge tuples from each sentence of unstructured IS. This approach neglects the context information in paragraph and paper. Therefore, we propose the CIB tuple extraction algorithm to extract tuples with context information from unstructured files. In addition to extract triple tuples from each sentence based on NLP algorithms, we will extract the core concepts of paragraphs and papers based on the largest contiguous frequency statistics. Then we make the core concepts as the context labels and add them into triple tuples hierarchically. For example, we extract triple tuples (a)-(b)-(c); in addition the core concepts of this paper and paragraph are A and B; we can combine them and get the thorough tuple (A B a)- (b)-(a B c) (each concept in tuple is reflected with an order vector); this approach will retain the context information with concept prefix. Resource description file is another conventional information carrier in IOT domain; it is used to describe various IOT resources such as device and data. In order to get more comprehensive knowledge, we propose the RDB tuple extraction model to extract tuples from resource description files. In this model, we make use of the resource description which conforms to thetreestructureandadoptsthekey-valuemodel, then formulates three rules in Section to extract tuples. (2) In the normal module, we use the stem algorithm and WordNet dictionary to extract the concepts word stem and synonyms in tuples and we achieve the standardized representation for vocabularies. Because a vocabulary can have a diversified statement, such as word activity, abbreviation, and synonym (e.g., measure is the stem for measuring, IOT is the abbreviation for internet of things, and evaluation is the synonym of measure), thus, the same concept will have variety of representations in tuples. This will cause the redundancy in ontology; therefore, we need to normalize the vocabularies in tuples. Stemming of the words is to remove affixes and reduce inflected (or sometimes derived) words to their stem. Synonym is a unified process for a vocabulary set which has the same meaning in linguistics; we use one word to represent other words in a vocabulary set (e.g., measure, quantity, amount, evaluate,and standard constitutes a synonyms set; we will use measure to represent other words in tuples). This operation will

37 International Journal of Distributed Sensor Networks 5 provide a unique and standardized form for each vocabulary in tuples. (3) In the upper-ontology module, we investigate the existing IOT domain ontologies and select SSN as our upper ontology. In order to make full use of the most general concepts in IOT domain and support the interoperability with other ontologies, we need to select IOT domain ontologies as our upper ontology. But existing automatic construction models mostly neglect the interoperability between ontologies and skip this step. We collect more than ten IOT domain ontologies such as SSN, CSIRO, MMI, and CESN. Through evaluating them in five aspects (key concepts, author, status, complexity, and cited), we choose the SSN as our upper ontology. (4) In the similarity module, we calculate the structure similarity between concepts in tuples and merge the similar concepts to constitute a connected IOT ontology. Because the knowledge tuples in tuple extraction modulearediscrete,thus,thetuplescannotconstitute a knowledge network and cannot be used for human or machines to understand the relevant knowledge of a concept. Existing approaches use the linguistic similarity of vocabularies to merge synonyms and neglectthescenethatthesamewordindifferentcontexts has different meanings and represents different concepts. Take into account that we add the multilevel context labs into tuples in tuple extraction module and each concept in tuples is expressed with a vector. Thus the structure of vector will reflect the context information of concepts in tuples; we propose the TSB similarity algorithm to calculate the structure similarity between concepts. Our ontology construction model is an incremental construction model. We iterate the ontology through adding new knowledge tuple to existing ontology constantly. Thus we need to extract contrast vector from ontology; then we need to make the concept in new tuple as object vector and use the longest common subsequence (LCS) between vectors to get structural similarity part between two vectors. Through calculating information content (IC) in structural similarity part, we can quantify the structural similarity between two vectors and provide evidence for merging similar concepts. Through merging the similar concepts, we can iterate the IOT ontology constantly. Therefore, a closed loop for ontology construction is formed in the proposed approach. We will expound each module and related algorithm in subsequent sections Knowledge-Tuple Extraction Algorithm. Knowledge tuples use concepts and relationships to describe the domain knowledge; they are the basic unit in ontology construction. The concepts in knowledge tuples will map the concepts in ontology; the loss and error of information in extraction process will be preserved permanently; thus, the tuple extraction model is an important part in ontology automatic construction. According to the structural degree of information sources in IOT environment, we divided the information sources into two categories: unstructured information and structured information. For unstructured information sources, we introduced the context information into tuples and propose the context-information-based (CIB) knowledge-tuple extraction algorithm; for structured information sources, we use the resource description as representative and propose the resource-description-based (RDB) knowledge-tuple extraction model Context-Information-Based Knowledge-Tuple Extraction. The unstructured information source refers to information that either does not have a predefined data model or is not organized in a predefined manner, such as text and web page, which is the most common type of information we contact. The unstructured information provides a knowledge source for tuple extraction. Existing ontology automatic construction models commonly use the natural language processing algorithms to extracted knowledge tuples from each sentence of information source. At the same time, they neglect the impact of the context information on tuples. Because the concepts in tuples will map the concepts in ontology, thus, the loss and error of information in extraction process will be preserved permanently. Therefore, we propose a context-information-based knowledge-tuple extraction algorithm. Our algorithm consists of two parts: obtaining the triple knowledge tuples through natural language processing algorithms and then in order to retain the context information in tuple, we extract the core concepts of paragraphs and papers based on the chapter s titles and largest contiguous frequency statistics and then we make the core concepts as the context labels and add them into triple tuples and retain the context information with concept prefix. In the first part, we use the natural language processing algorithm from Stanford CoreNLP to carry out syntactic analysis for each sentence. We will take raw English language text input and give the base forms of words and their parts of speech, whether they are names of companies, people, and so forth or normalize dates, times, and numeric quantities and we will mark up the structure of sentences in terms of phrases and word dependencies and indicate which noun phrases refer to the same entities. Stanford CoreNLP is an integrated framework, which makes it very easy to apply a bunch of language analysis tools to a piece of text. Its analyses provide the foundational building blocks for higher-level and domain-specific text understanding applications. For example, for the sentence: Common data formats include XML and json. The Stanford parser produces the following representations in Figure 2. Based on the syntactic analysis consequence above, we use Trunk (Trunk provide rules to extract subject-predicateobject tuple from syntactic analysis consequence) to extract triple tuples. In this example, we extract four tuples from syntactic analysis on the right of Figure 2.Thesetuplesusethe subject-predicate-object structure to carry knowledge and provide basic units for ontology construction. But each tuple in this part only carries the knowledge from one sentence,

38 6 International Journal of Distributed Sensor Networks Common JJ(amod) and CC(cc) Formats NNS(root) include VGB(perp) XML NN(pobj) json NN(conj) Data NNS(nn) property Format Data property Format Common include Format json include Format XML Figure 2: Syntactic analysis and triple tuples. we can use the tuples in Figure 2 as examples; xml is a data format, but we will never know if this format is used in sensor or gateway. Actually, the concept of a word is determined by the entire paper; thus, we introduce the context lab to retain paper s information into tuples. In the second part, we extract multilevel core concepts in paragraph and make them as prefix for concepts in tuples to retain context information. Here we can also use the long tuple to retain the context information in context. We use two tuples device-has-name-is-sensor and location-hasname-is-china, for example, device and location are the core concepts in two paragraphs; it has the same redundancy compared with (device name)-is-(device sensor) and (location name)-is-(location name china), but consider the common ontology format is owl and it inherited the triple description framework from rdf, thus, it will split the long tuple device-has-name-is-sensor and location-has-nameis-china into four triple tuples: device-has-name, name-issensor, location-has-name, and name-is-china. Because the center concept name, we will get the wrong inheritance between location and sensor. Thus we give up the long tupleandusethewayofconceptprefixintuples.themodels to extract core concepts from unstructured documents can be divided into two types, The first type makes the original chapter titles in documents as core concepts. When a document has clear chapter titles, titles are the summary of paragraphs by author. The parent-child relationship between titles are the reflection of expert knowledge which have high reliability. At thesametime,theextractionprocessissimpleandhasasmall computation overhead. The second type is to extract core concepts from chapters through data mining. Because some documents do not have chapter titles, we need to use data mining to extract the core concepts. the Centrality of graph to extract the core concepts from concepts map are commonly used. We use the degree centrality to extract 62 core concepts from 5 IOT scientific papers; more than 70% core concepts are thesamewiththechaptertitles;thus,degreecentralitycanbe used as a supplementary means to extract the core concepts. We use these core concepts as context labs and add these labs into concepts in tuples as multilevel prefix. Through the research on IOT scientific papers and technical documents, more than 95% of the levels of document sections are less than 5. In order to obtain the complete context information of each concept, we will extract core concepts for each paragraph basedonthemultilevelchaptersandconstructmultilevel prefixforeachconceptintuples.atthesametime,the calculated amount to maintain the maximum level is bearable (it will not cause the change in computational complexity). The titles of upper and lower sections meet the one-to-many relationship (parent-child relationship), in order to provide more information for the following similarity calculation, the concepts prefix can use ordered vector to retain the parentchild relationship information. Use the sentence in Figure 2 as an example; we can calculate the concept with largest contiguous frequency in second-level chapter as sensor and first-level chapter as IOT. We can extract the concepts in triple tuples as in Figure 3. We can achieve the CIB knowledge-tuple extraction as in Algorithm 1. At the same time, we open a test interface in our prototype system to extract tuples from unstructured information sources: show/nlptuple Resource-Description-Based Knowledge-Tuple Extraction. IOT environment includes various resource descriptions, which are a type of structured documents. Resource description has been used to describe various IOT resources such as device and data; we list a resource description for sensor in Figure 4; it describes various attributes and values for a Temp sensor. We use the Xively (previously Pachube, one of the most famous IOT platform) as an example; it has more than twenty thousand of resource descriptions and exceeds the number of characters in other pages in this website. This helps to obtain a more comprehensive knowledge for tuple extraction. Existing ontology automatic construction models commonly neglect this type of information source. Therefore, we propose a resource-descriptionbased knowledge-tuple extraction model. Compared with unstructured information, resource description has a tree schema to describe the structure of documents and uses the key-value pair like <Name>Temp sensor</name> in Figure 4. The tree schema has been constructed by the IOT experts and provides an explicit superior-subordinate relationship; at the same time, we can extract subjectpredicate-object tuples conveniently based on the key-value pair. According to statistical analysis, the vocabularies in structured information are more concise and close to the topic. This will reduce the redundancy for ontology from uncorrelated words. Figure4 is a resource description file in xml format; based on the W3C s definition document, xml is a markup language with key-value model and tree structure. An xml element is everything from (including) the element s start tag to (including) the element s end tag; xml elements can have attributes in start tag. The attributes provide additional information about an element; the key-value model of xml defines the relationship between element and value; the tree structure of xml defines the parent-child and brothers relationships between elements; it determines the one-to-many relationship (parent-child relationship in data structure) between upper and lower elements in resource description. Based on the research of resource description files in existing

39 International Journal of Distributed Sensor Networks 7 IOTsensorformat property IOTsensordata IOT sensor format property IOTsensorcommon IOTsensorformat include IOT sensor json IOT sensor format include IOT sensor XML Figure 3 Input: text Output: context tuples texts.each do paragraph i (from paragraph 1 to paragraph M ) paragraph i.each do sentence j (from sentence 1 to sentence N ) tuple = NLP(sentence j ) Put tuple into tuple group i. end Extract core concept as context lab i from tuple group i. Add context lab i into each tuple in tuple group i. end Algorithm 1: CIB tuple extraction. IOT platforms, the upper and lower elements in resource description mostly have the parent-child or concept-property relations. Considering the property is a partial description of theconcept,wecanbuildparent-childrelationshipbetween upperandlowerelements. We can divide the elements of resource description into four categories: tag-name (like Sensor and Name), this kind of elements is determined by xml schema and has a tree structure;inthesamexmlschema,tag-nameanditsstructure are relatively stable; tag-value (like Temp sensor and Huawei), this kind of element is determined by specific resource and often different between different xmls; tag-property (like Tags andid),thiskindofelementisdeterminedbyxmlschema and its structure is relatively stable in the same schema; tagproperty-value (like Temp and 12), the same as tag value, the elements are determined by specific resource and are often different between different xmls. Basedontheelementscategoryabove,wecananalyzethe relations between elements and divide the relations into four kinds: hierarchical relations between tag-names, they have been determined by the structure of xml schema; one-toone correspondence between tag-name and tag-value, they have been determined by specific resource; one-to-many correspondence between tag-name and tag-property, they have been determined by the structure of xml schema; one-toone correspondence between tag-property and tag-propertyvalue, they have been determined by specific resource. By integrating tag-name tag-property and tag-property tag-property-value, we can get three kinds of relations and define three rules to extract tuples from xml. (1) The relationship between upper and lower tag-names is Subclass. For example, for the tag-name Sensor and tag-name Name in Figure 4, wecanextract ternary tuple Sensor has subclass Name. (2) The relationship between tag-name, tag-property and tag-property-value in the same level is corresponding property. For example, for the nodes Sensor, Tags, and Temp in Figure 4, we can extract ternary tuple Sensor has Tags Temp. (3) The relationship between tag-name and tag-value in the same level is Value. For example, for the tag-name Name and tag-value TempSensor in Figure 4, we can extract ternary tuple Name has value Temp sensor. Based on the extraction rules above, we can extract triple tuples from resource description files. But triple tuples only retain the relationship between two layers and give up the multilevel parent-child relationships in resource description; this results in a loss of context information. Thus we need to add the context information into triple tuples and retain the context information with concept prefix. We extract the fourteen tuples in Figure 4 as in Figure 5. We can achieve the RDB knowledge-tuple extraction algorithm as in Algorithm 2. At the same time, we open a test interface in our prototype system to extract tuples from XML: Tuple Normalization. Tuple normalization converts words from tuples into normal format that can be used by the subsequent steps in ontology construction. It reduces inflected words to their stems, achieves synonyms of each word for each tuple, and finally normalizes the expression of synonyms that are the basis for further operation. In this study, the tuple normalization involves the two processes of word stem and synonym. (1) Word stem: stemming is the process of reducing inflected (or sometimes derived) words to their stem, base, or root form. For example, measure is the stem for {measuring, measured, measurement}.stemming of the remaining words is to remove affixes (i.e., prefixes and suffixes), such that the extraction of

40 8 International Journal of Distributed Sensor Networks <?xml version= 1.0 encoding= utf-8?> <Sensor Tags= Temp ID= 12 > <Name>Temp sensor</name> <Vendor>Huawei</Vendor> <Power> <AD>Null</AD> <CD>5V</CD> </Power> <Location> <Longitude>116.36</Longitude> <Latitude>39.94</Latitude> </Location> </Sensor> Name Temp sensor AC Power Sensor DC 5V Tags: Temp ID: 12 Vendor Huawei Longitude Location Latitude Figure 4: Resource description in IOT. Sensor subclass subclass subclass subclass Name Sensor Power Sensor Vendor Sensor Location Sensor Name value TempSensor subclass Sensor Power AC subclass Sensor Power DC AC value Null DC value 5V value Sensor Vendor Huawei subclass Sensor Location Longitude subclass Sensor Location Latitude value Sensor Latitude value Sensor Latitude Figure 5 concepts and relations containing syntactic variations of the terms is allowed. The Porter Stemming Algorithm [19] is used in this research to stem text. We open a test interface in our prototype system: (2) Synonym: synonym is a unified process for a vocabulary-set which have the same meaning in linguistics; we use one word to represent other words in a vocabulary-set. This operation will provide a unique and standardized form for each synonymvocabulary-set; it will provide convenient for the tuple similarity algorithm. In IOT environment abbreviation is the most common type of factors which generate synonyms, such as Internet of Things and IOT are considered as the same concept and their relations belong to the equal concept. WordNet dictionary provides a comprehensive collection of English synonyms and defines a unique identifier id for each word. Therefore we use WordNet as the standard, makingthewordwhichhastheminimumidineach synonym-vocabulary-setastheuniqueformforthe set. We open a test interface in our prototype system: IOT Upper Ontology. Upper ontology (also known as a top-level ontology or foundation ontology) is an ontology which describes very general concepts that are the same across one domain. Selecting an appropriate existing domain ontology as the upper ontology will make full use of knowledge in existing domain ontology. At the same time, the different ontologies which face different systems will have a unified upper ontology. The concepts in different ontologies can be built relations based on the unified upper ontology; it realizes the knowledge sharing across different systems. In this section we collect and collate existing IOT ontologies and formulate an assessment program and select the appropriate domain ontology as our IOT upper ontology in construction. IOT ontology has a rapid development in recent years. W3C and other organizations and individuals construct a dozen of upper ontology; all these ontologies are not for a specific IOT system; they describe IOT concepts and relationships from different angles. These ontologies generally make the device and observation as their core concepts, then expand the concepts based on the device type, capacity, properties and observation type, precision, andandsoforth.wecollectedthefamousiotontology in Table 1 andevaluatetheminfiveaspects(keyconcepts, author, status, complexity, and cited). The key concepts can reflect the emphasis and scope of concepts for an ontology; we will select the upper ontology based on coincident degree between key concepts and our IOT system. The author reflects the authority of ontology. The status of ontology includes developing, maintaining, and ceasing; it will reflect

41 International Journal of Distributed Sensor Networks 9 Input: xml Output: context tuples xml.each do element i (from element 1 to element M ) tuple = xml rules(element i ) Extract context lab into tuple. end Algorithm 2: RDB tuple extraction. Table 1: IOT ontology. Ontology name Key concepts Author Status Complexity Cited SSN Stimulus, sensor, and observation W3C Incubator Maintained Complicated Yes CSIRO Sensor and process. Michael Compton Developing Simple None MMI Device, capability, and property Luis Bermudez Developing Ordinary None CESN Sensor and physical property. Holger Neuhaus Cease Simple None SWAMO Platform, process, and observation John Graybeal Developing Complicated Yes A3ME Device, data, service, and capability Arthur Herzog Maintained Simple None OntoSensor Sensor, capability, and measurand Danh Le Phuoc Cease Complicated Yes OBOE Observation, context, value, and measurement Kevin Page Maintained Simple None SeReS Feature and result and observation Krzysztof Janowicz Developing Complicated Yes SemSOS Observation, process, feature, and phenomenon Cory Henson Maintained Ordinary None Sensei O and M Observation, data, process, and service Payam Barnaghi Cease Simple None OOSTethys Process, system, and observation Luis Bermudez Developing Ordinary Yes SERONTO Abstract, physical, and reference Laurent Lefort Cease Complicated Yes the degree of perfection and stability for an ontology; we will select the ontology which has a high degree of perfection and stable maintenance as our upper ontology. The complexity reflects the amount and granularity for IOT knowledge in ontology; we hope the upper ontology has a fine-grained and hierarchical knowledge system. The cited will reflect the degree of recognition in IOT domain. We analyze Table 1 and get conclusions as follows. These domain ontologies mostly make sensor and observation as their core concepts. In terms of author, only the SSN has been built by authoritative W3C Incubator; other ontologies all have been built by individual; there are four ontologies maintained state; they have a high degree of perfection and stable maintenance. In the aspect of complex, we tend to the ontologies with complicated ratings, these ontologies have a fine-grained description for domain knowledge and have been built with an integrated hierarchy. In the aspect ofquote,therearesixontologiesthathavebeenquotedby other ontologies and systems; they have a high degree of recognition.takingtheseconditionsabove,wechoosethe SSN [20] ontologywhichhasbeenbuiltbyw3csemantic Sensor Network Incubator group as our upper ontology. Central to SSN is the Stimulus-Sensor-Observation ontology design pattern. The pattern links sensors, what they sense, and the resulting observations, encompassing three of the four perspectives the missing system perspective is more about system organization and deployments than sensing but clearly links the pattern. The SSO has been developed as minimal, common ground for heavy-weight ontologies for the Semantic Sensor Web, as well as to explicitly address the need for light-weight semantics in the Linked Data cloud. Upper ontology will serve as the initial ontology for ontology automatic construction; we will supplement the upper ontology with normalized knowledge tuples through algorithms in the next section. Increasing the scope of ontology knowledge gradually achieves the semantic representation for IOT system finally Similarity Algorithm. In this section, we need to calculate the similarity between concepts in tuples and merge the similar concepts to constitute a connected IOT ontology. Because the knowledge tuples in tuple extraction module are discrete; thus, the tuples cannot constitute a knowledge network and cannot be suited for human or machines to understand the relevant knowledge of a concept. Existing approaches use the linguistic similarity of vocabularies to merge synonyms and neglect the scene that the same word in different contexts has different meanings and represents different concepts. Taking into account that we introduce multilayer context labs into concepts in Section 3.2, thus, we propose an aggregation similarity algorithm (TSB) to calculate the similarity between concepts. Similarity (ontology alignment) is an important research point in ontology construction; it will affect the accuracy of the concepts and relations in ontology directly. In general

42 10 International Journal of Distributed Sensor Networks [21], the similarity measure can be categorized in lexical, linguistic, and structural measures. (1) In lexical measures, there are a number of string distance methods such as Levenshtein distance [22], Jaro-Winkler distance [23, 24], and smoa distance [25]. And based on [25], the smoa distance is shown to be the most performing distance for the ontology alignment problem (the smoa distance takes into account both commonalities and differences characterizing the entities at issue). We took into account the following two questions: the relations between words string similarity and words similarity, computational complexity. First the words string similarity cannot fully reflect the similarity between words, such as apply, application, and apple. With smoa distance, apple has been closer to apply. But in fact, apply and application have the same stem apply and they are more similar. In addition, smoa distance increases the amount of calculation in a large number. Thus basedontheresearchoftwoaspectsabove,weused the stem treatment for all words and then with the fully matched comparison, we can get the similarity relations between lexical effective. (2) Linguistic measure computes the similarity between ontology entities by considering linguistic relations such as synonymy and hypernym. We used the synonym treatment which based on WordNet dictionary deals with the linguistic measure in Section 3.3. (3) The structural similarity measure generally refers to the hierarchy distance measure [21] (basedonthe subsumption relation between classes in ontologies). And it solved the scene that the same word in different contexts has different meanings and represents different concepts; we proposed two knowledge-tuple extraction methods in Section 3.2 and these two extraction methods all retained the structural information in knowledge-tuple. Based on these above, we propose a tuple-structure-based similarity algorithm (TSB) in Section Our ontology construction model is an incremental construction model. We iterate the ontology through adding new knowledge tuple to existing ontology constantly. Because the SSN (upper-ontology) covers the general concepts that are the same across IOT domain, our automatic construction model makes SSN as the initialize ontology, and then expands the concepts and relations through extracting new knowledge tuples. When we expand the existing ontology, we need to identify whether the concept from new tuple can be merged with existing concepts in existing ontology or added as a new concept based on similarity calculation between concepts. In structure similarity calculation, we transform the concepts with multilevel context labs into vectors and assess the structure similarity between vectors. Before we compare the structural similarity, we need to extract contrasted concepts based on linguistics from existing ontology. Therefore, we can divide the structure similarity algorithm into the following two parts: making the linguistic similarity as metrics and extract a contrasted concept-set in existing ontology which is similar to the concepts in new tuple, and then building a contrasted vector group based on the concept-set and making the concept in new tuple as object vector and using the longest common subsequence (LCS) between object vector andcontrastedvectortogetstructuralsimilaritypartbetween two vectors. Through calculating information content (IC) in structural similarity part, we can quantify the structural similarity between two vectors and provide evidence for merging similar concepts. Through merging the similar concepts and adding new concepts, we can iterate the IOT ontology constantly Contrasted Concepts Extraction. Contrasted concepts are the concepts of existing ontology which are linguistically similar to the object concept of new tuple; they will be used to calculate the structural similarity. Take into account that we add multilayer context labs into concepts in Section 3.2 and normalize the vocabularies in Section 3.3, thus,when we calculate the linguistics similarity between object concept and concepts in existing ontology, we only need to estimate whether the concepts have the same postfix. If two concepts have the same postfix, we consider that they are in linguistics similarity. Through extracting the contrasted concepts in existing ontology, we can calculate the structural similarity between object concept and contrasted concept, then merge the similar concepts or supplement new concepts into ontology. Based on the extraction algorithm in Section 3.2, each new tuple is composed of two types of elements: concept with hierarchy labels and relationship. We use the tuple (A B C 1 )-(P 1 )-(A B C 2 ), for example; it contains two concepts A B C 1 and AB C 2 ; each concept has two-level context lab A and B. We use the A B C 1 asobject concept. Because the linguistic similarity only relates to postfix C 1 ; thus, we will extract the concepts in existing ontology which have the same postfix C 1 andconstructthe contrasted concept-set. (In order to support the structural similarity comparison, when we add a new concept into ontology, we need to retain the context information for this concept in ontology. Thus we will assign a unique identifier for each concept in ontology. One unique identifier could correspond several express forms and each form will have the same postfix.) In order to facilitate the structure similarity calculation, we transform the object concept and contrasted concept into ordered vectors. Because one object concept could correspond to several contrasted concepts in ontology and each contrasted concept could have several express forms, thus the contrasted concept-set can be represented by a two-dimensional vector group. We use the object concept IOT sensor name as an example, we assume that there are three concepts named name with different ids in existing ontology; each concept has several express forms and we list the vector group in Figure 6. Eachrowinthevectorgroup reflects various express forms for a concept in ontology. Through the combination and calculation above, we transform the structural similarity comparison between concepts in new tuple and existing ontology into the comparison

43 International Journal of Distributed Sensor Networks 11 Ontology Device Name(379) ID: 379 [Sensor name] [device sensor name] [device name] Thing Location Name(267) ID: 267 [Location name] [description location name] Data Name(482) ID: 482 [Data datapoint name] [data list data name] [data name] Figure 6: Contrasted vector group. between object vector and arbitrary vector in a contrasted vector group Vector Similarity Algorithm. Through constructing the contrasted vector in Section 3.5.1, wetransformthestruc- ture similarity comparison between new tuple and existing ontology into similarity comparison between object vector and arbitrary vector in a contrasted vector group. In order to integrate structural similarity into similarity algorithm, we use the longest common subsequence (LCS) between vectors to get structural similarity part between two vectors. Through calculating information content (IC) in structural similarity part, we can quantify the structural similarity between two vectors and provide evidence for merging similar concepts in the following. The longest common subsequence (LCS) of two vectors states the similar parts of information provided by the two vectors. A proper quantification of the LCS of vectors improves the degree of structural similarity between vectors. References [26 28] put forward their own similarity algorithm that combine LCS and information content (IC). Reference [29] proposed evaluating the IC of the least common subsumer of the compared concepts (LCS(c 1,c 2 ))as the representative of this shared information. The LCS of a pair of concepts is the most specific common ancestor that subsumes them; it is found in the taxonomy to which they belong. If the two concepts are not taxonomically connected and the LCS does not exist, they are considered maximally different. Otherwise their semantic similarity is computed as theamountoficprovidedbythelcs. Consider sim [29] (c 1,c 2 ) = IC (LCS (c 1,c 2 )). (1) With [26] s metric, any pair of concepts with the same LCS will result in exactly the same similarity value. To better differentiate concepts, both [27, 28] also consider the IC of the compared terms into the equations. Reference [27]measures the similarity as the ratio between the common information between concepts (i.e., IC(LCS)) and the information needed to fully describe them (i.e., the IC of each concept alone). Reference [28]proposedcalculatingtheconceptdistance(the opposite of similarity) as the difference between the IC of each concept and the IC of their LCS. Consider sim [30] (c 1,c 2 )= 2 IC (LCS (c 1,c 2 )) (IC (c 1 )+IC (c 2 )), sim [26] (c 1,c 2 )=(IC (c 1 )+IC (c 2 )) 2 IC (LCS (c 1, c 2 )). (2) Here we need to calculate the similarity between tuple vector and arbitrary vector in the vector group. Through LCS(V 1, V 2 ),wecangetthenumberofthesameconcepts directlyanditreflectsthesameportionoftwovectors.but any pair of vectors with the same LCS will result in exactly the same similarity value. Therefore we need to measure the ratio between the common information and all information. In (1) all information is a fixed value and reflects the information of universal set. But in our algorithm all information only reflects the total information of two vectors and it will be changed with vectors. We can get vector similarity algorithm as follows: IC (LCS (V 1, V 2 )) = log ( LCS (V 1, V 2 ) ). (3) V 1 V 2 Then we need to merge similarity concepts based on vector similarity and threshold. According to the contrasted vector group in Figure 6, each row in the vector group reflects various express forms for a concept in ontology. We can divide the similarity computation results into three types: first type, there is not a vector similar to object vector in contrasted vector group, we will add the object concept into ontology as new concept; second type, there is only one row which has vector similar to object vector in contrasted vector group, we will merge object concept with the concept which corresponding to the one row, at the same time we add the new express form into this row in group; third type, there are more than one row which have vector similar to object vector in contrasted vector group, we need to merge all similar concepts and merge the corresponding express forms. In the third type, the new concept will change the partialstructureofexistingontology.thesimilaritythreshold defines the maximum similar distance between vectors, but each concept in ontology does not define its absolute standard express forms (centre); therefore, the similar space for a concept is a dynamic space with the change of concept s express forms. With the expanding of express forms for a concept, the similar space is also increasing. We can get the edge distance between multiple concepts based on the similar space of each concept in ontology; when the edge distance is smaller than the maximum diameter of the similar space of new express forms, the new express forms may cause the combination of multiple concepts in ontology and this is the occurrence of third type. With the extension of concepts and express forms in ontology, the edge distance between polysemous concepts will be reduced gradually. Heresomenoisetuples(Becauseofwiderangeofinformation source, it is difficult to ensure that each tuple complies with the basic knowledge of IOT domain; some artificial

44 12 International Journal of Distributed Sensor Networks Input: existing ontology,new concept Output: new ontology concepts(in existing ontology).each do concept i (from concept 1 to concept M ) If concept i.postfix = new concept.postfix concept i.forms.each do form j (from form 1 to form N ) similarity (new concept,form j )= log ( LCS (new concept,form j) new concept form j ) If similarity(new concept, form j ) threshold Put concept i into merged group. break end end end end If merged group.size = 0 Put new concept into existing ontology as new concept. elsif merged group.size 1 Merge new concept and all concepts in existing ontology into one concept. Merge all forms for one concept. end Algorithm 3: Structure similarity. nonstandard information will be used as tuples to iterate IOT ontology) may cause the collapse (since the individual noise information led to the concept of the ontology polysemy merge, resulting in loss of information) of partial ontology concepts. This paper only presents this possible problem and does not make a detailed exposition. In future studies we intend to supplement the merger control factor to control the merging between polysemy concepts in ontology. Wecalculatethecomplexityofthestructuresimilarity algorithm (Algorithm 3). When adding a concept, we suppose that the number of concepts in ontology is m, the number of concepts expression is n, and the length of the expression is t.we can calculate the timecomplexity as O(m n tlog(t)). Consider the length of the expression does not increase with the expansion of ontology; thus, t log(t) can be considered as a constant, and the time complexity can be simplified as O(m n). 4. Prototype and Evaluation In this section, we build a prototype system and collect files from six IOT platforms. These files include platform introductions, technical documents, and resource descriptions. We put these files and SNS ontology into prototype system, select an appropriate threshold, and get the IOT ontology. Then we evaluate the ontology based on the experts knowledge and evaluating system Prototype System. Weuserubyonrailstobuildthe prototype system and deploy it in ubuntu environment with two 2.26 GHz Xeon processor and 4 G of memory. Prototype system includes two main modules: basic components andcompleteprocess.wehavedevelopedmorethanten atomic functions in basic component. Include searching synonym, hypernym, and hyponym based on WordNet; removing stopwords; achieving stem based on [19]; segmenting Chinese Word; calculating tags based on random walk; extracting knowledge tuple based on stanford-corenlp, XML,and ontology; building ontology in owl format based on knowledge tuples. All these functions have been open with restful API and we list them in Table 2 (detail in By invoking the APIs in Table 2,weassembledseveral types of automatic construction processes which adapt to different needs in complete process module. These processes include the upper-ontology-based approach for automatic construction of IOT ontology in this paper ( Users can enter the url of upper ontology, xml, and text in web page; then system will build an ontology by invoking all atomic functions in Section Data Sources and Construction Result. We collect data source files which include platform introductions, technical documents, and resource descriptions from six IOT platforms. These files are expressed by text, web page, xml, or json (as a structured document, json can be converted to xml) and we classify the files based on data format and application scenarios in Table 3.The websites of six platforms are listed below the table: Xively, previously Pachube, is a web service which enables users to connect devices into platform and share data; Cloudsensing is our WOT platform; we put in various sensors, controllers, and open data in each layer; Evrythng, Thingspeak, and Yeelink are all IOT applications which store and retrieve data from things using RESTful API. Exosite has a function to generate strategy; we can formulate

45 International Journal of Distributed Sensor Networks 13 Table 2: Basic components API. Name URL Method Input Output Word segmentation POST text words Stem GET word stem Synonym GET word synonyms Hypernym GET word hypernyms Hyponym GET text hyponyms Remove stop words POST text words XML tuple POST urls tuples NLP tuple POST text tuples OWL tuple POST urls tuples OWL packaging POST tuples url Table 3: Data sources. Resource Type C1 C2 E1 E2 T Y Total Text Platform introduction Technical document Smart home Smart agriculture Resource description Energy monitor Air quality Ocean Observations Weather station C1: Xively: C2: Cloudsensing: E1: Evrythng: T: Thingspeak: Y: Yeelink: events and responds which have been triggered by device data in Exosite. The six IOT platforms have a similar IOT platform architecture; open resources in three levels: gateway, device, and data. We collected 14 texts and 502 resource description files from the platforms. In order to get the relations between the number of imported information sources with the IOT ontology, we divide all information sources into five parts and import them into prototype gradually. At the same time, we make the SSN as upper ontology and it will serve as initial ontology for ontology automatic construction, the concepts in SSN describes very general concepts that are the same across one domain; thus, we consider the concepts in SSN as unique (reflect the same concept in different contexts) in IOT domain, and the linguistic similarity will reflect the concept similarity. Wecangettherelation-tableasinTable 4. Based on Table 4, we can get the conclusions as follows. The number of concepts in ontology is increasing with the number of imported information sources, but the growth rate is reducing gradually. Because of the unceasing expansion of ontology,thereareagrowingnumberofconceptsinontology. At the same time, the probability of finding a structuresimilar concept in existing ontology with the new concept in tuple is increasing constantly; therefore, the growth rate of concepts is reducing. After importing the 14 texts and 502 resource description files, the IOT ontology contains 3045 concepts totally. These concepts cover general concepts in Table 4: Number of information sources and ontology. Number of elements in Number of information sources Stage ontology Text Resource description Concept IOT domain such as gateway, device, and observation. At the same time, they also reflect the features of six IOT platforms, such as event and alert. We use device and observation as centre to unfold related parts of ontology with protégéandgettheviewsinfigures7 and Ontology Evaluation. In order to evaluate the performance of the construction approach, we investigate the general method of ontology evaluation and define an assessment scheme to estimate the effectiveness of IOT ontology. There are three main approaches to ontology evaluation. (1) Gold standard evaluation: this approach compares an ontology with another ontology that is deemed to

46 14 International Journal of Distributed Sensor Networks Thing Timezone Active time ID Device type Device type Thing Interface Node Port Protocol Timestamp Observation (SSN) Device (SSN) Connections Devices list Resources Device info Location Type Model Longitude Resource Attribute Deploy info Manager Storage Alias Name Serial number GPS Address Owner Local coordinate Latitude Unit Url Power Vendor Illumination Figure 7: Construction result (device). Thing Time Value Alert name Reference event Event action Recipient Alert interval Feeds Entry Id Longitude Description Created at Latitude UpdatedAt ID Observation name Feed Elevation Channel Calculation Log Name Units Format RID Observation (SSN) Alert Current value Observation list Graph Device (SSN) Condition type Event Reference device Alias Subject Message Event name Comparison Constant Reference obervation Figure 8: Construction result (observation). be the benchmark. Typically, this kind of evaluation is applied to an ontology that is generated (semiautomatically or according to a learning algorithm) to assess the effectiveness of the generating process. Maedche and Staab [30] give an example of a gold standard ontology evaluation and they propose ways to empirically measure similarities between ontologies both lexically and conceptually based on the overlap in relations. These measures determine the accuracy of discovered relations generated from their proposed ontology learning system compared with an existing ontology but are not so useful outside thedomainofontologylearningbecauseifaknown gold standard ontology exists then there is no need to evaluate other ontologies. (2) Criteria-based evaluation: this approach takes the ontologyandevaluatesitbasedoncriteria[18]suchas consistency, completeness, conciseness, expandability, and sensitivity. It depends on external semantics to perform the kind of evaluation that only humans are currentlyabletodo,sinceitisdifficulttoconstruct automated tests to compare ontologies using such criteria [11]. These criteria focus on the characteristics

47 International Journal of Distributed Sensor Networks 15 Resources Xml All element Xml element Semantic IOT ontology Related concepts Evaluation Precision/recall Figure 9: Architecture of task-based evaluation. of the ontology in isolation from the application. So, while ontology criteria maybe met, it may not satisfy all the needs of the application even if some needs may correspond with the ontology criteria. (3) Task-based evaluation: this approach evaluates an ontology based on the competency of the ontology in completing tasks. In taking such an approach, we can judge whether an ontology is suitable for the application or task in a quantitative manner by measuring its performance within the context of the application. The disadvantage of this approach is that an evaluation for one application or task may not be comparable with another task. Hence, evaluations need to be taken for each task being considered. Based on the above three types of evaluation methods, we can get conclusions as follows: existing ontologies describe the basic concepts of IOT in a coarse-grained way, but because of the limitations of the artificial construct speed and rapid development of IOT technology, these domain ontologies lack a description for system features and emergingconcepts,andtheyarenotsuitablefordirectuseinour specific IOT system. Therefore they cannot support Gold standard evaluation as the Gold standard; the IOT ontology which we build contains thousands of concepts and relationships; it is too onerous for manual evaluation. And our IOT system resources are increasing rapidly with the equipments access to our IOT system; the ontology also is expanding continually. We need an automated ontology evaluation method to evaluate the existing ontology in real time; thus, criteria-based evaluation is not suitable for our assessment program; ontology can be used in many IOT scenes such as heterogeneous resource search and service development, and ontology-based semantic search is the basic task of various scenes; therefore, we can use the semantic search as the task for task-based evaluation, making precision and recall as the evaluation index, and achieving the automatic ontology evaluation.webuiltthetask-basedevaluationarchitecture in Figure 9. Evaluation architecture mainly contains the following three steps: importing a keyword from xml file which is to be searched, carrying out semantic expansion in IOT ontology based on the keyword, and calculating the precision and recall between semantic expansion concepts and all vocabularies in xml file which is to be searched. (1) In the first step, we choose the resource description files as the object to be searched. Resource description is a kind of structured xml document, widespread in the IOT systems, and is used to describe various Keywords Our ontology Table5:Precisionandrecall. Precision Contrastive ontology Our ontology Recall Contrastive ontology % 76.52% 34.43% 41.37% % 76.29% 54.36% 55.29% % 76.78% 70.29% 70.54% % 76.32% 78.73% 78.83% types of open resource. The resources search in IOT systems are performed for resource description files, thus making resource description as the searched files that will conform to the IOT scene. We select 1 3 words from resource description to simulate the user s search input and put them into second step. At the same time, all vocabulary set in resource description will be imported into third step as an ontology evaluation input. (2)Inthesecondstep,weachieveseveralconceptsinthe ontology which are corresponding to the keywords in first step, and then get their parent-child concepts as semantic expand set. Then we make the semantic expand set as an input and import it into third step. (3) In the third step, we evaluate the ontology based on all vocabulary set in resource description (G w ) and semantic expand set (G c ) andgettheprecisionand recall index. Consider Precision = Recall = G w G c (G w G c ) (G c G w ), G w G c (G w G c ) (G w G c ). In order to compare the ontology-based semantic search effects with different ontologies, we construct a contrastive ontology with traditional methods. We used the information sourcesin Table 3 to ensure knowledge consistency. Then we use the natural language processing algorithms to extracted triple tuples from information sources. Through merge synonyms in tuples with the linguistic similarity of vocabularies, we get the connected contrastive ontology. We randomly select 100 resource description files from information sources and make them as searching objects. According to statistical analysis with google, the situation which users enter 1 4 keywords to search an object cover more than 95%. Therefore, we will randomly select 1 3 keywords in each resource description and import the keywords into our ontology and contrastive ontology. Through getting theparent-childconceptswithin3jumpanddirectproperties with keywords in ontologies, we can get two semantic expand sets as G w1 (our ontology) and G w2 (contrastive ontology). Thenwecollectallvocabulariesinoneresourcedescription as G w andcalculatetheprecisionandrecallbasedonformula (4). We divide the evaluation table based on the number of (4)

48 16 International Journal of Distributed Sensor Networks Keyword Concept Figure 10: Architecture of task-based evaluation. keywords and two ontologies, calculate the average values of 100 resource descriptions, and get precision-recall in Table 5. Based on Table 5, we can get the conclusions as follows. Using the same ontology, precision remains unchanged as the keywords increase. The precision of our ontology is 11% higher than the contrastive ontology s. With the same ontology, recall is increasing with the number of keywords, but the growth rate is reducing gradually. With the same number of keywords, the recalls of two ontologies are the same. In the aspect of precision, because of the contrastive group using triple tuples and linguistic similarity to construct the ontology, some concepts inherited irrelevant upper concepts. We use a couple of tuples as example: device-has-name-issensor and location-has-name-is-china, if we use the way of triple tuples, we can get device-has-name, name-is-sensor, location-has-name, and name-is-china. After merging the concept name based on linguistics, the sensor will inherit device and location ; this will cause the decline in precision; in the aspect of recall, with the increasing of keywords number, we can get more parent-child concepts within 3 jump and direct properties in ontology; thus, the recall is increasing. At the same time, with the increasing of keywords number, it is more possible to have an intersection between concept sets with different keywords. We reveal this situation in Figure 10; red point represents the keyword and blue point represents the relevant concept in ontology; because of the intersection between sets, the number of blue points does not have a linear growth with red point. Therefore the growth rate is reducing gradually. 5. Conclusions and Future Work An IOT ontology can help users and computers to grasp the knowledge of IOT system. Rapid, accurate construction of ontology has become an important topic for researchers working on semantic web. This paper proposed a new architecture for ontology automatic construction: selected the appropriate existing domain ontology as the upper ontology, made full use of knowledge in existing domain ontology and reduced the number of extracting new concepts and relationships; proposed the context-information-based knowledge-tuple extraction algorithm and tuple-structurebased similarity algorithm and ensured the accuracy of concepts and relationships in ontology; proposed the resourcedescription-based knowledge-tuple extraction model and made full use of the structured information in IOT systems. We constructed a semantic search task-oriented ontology evaluation architecture and evaluated different IOT ontology construction architectures from two aspects: precision and recall. The construction architecture in this paper can improve the search recall well. Our opinions and methods in this paper could be a starting point for automates construction which utilize various information source in IOT. Conflict of Interests The authors declare that there is no conflict of interests regarding the publication of this paper. Acknowledgments The work is supported by the National 3rd Key Program project (no. 2011ZX ) and Project on the Architecture, Key technology research and Demonstration of Web-based wireless ubiquitous business environment (no. 2012ZX ). References [1] S. Steffen and S. Rudi, Eds., Handbook on Ontologies, International Handbooks on Information Systems, [2] S. Hachem, T. Teixeira, and V. Issarny, Ontologies for the internet of things, in Proceedings of the 8th Middleware Doctoral Symposium (MDS 11), [3] M. Compton, P. Barnaghi, L. Bermudez et al., The SSN ontology of the W3C semantic sensor network incubator group, Web Semantics: Science, Services and Agents on the World Wide Web, vol. 17, pp , [4] Work on the SSN ontology. [5] M. Y. Dahab, H. A. Hassan, and A. Rafea, TextOntoEx: automatic ontology construction from natural English text, Expert Systems with Applications, vol.34,no.2,pp , [6] R.-C. Chen, J.-Y. Liang, and R.-H. Pan, Using recursive ART network to construction domain ontology based on term frequency and inverse document frequency, Expert Systems with Applications,vol.34,no.1,pp ,2008. [7]X.Hou,S.K.Ong,A.Y.C.Nee,X.T.Zhang,andW.J.Liu, GRAONTO: a graph-based approach for automatic construction of domain ontology, Expert Systems with Applications,vol. 38,no.9,pp ,2011. [8] R. Subhashini and J. Akilandeswari, A survey on ontology construction methodologies, International Journal of Enterprise Computing and Business Systems,vol.1,no.1,14pages,2011. [9] I. Niles and A. Pease, Towards a standard upper ontology, in Proceedings of the International Conference on Formal Ontology in Information Systems (FOIS 01),2001. [10] C.-W. Shih, M.-Y. Chen, H.-C. Chu, and Y.-M. Chen, Enhancement of domain ontology construction using a crystallizing approach, Expert Systems with Applications, vol. 38, no. 6, pp , [11] C.-S. Lee, Y.-F. Kao, Y.-H. Kuo, and M.-H. Wang, Automated ontology construction for unstructured text documents, Data &KnowledgeEngineering, vol. 60, no. 3, pp , [12] A. Zouaq, D. Gasevic, and M. Hatala, Towards open ontology learning and filtering, Information Systems, vol.36,no.7,pp , 2011.

49 International Journal of Distributed Sensor Networks 17 [13] M. Kamel and N. Aussenac-Gilles, Ontology learning by analysing XML document structure and content, in Proceedings of the Knowledge Engineering and Ontology Development (KEOD 09), pp , Madère, Portugal, [14] Y. Jung, J. Ryu, K.-M. Kim, and S.-H. Myaeng, Automatic construction of a large-scale situation ontology by mining howto instructions from the web, Web Semantics: Science, Services andagentsontheworldwideweb,vol.8,pp ,2010. [15] D. Sánchez and A. Moreno, Learning non-taxonomic relationships from web documents for domain ontology construction, Data & Knowledge Engineering, vol.64,no.3,pp , [16] R.-C. Chen, C.-T. Bau, and C.-J. Yeh, Merging domain ontologies based on the wordnet system and fuzzy formal concept analysis techniques, Applied Soft Computing, vol.11,no.2,pp , [17] A. Maedche and S. Staab, Measuring similarity between ontologies, in Knowledge Engineering and Knowledge Management: Ontologies and the Semantic Web, vol.2473oflecture Notes in Computer Science, pp , [18] [19] M. Porter, An algorithm for suffix stripping, Program,vol.14, no. 3, pp , [20] W3C, The SSN Ontology of the W3C Semantic Sensor Network Incubator Group, [21] G. Acampora, V. Loia, S. Salerno, and A. Vitiello, A hybrid evolutionary approach for solving the ontology alignment problem, International Journal of Intelligent Systems, vol. 27, no. 3, pp , [22] V. I. Levenshtein, Binary codes capable of correcting deletions, insertions, and reversals, Soviet Physics Doklady, vol. 10, no. 8, pp , [23] M. A. Jaro, Probabilistic linkage of large public health data files, Statistics in Medicine,vol.14,no.5 7,pp ,1995. [24] W. Winkler, The state record linkage and current research problems, Tech. Rep., Statistics of Income Division, Internal Revenue Service Publication, [25]G.Stoilos,G.Stamou,andS.Kollias, Astringmetricfor ontology alignment, in Proceedings of the 4th International Semantic Web Conference (ISWC 05),pp ,2005. [26] P. Resnik, Using information content to evaluate semantic similarity in a taxonomy, in Proceedings of the 14th International Joint Conference on Artificial Intelligence, (IJCAI 95), C.S. Mellish,Ed.,vol.1,pp ,MorganKaufmannPublishers, Quebec, Canada, [27] D. Lin, An information-theoretic definition of similarity, in Proceedings of the 15th International Conference on Machine Learning (ICML 98), J.Shavlik,Ed.,pp ,Morgan Kaufmann, Madison, Wis, USA, [28] J. J. Jiang and D. W. Conrath, Semantic similarity based on corpus statistics and lexical taxonomy, in Proceedings of the 10th International Conference on Research in Computational Linguistics (ROCLING 97),pp.19 33,Taipei,Taiwan,1997. [29] OGC SWE, OpenGIS Observations & Measurements Implementation Specification,2007. [30] W3C, Web Services Description Language (WSDL) 1.1,2001.

50 Hindawi Publishing Corporation International Journal of Distributed Sensor Networks Volume 2014, Article ID , 18 pages Research Article An Effective Routing Protocol with Guaranteed Route Preference for Mobile Ad Hoc Networks Feng-Tsun Chien, 1 Kuo-Guan Wu, 2,3 Yu-Wei Chan, 4 Min-Kuan Chang, 2,3 and Yi-Sheng Su 5 1 Department of Electronics Engineering, National Chiao Tung University, Hsinchu City 30010, Taiwan 2 Graduate Institute of Communication, National Chung Hsing University, Taichung City 402, Taiwan 3 Department of Electrical Engineering, National Chung Hsing University, Taichung City 402, Taiwan 4 Department of Information Management, Chung Chou University of Science and Technology, Yuanlin 510, Taiwan 5 Department of Computer Science and Information Engineering, Chang Jung Christian University, Tainan City 71101, Taiwan Correspondence should be addressed to Yu-Wei Chan; ywchan@sslab.cs.nthu.edu.tw Received 7 November 2013; Revised 7 February 2014; Accepted 7 February 2014; Published 20 March 2014 Academic Editor: Luis Javier Garcia Villalba Copyright 2014 Feng-Tsun Chien et al. This is an open access article distributed under the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited. Battery power and queue conservation are critical issues in mobile ad hoc network (MANET). These two factors not only affect the delivery ratio but also the lifetime of the network. In this work, we will propose a simple and effective routing protocol to extend the lifetime and to evenly distribute the traffic loads of the networks as possible. Furthermore, a concept of serving capacity is introduced to reflect the level of congestion around a node. In this way, the proposed routing protocols can avoid network congestion and achieve higher packet delivery ratios. The extensive computer simulation is conducted to compare the proposed protocol against many existing routing protocols. The results show that the proposed routing protocol can have better performance in terms of queueing length, lifetime, and packet delivery ratio and have comparable end-to-end delays. 1. Introduction Recently, with the emergence of mobile applications [1 3], mobile ad hoc networks (MANETs) have attracted significant research activities due to their flexibility. In MANETs, wireless nodes form a temporary network without the help of any existing infrastructure and are allowed to move freely. Thus, two arbitrary nodes, which would like to communicate with each other, may not be in the radio range of each other. In this case, direct communication is not possible. Therefore, a routing path consisting of other nodes in the networks has to be established before the actual communication can be ensured. In view of this, the routing protocol becomes the key to the success of MANETs and has been an active field of research in MANETs [4 6]. In MANETs, mobile nodes are battery-powered and have limited queue size. When a node falls short of battery power, it would not be able to provide any service to other nodes in the network. As the number of such nodes increases, to some extent, the network becomes partitioned and communication among some nodes cannot be possible. On the other hand, when the queueing length of a node increases, the delay experienced by a packet arriving at this node increases as well. If no measure is taken, the overflow occurs and the packets start to get dropped. This degrades the performance of network. In light of these, balancing the loading and avoiding the lowbattery-power nodes to prolong the lifetime of the network are two important issues when designing a routing protocol for MANET. These two issues inspire two main categories of routing protocols, namely, the power-aware routing protocols [7 12] and the queueing-aware routing protocols [12 15]. The power-aware routing protocols take the available battery power of nodes in the network as a whole and pay moreattentiontohowtotapthisresourceinanefficient way and how to extend the lifetime of the network as much as possible. On the other hand, the queueing-aware routing protocols use metrics as functions of the queueing length for helping search the route. The protocols in this category mainly aim at distributing the traffic evenly over the network and avoiding the selection of those nodes with longer queueing length as much as possible. Both categories have their advantages and disadvantages. It would be more

51 2 International Journal of Distributed Sensor Networks desirable if a routing protocol can take both the factors of battery power and queueing length into account. The merits of taking both factors into consideration have been shown in [12]. In this work, we will propose a new metric to help us find a route with preferable remaining battery power level and queueing length. In addition, how to avoid the network congestion [15 18] is also essential when selecting a route for communication. Traditionally, the way to judge the level of contention is based on either the queueing length of a node [15, 16, 19], packet drops [19], or delay [17, 18, 20]. However, the available bandwidth around a node can also reflect the congestion around that node. The reason behind it is that the bandwidth in the wireless environment is a shared medium. In power-aware routing protocols or queueingaware ones, a node with higher remaining battery power or shorter queueing length would be selected with higher chance. However, if the available bandwidth around this node is little, the network in this area will become congested and this will consequently degrade the performance of the network. We will introduce a concept of serving capacity to prevent this from happening. Section 2 will introduce some related works. Section 3 shows the details of the proposed protocol. Section 4 presents the simulation results of the proposed protocol and other protocols. The concluding remark is given in Section Related Works In MANETs, many routing protocols were proposed and each of them adopted its own metric to help select a suitable route for a pair of nodes. Among these routing protocols, Ad hoc On Demand Distance Vector (AODV) [21] and Dynamic Source Routing (DSR) [22] arethemostcommonrouting protocols in MANETs due to their simplicity. AODV and DSR have their merits and demerits [23]. Since the proposed routing protocol is based on AODV, a brief description of AODV will be given below. In AODV, the route discovery of these routing protocols is achieved by flooding the route request (RREQ) messages towards the destination node. When an intermediate node receives a RREQ message for the first time, it creates a reverse link to the upstream node of the received RREQ message and rebroadcasts the RREQ message immediately. The later arrived RREQ messages will be dropped. When the destination receives the first RREQ message, it replies a route reply (RREP) message to the source node along the traversed route of the replied RREQ message and creates the forward link. If the later received RREQ message has less hop counts than the replied RREQ message, the destination node replies a RREP message again to tell source node using the shorter one. In this way, a route with the minimum hop counts can be found. However, AODV does not take other factors such as the remaining battery power and the queueing length into consideration when selecting a route for a pair of nodes. The objective of the power-aware routing protocols is to prolong the network lifetime. The lifetime can be extended by avoiding those nodes with low remaining battery power and directing the traffic to those nodes with abundant remaining battery power. To this end, much of the literature [7 10, 24] proposed avoiding selection of the low-battery-power nodes. In these protocols, the mechanisms of route selection are modified versions of [21, 22]. In the route searching, when an intermediate node receives a RREQ message, it delays for a while to collect RREQ messages and selects the RREQ message with minimum cost. The cost of current node is calculated and the cost of the RREQ message will be updated. Then, the node rebroadcasts the RREQ message with minimum cost. The destination also selects the RREQ message with minimum cost and replies an RREP message along the traversed path of selected RREQ message. The main difference of the proposed protocols is the designofthecostfunction.in[7, 8], the cost of the RREQ message is the inverse of the summation of the remaining battery power of the traversed nodes. For [9, 10], the cost is the inverse of the minimum remaining battery power of the traversed nodes. Using either type of the cost, a route with more remaining battery power can be found and the lifetime of the network can be prolonged. However, the main disadvantage of the power-aware routing protocol lies in the unbalanced usage of battery power of a node. The extension of the network lifetime is achieved at the expense of extensive usage of battery power of those nodes favored by the metrics adopted in those protocols. Another category of prolonging the lifetime in the network is to balance the loading of the nodes according to the queueing length of the nodes [13, 14]. The rationale behind the queueing-aware routing protocols is that the bottleneck of the network happens at nodes with high load and with long queueing length. By distributing load more evenly throughout the entire network, hot spots can be reduced and the lifetime of the network is expected to extend. The assumption behind this is that the battery power of each node in the network is similar. When the remaining battery power of nodes is highly diverse, the advantage of queueingaware protocols is compromised. Thus, it is sensible to adopt a routing protocol considering both the remaining battery power and the queueing length of nodes in the network. The work in [12] proposedaprotocolwhichtakesboth remaining battery power and queueing length into consideration. They found that the lifetime can be further improved when using a metric as a function of both remaining battery power and queueing length. However, in these protocols [7 10, 12 14], the available bandwidth around a node is not considered when choosing a route for a given communicating pair.thus,itmayrunintoasituationwhereanodehas abundant remaining battery power and available queue size but little available bandwidth is chosen to provide service. This will cause extra contention, incur unwanted packet drops, and, as a result, induce additional delay. To avoid this, in this work, we will adopt a metric called serving capacity tohelpfindaroutewithmoreavailablebandwidth.by considering the battery power, queueing length, and serving capacity, the proposed routing protocol can enjoy higher delivery ratio, shorter queueing length, longer network lifetime, and comparable end-to-end delay when comparing with the existing protocols.

52 International Journal of Distributed Sensor Networks 3 3. The Proposed Routing Protocol We assume each node in the mobile ad hoc network is powerlimited and has limited queue size. Here, let P i be the remaining battery power and Q i be the current queueing length of node i. P i and Q i are normalized by the maximum battery power and maximum queue size, respectively. We assume that the maximum battery capacity and the maximum queue sizeofeachnodeinthenetworkarethesame.weletthe maximum receiving rate and the maximum transmission rate of node i be R s i and R t i,respectively.rs i and R t i may not be the same due to the effect of the channel. R s i and R t i can be determined by first evaluating the SNR of received signals from the upstream node and the downstream one and adopting the Shannon capacity to estimate the receiving and transmission rate. The proposed routing protocol takes three factors into account. The first one is the transmitting or receiving capability of a node, the second one is the minimum battery power ofaroute,andthethirdoneisthemaximumqueueinglength of a route. The first factor reflects the availability of a node. When either the transmitting or receiving capability of a node cannot support further request, a node will not participate in the current route search. In this work, a serving capacity of a node is defined in the following section to judge whether a node is able to respond to route requests. To extend the lifetime of a chosen route, it is necessary to search for a route whose minimum battery power should be as large as possible. In view of this, the second factor is taken into account when designingtheproposedprotocol. However, with first two factors only, we can imagine that those nodes having better serving capacity and higher batter powerwillbechosenmoreoftenthanthosewhodonot.in this case, the queue could be built up at those nodes and this would cause excessive queueing delay at those nodes. Toavoidthis,theproposedroutingprotocolalsowouldlike to find a route whose maximum queueing length can be minimized as much as possible at the same time. In the following subsections, we will first give the route selection criterion of the proposed routing protocol and its reason behind following the route search mechanism and route maintenance mechanism Route Selection Criterion. In this subsection, we will present the route selection criterion and the reason why we choose such a criterion. We first define Λ i = max{min{r t i r i,r s i r i }, 0} as the serving capability of node i, wherer i and r i are the current transmission rate and receiving rate, respectively. The defined serving capacity is used to reflect the transmitting or the receiving capacity of a node. When the serving capacity is zero, it means that a node cannot support additional incoming or outgoing traffic. When this happens, a node cannot participate in any further route requests. In addition, since serving capacity indicates the level of traffic load of node i, it can also reflect the level of congestion around anode.ifanodehasasmallerservingcapacity,thismeans that either its available outgoing rate or its available incoming rate is smaller. Compared to the regions around the nodes with larger serving capacities, the level of the congestion of theregionaroundthisnodewouldbeseverer.consequently, ifmoretrafficisinjectedintothisregion,itwouldbeeasier to get congested than the regions around those nodes having larger serving capacities. Therefore, the serving capacity in this work is used to achieve two purposes: the first one is to decide whether a node can participate in a route discovery and the second one is to be used to reflect the level of congestion around a node. Oneofthegoalsoftheproposedroutingprotocolisto choose a route with less chance of the occurrence of congestion. As mentioned above, the serving capacity can be used toreflectthecongestionlevel.basedonthis,underwhat conditions will a route have less chance of having congestion? First, let us see what happens if the serving capacities of nodes along a route are similar. When the serving capacities of nodes along a route are close, with high chance, the levels of congestion of nodes along the route would be similar as well. Nevertheless, under this situation, if the maximum servingcapacityofthisrouteissmall,thelikelihoodofhaving congestion along the route would be high. In addition, multiple points of bottleneck could occur along this route. Thus, if we would like to choose a route to avoid congestion occurring at multiple nodes, the chosen route has to fulfill two conditions in terms of serving capacity. The first condition is that the maximum serving capacity of thisrouteshouldbeaslargeaspossible.thesecondoneisthat the serving capacities of nodes along the chosen route should be as similar as possible. This condition would lead to the ratio of the minimum and the maximum serving capacities being as large as possible. To achieve these two conditions, we adoptthefollowingmetrictojudgewhetheraroutecanpossibly fulfill these criteria. Let Ψ i be the set of routes perceived by node i for a given communicating pair and let Ψ ij be the jth route in Ψ i, which contains the information of the traversed nodes up to node i. The proposed metric is defined as ( Λ min Ψ ij Λ max Ψ ij ) exp Λ max Ψ ij / Λ max Ψ i, (1) where Λ min Ψ ij = min κ Ψij Λ κ, Λ max Ψ ij = max κ Ψij Λ κ, and Λ max Ψ i = max j max κ Ψij Λ κ.themetricin(1) servestwo purposes. The first term in (1) is called the service ratio of the jth route perceived at node i and is used to judge the closeness of the minimum and maximum serving capacity of are close, the ratio between these two quantities will be close to one. Thus, as the service ratio becomes larger, the difference between the minimum and maximum serving capacity normalized by the maximum serving capacity becomes smaller. However, using this ratio alone will lead to the a given route up to the current node. When Λ min Ψ ij and Λ max Ψ ij situation of choosing a small Λ max Ψ ij while Λ min Ψ ij /Λ max Ψ ij is large. In addition, the purpose of the second term in (1)istoprevent this situation from occurring too often. It is clear that the second term is large when Λ max Ψ ij is close to Λ max Ψ i.thisfactor will help us choose a route whose maximum serving capacity is close to the overall maximum serving capacity of all perceived nodes at an intermediate node or at the destination.

53 4 International Journal of Distributed Sensor Networks Combining the above two terms, the metric would have the following properties. First, if two or more potential routes havethesameserviceratio,thesecondtermwillguarantee that a route with higher maximum serving capability will be selected. This means that the traffic will be directed to a route with less traffic and the congestion can be prevented. Secondly, if two or more potential routes have the same second term, the first term ensures a route with higher service ratio and it will be chosen. That is, the route with smaller discrepancy in the minimum and the maximum serving capacities will be favored against those routes with larger discrepancy when the second terms of (1) of different routes perceived at a node are the same. Thus, under this situation, this metric routes the traffic towards area with lighter load and bypasses the congestion region. These two properties suggest the proposed metric distributes the traffic load evenly throughout the entire network by way of congestion speculation based on the serving capacities of nodes of routes. The metric in (1)onlyservestochoosearoutewithlarger Λ min Ψ ij /Λ max Ψ ij and larger Λ max Ψ ij as much as possible. However, the preference of a route in terms of battery power and queueing length is not well addressed. To guarantee the route preference, we adopt the following weighting function to help choose a route whose minimum and maximum serving capacity can be close while its preference can be fulfilled. In this work, the weighting function for the jth route perceived at node i is defined as f(p j,min,q j,max )= (Pj,min P TH ) 2 (Q TH Q j,max ) 2 (1 P TH ) 2 (Q TH 0) 2, (2) where P TH and Q TH are the preference of a route, P j,min is the minimum battery power of the jth route, and Q j,max is the maximum queueing length of the jth route. Also, f i (1, 0) = 1, f i (P TH,Q i ) = 0,andf i (P i,q TH ) = 0.This weighting function gives a route whose minimum battery power and maximum queueing length are far away from the respective thresholds, namely, P TH and Q TH. In this fashion, this proposed weighting function can help further differentiate routes in terms of their minimum battery power and the maximum queueing length. With the proposed weighting function and the metric, we define the cost of the jth route at an intermediate node, say node i,as J(P j,min,q j,max,λ i ) Λ min Ψ ij = f(p j,min,q j,max ) [( Λ max ) exp Λ max Ψ ij / Λ max Ψ i ]. Ψ [ ij ] A route for a given communication pair is chosen if that route can yield the minimum cost. By incorporating this weighting function, we hope to find a route whose service ratio and minimum power can be as large as possible and the maximum queueing length can be as small as possible. To show the capability of the proposed route selection criterion in (3), we conduct a theoretical simulation of the proposed criterion along (3) with other existing criterions such as Max-Min power [7, 10], Min-Max queueing length [13, 14], Max average power [9], andminaveragequeueinglength[12]. The details of these criteria for comparison can be found in the respective reference. In the simulation, we assume that the number of perceived routes at a node is 10 3.Inthisway,wecanobtainthe asymptotic behavior of those criteria. The remaining battery power and the queueing length of a node are randomly chosen between 0 and 1. Then we observe the behavior of the service ratio, the minimum remaining battery power, and the maximum queueing length of a chosen route under different route lengths. These results are shown in Figures 1(a) 1(c). Figures1(a) 1(c) show that the proposed selection criterion can select a route to find a balance point among the service ratio, the minimum remaining battery power, and the maximum queueing length. Furthermore, from Figures 1(a) 1(c), though Max-Min power and Max average power have better minimum remaining battery power than the proposed criterion, their maximum queueing lengths are higher than the proposed criterion. Thus, the advantage of better minimum remaining battery power of these two criterions is compromised by having longer maximum queueing lengths. Longer maximum queueing length means more battery power will be spent on transmitting queued packets. Another example of the route chosen by the proposed criterion is illustrated in Figure 2. AsshowninFigure2, the proposed criterion chooses a route, S-5-6-D for the source node S and the destination node D. Itisclearfromthis example that the proposed selection criterion can choose a route to avoid the congested area of the network Route Search Mechanism. The purpose of the proposed routing protocol aims at finding a route to have minimal service ratio, larger minimum remaining battery power, and smaller maximum queueing length as much as possible. To this end, the proposed cost function plays an essential role. As seen in the previous subsection, the proposed cost function is able to achieve these goals to some extent. Next, we would like to describe the proposed routing protocol, which is comprised of the route search mechanism and the route maintenancemechanism.westartwiththeroutesearchmechanism. It is worthwhile to mention at this point that the proposed route search mechanism is based on AODV [21] withmodifications. The way to handle the creation of reverse link and forwardlinkandroutereplyisthesameasaodv[21]. Assume that node s wouldliketoestablishalinkwith node d. Nodes sends a route request (RREQ) packet out to seek a route. Initially, this RREQ packet contains the IP address of node s, the destination IP address, the address of node d, source sequence number (SSN), the request transmission rate, the delay constraint, W TH,theconstraint of minimum queueing size, Q TH, the constraint of the minimum remaining battery power, P TH,thetimestamp, T RREQ, the current maximum serving capability, and the current minimum serving capacity. T RREQ records the time when node s generates its RREQ packets. Upon receiving a new RREQ packet, node i starts a predefined waiting window to see if there are more RREQ packets with the same SSN, source, and destination. Further

54 International Journal of Distributed Sensor Networks 5 1 Service ratio 1 Minimum battery power Route length Route length (a) Maximum queueing length (b) Route length Max metric Proposed criterion Maxmin power Minmax queueing length Max average power Min average queueing length (c) Figure 1: The comparison of the proposed cost function in (3) against other criteria. RREQ packets are dropped after the expiration of the waiting window. Once this waiting window expires, node i first checks whether it is able to participate in this route search by comparing its current remaining battery power and the queueing length against P TH and Q TH,respectively.Ifoneor both constraintsare violated, node i drops this RREQ packet. Otherwise, node i calculates its serving capacity Λ i to see whether it can accommodate the traffic of node s. Ifeither Λ i =0or Λ i is smaller than the request transmission rate, node i simply ignores the route request of node s. Ifnodei satisfies all these requirements, it computes the costs of all potential routes that it perceives during the waiting window forthesamessn,source,anddestinationandchoosestheone with minimal cost. After this, it updates the current maximum serving capacity and the minimum serving capacity and broadcasts the RREQ packet with minimal cost. Once the RREQ packet is sent, node i creates a reverse link towards its upstream node of the sent RREQ packet and starts a RREQ timer same as AODV [21]. If the route reply is not received before the expiration of the RREQ timer, node i neglects this route request from node s. When the route request packets arrive at node d,sameas node i, it begins a predefined waiting window to collect as many route requests as possible. Node d will find the costs of all routes arriving during this waiting window and choose the one with minimal cost. Once the route with minimal cost is

55 6 International Journal of Distributed Sensor Networks 2 P: 0.8 Q: 0.4 S: 0.2 D Proposed: S-5-6-D Maxmin power: S-1-2-D Minmax queueing length: S-3-4-D Max average power: S-1-2-D Min average queueing length: S-3-4-D Maxmin serving capacity: S-3-6-D Max average serving capacity: S-3-6-D 1 Battery power S P: 0.7 P: Q: 0.6 Q: 0.2 S: 0.2 S: 0.2 P: Q: 0.3 P: S: 0.7 Q: 0.1 S: 0.7 Queue Proposed P: Q: 0.2 S: 0.6 P: remaining battery power Q: queueing length S: available serving capacity Figure 2: An illustration of the route chosen by the proposed criterion. chosen, node d will reply the route request from node s along thechosenroute.uponreceivingtheroutereplyfromnoded, each node along this route allocates the resource requested by node s.when node s gets the route reply from node d,these two nodes can start to communicate Route Maintenance Mechanism. The chosen route could fail to function normally either when one or more nodes are outoffunction,whenthecommunicationchannelisbadfor transmission, or when they are unable to provide services due to lack of available battery power or available queue size. To deal with these three situations, we propose the following route maintenance mechanism shown below. For the first situation, let node k bethenoderightbefore the first broken link along the route. Under this situation, node k on behalf of node s initiates a route search starting from itself to the destination node. The route maintenance packet generated by node k contains the address of node s,the destination address, the address of node k,the request transmission rate, delay requirement, current time stamp of node k minus the delay, τ sk,fromnodes to k, themaximumand minimum serving capabilities set to the serving capability of node k, andthepartialinformationofthisrouteuptonode k. Each node receiving this route maintenance packet treats thisasanrreqpacketandfunctionsaccordingly.atthe node d, when it receives the route maintenance packets from node k, it functions as in the route search stage but replies the information of the new chosen route to node k instead of node s.whennodek obtains this reply, node k continues to forward the packets from node s. At the same time, when the route initiated by node k is found, node d also initiates a notification packet back to the node s along the newly repaired route. The notification packet will collect the information of the maximum and minimum serving capacities, the minimum remaining battery, and the maximum queueing length when it traverses this route. When node s receives the notification from node d, it compares the minimum remaining battery power or the maximum queue length of this notification. If the minimum remaining battery power is smaller than the power threshold, P TH, or the maximum queue length is greater than the queueing threshold, Q TH,thenodes initiates a new route search. Otherwise, it can decide to adopt this new route or to initiate a new route search according to a probabilistic function. The larger value of the probabilistic function means that the chance to adopt this new route would be higher. If node s decides to initiate a new route search, it generates a new RREQ packet with new SSN to discover a new route. For the second situation, also let node k be the node right before the link with bad channel condition along the route. It is known that wireless fading channel is time-varying [25, 26]. Thus, when a channel is in bad condition, with high chance, it would return to the good condition after some time duration.evenwhenthechannelisinbadcondition,the reliable communication is still possible by giving the packet more protection at the expense of reduction in transmission rate. Therefore, in this situation, the route cannot be taken as a broken route completely. The proposed route maintenance mechanism under this situation is as follows. When N 1 consecutive NAKs for the same packet are received from the next hop, node k perceives the channel is bad in all likelihood. When this happens, node k takes the following actions. (i) Node k choosestoreducethecoderateofthechannel coding to a lower value and sends the same packet again. It is possible that the original packet will be divided into smaller packets due to the change of the code rate. If a NAK is received again, node k reduces thecoderatefurther. (ii) If N 2 consecutiveacksforaseriesofpacketsare received after the reduction of code rate, node k increases the code rates until the normal code rate. (iii) If N 3 consecutivenaksforthesamepacketare received again after the reduction of code rate, node k

56 International Journal of Distributed Sensor Networks 7 initiates a route maintenance packet and takes action like in the first situation. In this situation, N 1, N 2,andN 3 are chosen to minimize the need of seeking a new route and avoid resource wasting as much as possible. N 1 has to be chosen to minimize the false alarm of link failure without introducing excessive delay of apacketasmuchaspossible.whenaseriesofacksare received, this means that the channel state is getting better alongthetime.thecoderateshouldbeadjustedtonormal code rate as soon as the channel state returns to the original channel state before the reduction of code rate. Thus, N 2 has to do with the turnaround time of the fading channel. N 3 is chosen to avoid excessive wait for the channel state to return totheoriginalchannelstatebeforethereductionofcoderate. Note that N 1, N 2,andN 3 are design parameters. For the third situation, when the new route is found, the destination initiates a notification packet back to the source periodically. It records the minimum remaining battery power and the maximum queueing length of the route. When the source receives the notification packet, if the minimum remaining battery power is smaller than P TH or the maximum queueing length is greater than Q TH, two actions are taken in this situation: (i) lack of available battery power: when the source receives the notification of the lack of the available battery power. It has no choice left. The source initiates a new route search to seek a new route; (ii) lack of available queueing buffer: if the source node receives consecutive N 4 notifications of the lack of queueing buffer, the source initiates a new route search to seek a new route. It is possible that a node originally with insufficient queueing buffer can digest the queued packets timely and its queueing length can meet the constraints once again. Spending time waiting for that to happen costs less than initiating a new round of route discovery. Note that N 4 is the system parameter such that the impact of lack of available queueing size on the system can be minimized as much as possible. 4. Simulation Results 4.1. Simulation Setup. Our simulation is evaluated using QualNet 4.5 simulator. Simulations are running over a 1000 m 1000 m area. The number of mobile nodes is topologies of the network are randomly generated. For each topology, the initial battery power and queueing length of a node are randomly generated and the simulation results are the average over all realizations. The mobility model we use is the random waypoint model. The nodes in the networksareeitherstaticormovingatvariousvelocities (1 m/s, 5 m/s, and 20 m/s). There are 25 CBR sessions for each topology and the data rates are 2, 1, and 0.5 packets/second, which correspond to high load, medium load, and low load, respectively. In the simulation, each packet size is 512 bytes. FIFO queue is adopted, the maximum queue size of a node is 3000 bytes, and the queueing model is assumed to be an M/M/1/k. In the simulation, IEEE b is utilized with rate of 2 Mbps. The transmission power is 15 mw and the radio range is 250 m. In addition, the battery model is a linear model. The power consumption is the duration of the packet transmission (or the reception) times the power of transmission (or the reception). In addition, the lifetime in this work is defined as the n-of-n lifetime, which is the time whenthefirstnodedies[27]. The power-aware routing protocols to be compared against the proposed one are PAR-AODV [9], LPR-AODV [8], and PSR-AODV [7] routing protocols. The queueingaware routing protocols to be compared against the proposed one are AODVM [13], WLBR [12], and QoS-AODV [14]. In addition, we also compare the proposed protocol with three extreme protocols. One aims at finding a route whose maximum remaining battery power is the largest among all the maximum remaining powers of all perceived routes at the intermediate nodes and the destination and it is called max battery power. The other aims at finding a route whose minimum queueing length is the smallest among all the minimum queueing lengths of all perceived routes at the intermediate nodes and the destination and it is called min queueing length. Another one is to find a route whose metric is the largest among all the maximum remaining powers of all perceived routes at the intermediate nodes and the destination and is called max metric. These three protocols serve as the benchmarks to see how good a newly route is in terms of remaining battery power, queueing length, and service ratio, respectively Simulation Results of Route Discovery Only. In this subsection, we will focus on the comparison of the battery power and serving capacity of a newly discovered route among the proposed and other existing protocols under the condition that the network size is fixed to 1000 m 1000 m. First, we look at the maximum, minimum, and average battery power of a newly discovered route, which are shown in Figures 3(a), 3(b), and 3(c), respectively. The proposed protocol is able to find a route which has the largest minimum battery power, the average battery power, and the good maximum battery power among all protocols. The proposed protocol has comparable performance to max battery power in terms of the maximum battery power. However, the minimum battery power of the route found by max battery power is much worse than the proposed protocol. Next, let us turn to the serving capacity. We will compare the maximum and minimum serving capacity and the service ratio. The results are shown in Figure 4. Asillustratedin Figure 4, all the protocols have similar maximum serving capacity. However, since the proposed protocol has the largest minimum serving capacity, the service ratio of the proposed protocol is the best among all the protocols. This means that the proposed protocol is capable of discovering a route whose serving capacity is more consistent. The consistency in serving capacity of the chosen route helps distribute the traffic load of the network more evenly. The last two things wewouldliketoshowarethecomparisonofthehopcount andinitialdelayofanewlydiscoveryroute.figure5 illustrates the comparison of the hop count of a newly discovery route

57 8 International Journal of Distributed Sensor Networks Normalized battery power Nodes (a) Maximum battery power Normalized battery power Nodes (b) Minimum battery power Normalized battery power Nodes Proposed AODV PAR-AODV LPR-AODV Max battery power (c) Average battery power Min queueing length WLBR AODVM QoS-AODV Max metric Figure 3: Comparison of the maximum, minimum, and average battery power. among all protocols. This figure shows the proposed protocol is able to find a route, which yields the smallest hop count among all protocols. Meanwhile, in Figure 6, we can see that the route found by the proposed protocol has the shortest initial delay. The reason for this is that the proposed protocol can find a route smallest hop count, which compensates the negative effect of the delayed broadcast at nodes during the phase of route discovery Simulation Results of Overall Proposed Protocol: AWGN Channel. Figure 7 showsthesimulationresultsunderthe high traffic loading under AWGN channel. The results of average queueing length of all protocols are shown in Figure 7(a). Theproposedprotocolenjoystheshortestaverage queueing length. However, as shown in Figure 7(e), the shorter average queueing length of the proposed protocol does not necessarily imply that the proposed protocol will havethebestend-to-enddelay.theproposedprotocolhasthe best end-to-end delay only when the moving speed of node is high. When the moving speed is low, the proposed protocol is worsethanaodv,wlbr,andqos-aodv.thecomparison of lifetime among all protocols is given in Figure 7(b). The proposed protocol has the longest lifetime among all protocols. However, the lifetimes of the rest of the protocols are similar. This might be due to the fact that the metrics adopted in these protocols favor some certain nodes and, thus, create unwanted hot spots at these nodes. Consequently, these nodes quickly run out of their battery power, which results in comparable lifetime of these protocols. Next, we will look at the delivery ratio of these protocols presented in

58 International Journal of Distributed Sensor Networks 9 Normalized serving capacity Nodes (a) Maximum serving capacity Normalized serving capacity Nodes (b) Minimum serving capacity Normalized serving capacity Nodes Proposed AODV PAR-AODV LPR-AODV Max battery power (c) Service ratio Min queueing length WLBR AODVM QoS-AODV Max metric Figure 4: Comparison of the maximum and minimum serving capacity and service ratio. Figure 7(c). Due to the high traffic load, the delivery ratio of all protocols is low as expected. For most protocols, the delivery ratio is lower than 0.18, but for the proposed protocol the delivery ratio is about 0.22, which is relatively higher than others. Even though the proposed protocol has many advantages over other protocols, the main disadvantage of the proposed protocol is the higher overhead as shown in Figure 7(d). Figures 8 and 9 show the performance under the medium traffic loading and low traffic loading, respectively. Because the traffic rate decreases, the average queueing length of medium and lower traffic loading decrease as shown in Figures 8(a) and 9(a) and the delivery ratios in Figures 8(c) and 9(c) increase. The proposed protocol has comparable low end-to-end delay as shown in Figures 8(e) and 9(e). The total number of RREQ messages of our proposed protocol for all traffic loadings are more than others as shown in Figures 7(d), 8(d), and 9(d). In the high loading scenario, the number of RREQ messages of our proposed protocol is about 1.4 times more than other protocols. In the medium and the lowtrafficloadingscenario,thenumberofrreqmessages of our protocol is about 1.2 times more than other protocols. This is because the proposed protocol has several mechanisms of handling the link failure and the lack of resource. When the queueing length is near full or the battery power of a node is going to be exhausted soon, the proposed protocol takes action to perform the local route maintenance or reroute to relieve the loading of busy nodes. As the results, the number of overheads in the proposed protocol is more than other protocols, especially in the high traffic loading scenario.

59 10 International Journal of Distributed Sensor Networks Hops Nodes Proposed AODV PAR-AODV LPR-AODV Max battery power Min queueing length WLBR AODVM QoS-AODV Max metric Figure 5: Comparison of average hop count of a newly discovery route. (s) Nodes Proposed AODV PAR-AODV LPR-AODV Max battery power Min queueing length WLBR AODVM QoS-AODV Max metric Figure 6: Comparison of average initial delay of a newly discovery route. Another thing we should note is why queueing-aware routing protocol does not have good queueing performance in the high traffic loading scenario. The main reason behind this is that the service capacity is not taken into account. The selected route may not be able to support additional traffic. This would cause the queue accumulation. Those routing protocols taking only the battery power into consideration also have the similar problem. As we can see in Figure 7(b), the lifetime of power-aware routing protocols in high loading is a little better than AODV and queueing-aware protocols. But when the traffic loading decreases, the lifetime of poweraware protocols is worse than AODV and queueing-aware protocols. That proves that taking only the remaining battery power into consideration cannot help in prolonging the network lifetime Simulation Results of Overall Proposed Protocol: Fading Channel. We also simulate the proposed protocols and the other existing protocols under fading channel to see how the fading channel variation affects the performance of these protocols. It is clear that the performance of all the protocols under fading channel is worse than that under nonfading channel. The simulation results of queueing size under fading channelareshowninfigures10(a), 11(a),and12(a) for highrate, medium-rate, and low-rate cases, respectively. For the proposed protocol, the average queueing length is 0.3, 0.22, and 0.12 for high-rate, medium-rate, and low-rate cases, respectively. From these three figures, we can see that the proposed protocol outperforms other existing protocols in the case of fading channel. Figures 10(e), 11(e), and12(e) show the end-to-end delay in high-rate, medium-rate, and low-rate cases, respectively. In both high-rate and medium-rate cases, the proposed protocol has the smallest end-to-end delay among all the protocols. However, when being in the low-rate case, the performance of theend-to-enddelayoftheproposedprotocoliscomparable to other protocols and is slightly better than others when the mobility is high. The results of the delivery ratio are shown in Figures 10(c), 11(c), and 12(c) for the high-rate, medium-rate and low-rate cases, respectively. The proposed protocol has the highest delivery ratio in all three cases. Comparing with the situation when the channel is AWGN, we find that the delivery ratio under the fading environment is much lower than that under theawgnchannel.fromtheabovesimulationresults,itis clear that the data forwarding under the fading channel is harder than that under nonfading channel. A packet stays longerinthequeueofanodeandtakesmoretimetoreachthe destination due to the increased number of retransmission under the fading channel. For the proposed protocol, we have an adaptive retransmission mechanism to overcome the channel variation due to fading. The results show the effectivenessoftheproposedmechanism,whichexplainswhy the proposed protocol has better performance in the fading channel. The lifetimes under the fading channel are shown in Figures10(b), 11(b), and12(b) for high-rate, medium-rate, and low-rate cases, respectively. As the rate increases, the lifetime under the same setting becomes smaller. Compared with the results under AWGN channel, the performance of lifetime of the proposed protocol is degraded. This is due to the fact that more energy is spent in the retransmission, which in turn reduces the lifetime of the network. However, the proposed protocol still has the best lifetime performance among all the protocols. The fading channel increases the chance of the link failure. As a result, the overhead grows under fading channel. In

60 International Journal of Distributed Sensor Networks 11 Normalized queueing length Moving speed (m/s) (a) Queueing length (s) Moving speed (m/s) (b) Lifetime Ratio Moving speed (m/s) (c) Delivery ratio Number of RREQ packets Moving speed (m/s) (d) Number of RREQ packets (s) Moving speed (m/s) Proposed AODV PAR-AODV LPR-AODV WLBR AODVM QoS-AODV (e) End-to-end delay Figure7:High-ratecaseunderAWGNchannel.

61 12 International Journal of Distributed Sensor Networks Normalized queueing length Ratio Moving speed (m/s) (a) Queueing length Moving speed (m/s) (c) Delivery ratio Number of RREQ packets (s) Moving speed (m/s) (b) Lifetime Moving speed (m/s) (d) Number of RREQ packets (s) Moving speed (m/s) Proposed AODV PAR-AODV LPR-AODV WLBR AODVM QoS-AODV (e) End-to-end delay Figure 8: Medium-rate case under AWGN channel.

62 International Journal of Distributed Sensor Networks 13 Normalized queueing length Moving speed (m/s) (s) Moving speed (m/s) 1.1 (a) Queueing length (b) Lifetime Ratio Number of RREQ packets Moving speed (m/s) Moving speed (m/s) (c) Delivery ratio (d) Number of RREQ packets (s) Moving speed (m/s) Proposed AODV PAR-AODV LPR-AODV WLBR AODVM QoS-AODV (e) End-to-end delay Figure 9: Low-rate case under AWGN channel.

63 14 International Journal of Distributed Sensor Networks Normalized queueing length (s) Moving speed (m/s) (a) Queueing length Moving speed (m/s) (b) Lifetime Ratio Moving speed (m/s) (c) Delivery ratio (s) Number of RREQ packets Moving speed (m/s) Moving speed (m/s) (d) Number of RREQ packets Proposed AODV PAR-AODV LPR-AODV WLBR AODVM QoS-AODV (e) End-to-end delay Figure 10: High-rate case under fading channel.

64 International Journal of Distributed Sensor Networks Normalized queueing length Moving speed (m/s) (a) Queueing length (s) Moving speed (m/s) (b) Lifetime Ratio Moving speed (m/s) (c) Delivery ratio (s) Number of RREQ packets Moving speed (m/s) Moving speed (m/s) (d) Number of RREQ packets Proposed AODV PAR-AODV LPR-AODV WLBR AODVM QoS-AODV (e) End-to-end delay Figure 11: Medium-rate case under fading channel.

65 16 International Journal of Distributed Sensor Networks Normalized queueing length Ratio Moving speed (m/s) (a) Queueing length Moving speed (m/s) (c) Delivery ratio 3 (s) Number of RREQ packets Moving speed (m/s) (b) Lifetime Moving speed (m/s) (d) Number of RREQ packets 2.5 (s) Moving speed (m/s) Proposed AODV PAR-AODV LPR-AODV WLBR AODVM QoS-AODV (e) End-to-end delay Figure 12: Low-rate case under fading channel.

66 International Journal of Distributed Sensor Networks 17 Figures 10(d), 11(d),and12(d), we find that numbers of RREQ messages under the fading channel in all cases are more than those under the nonfading channel. 5. Conclusions In this work, we proposed a simple and effective routing protocol with a new route selection criterion. The adopted route selection criterion aims at finding a balance among the remaining battery power, the queueing length, and the service capacity. As we can see from the simulation results, by making use of the proposed routing protocol, the network lifetime can be further extended, the delivery ratio is improved, the average queueing length can be reduced, and the end-to-end delay is greatly improved especially under the fading channel. Conflict of Interests The authors declare that there is no conflict of interests regarding the publication of this paper. References [1] K. S. Chung and J. E. Lee, Design and development of m-learning service based on 3G cellular phones, Journal of Information Processing System,vol.8,no.3,pp ,2012. [2] H.LuoandM.L.Shyu, Qualityofserviceprovisioninmobile multimedia a survey, Human-Centric Computing and Information Sciences, vol. 1, no. 1, article 5, pp. 1 15, [3] J. K.-Y. Ng, Ubiquitous healthcare: healthcare systems and applications enabled by mobile and wireless technologies, Journal of Convergence,vol.3,no.2,pp.31 36,2012. [4] R. Sumathi and M. G. Srinivas, A survey of QoS based routing protocols for wireless sensor networks, Journal of Information Processing System,vol.8,no.4,pp ,2012. [5]A.U.Bandaranayake,V.Pandit,andD.P.Agrawal, Indoor link quality comparison of IEEE a channels in a multiradio mesh network testbed, JournalofInformationProcessing System,vol.8,no.1,pp.1 20,2012. [6] X. Li, N. Mitton, I. Stojmenovic, and A. Nayak, Achieving load awareness in position-based wireless ad hoc routing, Journal of Convergence,vol.3,no.3,pp.17 23,2012. [7] M. Maleki, K. Dantu, and M. Pedram, Power-aware source routing protocol for mobile ad hoc networks, in Proceedings of the International Symposium on Low Power Electronics and Design ( ISLPED 02), pp , August [8] M. Maleki, K. Dantu, and M. Pedram, Lifetime prediction routing in mobile ad hoc networks, in Proceedings of the IEEE Wireless Communications and Networking (WCNC 03), vol.2, pp ,NewOrleans,La,USA,March2003. [9] S.Singh,M.Woo,andC.S.Raghavendra, Power-awarerouting in mobile ad hoc networks, in Proceedings of the 4th Annual ACM/IEEE International Conference on Mobile Computing and Networking (MobiCom 98), pp , October [10] C.-K. Toh, Maximum battery life routing to support ubiquitous mobile computing in wireless ad hoc networks, IEEE Communications Magazine,vol.39,no.6,pp ,2001. [11] Y. Cui, Y. Xue, and K. Nahrstedt, A utility-based distributed maximum lifetime routing algorithm for wireless networks, IEEE Transactions on Vehicular Technology, vol.55,no.3,pp , [12] A.RaniandM.Dave, Weightedloadbalancedroutingprotocol for MANET, in Proceedings of the 16th International Conference on Networks (ICON 08), pp.1 6,NewDelhi,India,December [13] S. Puri and S. R. Devene, Congestion avoidance and load balancing in AODV-multipath using queue length, in Proceedings of the 2nd International Conference on Emerging Trends in Engineering and Technology (ICETET 09), pp , Nagpur, India, December [14] K. A. M. AlSoufy and A. M. Abbas, Lifetime and queue length constrained quality of service routing for mobile ad hoc networks, in Proceedings of the IEEE Conference and Exhibition on Control, Communications and Automation (INDICON 08),vol. 1, pp , Kanpur, India, December [15]X.Gao,X.Zhang,D.Shi,F.Zou,andW.Zhu, Contention and queue-aware routing protocol for mobile ad hoc networks, in Proceedings of the International Conference on Wireless Communications, Networking and Mobile Computing (WiCOM 07), pp , Shanghai, China, September [16] D. A. Tran and H. Raghavendra, Congestion adaptive routing in mobile ad hoc networks, IEEE Transactions on Parallel and Distributed Systems, vol. 17, no. 11, pp , [17] W. Wu, Z. Zhang, X. Sha, and D. Qin, A study of congestionaware routing protocols for wireless ad hoc networks, in Proceedings of the 4th International Conference on Wireless Communications, Networking and Mobile Computing (WiCOM 08), pp. 1 4, Alian, China, October [18] S. S. Baboo and B. Narasimhan, An energy-efficient congestion-aware routing protocol for heterogeneous mobile ad-hoc networks, in Proceedings of the International Conference on Advances in Computing, Control and Telecommunication Technologies (ACT 09), pp , Thiruvananthapuram, India, December [19] H. M. P. Shekhar and K. S. Ramanatha, Mobile agents aided congestion aware multipath routing in mobile ad hoc networks, in Proceedings of the 9th International Conference on Telecommunications (ConTEL 07), pp.65 72,Zagreb,Croatia,June [20] M. Hiyama, E. Kulla, T. Oda, M. Ikeda, and L. Barolli, Application of a MANET testbed for horizontal and vertical scenarios: performance evaluation using delay and jitter metrics, Human- Centric Computing and Information Sciences, vol.1,no.3,pp. 1 14, [21] C. E. Perkins and E. M. Royer, Ad-hoc on-demand distance vector routing, in Proceedings of the 2nd IEEE Workshop on Mobile Computing Systems and Applications (WMCSA 99),pp , New Orleans, La, USA, February [22] D. B. Johnson, D. A. Maltz, and J. Broch, DSR: the dynamic source routing protocol for multi-hop wireless ad hoc networks, in Proceedings of the ACM Ad Hoc Networking, pp , [23] E. M. Royer and C.-K. Toh, A review of current routing protocols for ad hoc mobile wireless networks, IEEE Personal Communications,vol.6,no.2,pp.46 55,1999. [24] A. Sinha and D. K. Lobiyal, Performance evaluation of data aggregation for cluster-based wireless sensor network, Human- Centric Computing and Information Sciences, vol.3,no.1,pp. 1 17, [25] A. Anastasopoulos and K. M. Chugg, An efficient method for simulation of frequency selective isotropic rayleigh fading, in Proceedings of the 47th IEEE Vehicular Technology Conference (VTC 97),vol.3,pp ,Phoenix,Ariz,USA,May1997.

67 18 International Journal of Distributed Sensor Networks [26] W.-H. Chung and K. Yao, Modified hidden semi-markov model for modelling the flat fading channel, IEEE Transactions on Communications,vol.57,no.6,pp ,2009. [27] I. Dietrich and F. Dressler, On the lifetime of wireless sensor networks, ACM Transactions on Sensor Networks, vol.5,no.1, article 5, 2009.

68 Hindawi Publishing Corporation International Journal of Distributed Sensor Networks Volume 2014, Article ID , 10 pages Research Article Self-Organized Cognitive Sensor Networks: Distributed Channel Assignment for Pervasive Sensing Li-Chuan Tseng, 1 Feng-Tsun Chien, 1 Abdelwaheb Marzouki, 2 Ronald Y. Chang, 3 Wei-Ho Chung, 3 and ChingYao Huang 1 1 Department of Electronics Engineering, National Chiao Tung University, Hsinchu 30010, Taiwan 2 Institut Mines-Télécom, Télécom SudParis, Évry, France 3 Research Center for Information Technology Innovation, Academia Sinica, Taipei 11529, Taiwan Correspondence should be addressed to Feng-Tsun Chien; ftchien@mail.nctu.edu.tw Received 21 October 2013; Accepted 9 February 2014; Published 17 March 2014 Academic Editor: Luis Javier Garcia Villalba Copyright 2014 Li-Chuan Tseng et al. This is an open access article distributed under the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited. We study the channel assignment strategy in multichannel wireless sensor networks (WSNs) where macrocells and sensor nodes are overlaid. The WSNs dynamically access the licensed spectrum owned by the macrocells to provide pervasive sensing services. We formulate the channel assignment problem as a potential game which has at least one pure strategy Nash equilibrium (NE). To achieve the NE, we propose a stochastic learning-based algorithm which does not require the information of other players actions and the time-varying channel. Cluster heads as players in the game act as self-organized learning automata and adjust assignment strategies based on their own action-reward history. The convergence property of the proposed algorithm toward pure strategy NE points is shown theoretically and verified numerically. Simulation results demonstrate that the learning algorithm yields a 26% sensor node capacity improvement as compared to the random selection, and incurs less than 10% capacity loss compared to the exhaustive search. 1. Introduction In wireless sensor networks [1], spatially distributed, lowpower, and low-cost sensor nodes are deployed in a geographical area to monitor the environment. The sensor nodes usually form clusters, and in each cluster there is an energyrich sensor node acting as the cluster head, while other sensor nodes are referred to as cluster members. A cluster head is a special sensor node with better cognitive radio (CR) [2, 3] functionality and is responsible for the spectrum sensing and the channel assignment among its cluster members. To enable the various kinds of services [4 6] provided by a pervasive sensing system, proper radio resource management [7] is important. Due to the spectrum scarcity and the ad hoc nature of sensor network deployment, it couldbehardtoassignlicensedbandstosensornetworks. Therefore, the CR technology has been considered as a promising solution to the channel assignment problem of sensor networks. CR technology enables dynamic spectrum access (DSA) of unlicensed users in distributed networks. The key idea for CR operation is to allow the active sensing of the dynamic radio environment so as to improve the spectrum utilization. Akan et al. [8] providedasurveyoncognitive radio sensor networks. By utilizing the CR technology, the sensor networks are able to attain high data rate due to available spectrum holes. In addition, dynamic spectrum access helps mitigate the interference incurred by dense deployment of sensor nodes. Despite the promising features of cognitive sensor networks, the deployment of such heterogeneous networks with sensor clusters underlying the same spectrum as macrocells and in the same geographical area brings new technical challenges. In particular, we are interested in the case of densely populated sensor networks where, due to extensive frequency

69 2 International Journal of Distributed Sensor Networks reuse, the cochannel interference (CCI) among sensor nodes and the cross-tier interference (between the macrocell and sensor networks) affect the system performance. In the absence of a central controller, channel assignment in cognitive sensor networks is implemented in a distributed manner. In this paper, we consider the distributed channel assignment for self-organized cognitive sensor networks from a game-theoretic perspective. The main contributions of this paper are summarized as follows. (i) We model the femtocell channel assignment problem as an ordinal potential game (OPG). The game considers time-varying channel availability as its external state. (ii) We propose a fully decentralized channel assignment algorithm in which the channel is selected by each cluster head independently based on its action-reward history. The convergence property of the algorithm to a pure strategy NE point is verified theoreticallyaswellasnumerically. Thispaperisorganizedasfollows.Wereviewtherelated work and compare it with our study in Section 2.InSection 3, the system model for a cognitive sensor network is presented. Section 4 describes the game-theoretic model of the distributed channel assignment problem. Section 5 presents the stochastic learning procedure carried out by the cluster heads. Finally, numerical results are given in Section 6, with the conclusion drawn in Section 7. Notations. Normal letters represent scalar quantities; uppercase and lowercase boldface letters denote matrices and vectors, respectively. Given a finite set A, Δ(A) represents the set of all probability distributions over the elements of A. 1 {cond} is the indicator function which equals one if the condition cond is satisfied and zero otherwise. 2. Related Works Distributed channel assignment has been extensively investigated for different networking applications where concurrent transmissions among neighboring wireless links exist. In an interference avoidance scenario, different channels must be assigned to neighboring links. In femtocell networks [9], different methods have been proposed to assign different spectrum to adjacent femtocells. Examples include distributed random access [9], dynamic frequency planning [10], and clustering [11]. For sensor networks with multiple channels, graph theory-based methods have also been considered [12]. These methods can be viewed as variations of frequency planning and usually rely on negotiations among neighboring links. Graph theory is also a popular approach as the interference condition can be represented as nodes and edges in a graph. In sensor networks, Chowdhury et al. [13] proposed the dynamic channel allocation (DCA) and studied the related protocol design. Yu et al. [14] considered a game theory-based approach which takes into account both the network topology and transmission routing information. MBS R BS x a=10m a a x Active MBS MUE and its avoiding region Figure 1: Dual-stripe deployment of sensor clusters. Channel selection for multicell orthogonal frequency division multiple access- (OFDMA-) based networks using graph framework was considered in [15]. On the other hand, in an interference mitigation scenario, mutual interference is tolerated. Channel assignment for cognitive sensor networks has been studied in [16]. Recently, self-organization of distributed agents based upon reinforcement learning (RL) mechanisms [17, 18] has been showntobeeffectiveintheliterature.multiagentq-learning (MAQL) was applied to femtocell networks in [19, 20]. MAQL involves the actions of other agents as the external state and thus requires the sharing of the knowledge of all agents actions. The stochastic learning (SL), in contrast, updates the actions of users based on their individual action-reward history. Nie and Comaniciu [21] considered the channel selection in cognitive radio using interference mitigation game formulation. SL was also applied to the opportunistic spectrum access in cognitive radio networks [22] to achieve the Nash equilibrium (NE) strategy. However, fully distributed resource allocation in cognitive sensor networks has not been extensively investigated. 3. System Model We consider a cognitive radio sensor network consisting of one MBS and N sensor clusters under the coverage of the MBS. The method of sensor node clustering and cluster head selection [23] are also interesting topics but are out of the scope of this paper. Also we consider only the single-hop transmission and omit the multihop routing issue for ad hoc networks [24].Thesensorsaredeployedinanapartment block with a dual-stripe room layout, as shown in Figure 1. In our considered system, the medium access control (MAC) function in a cluster resembles that of cellular systems. The time domain is divided into frames, and a frame is further divided into time slots. In each frame, a cluster head allocates its cluster members (i.e., sensor nodes) in different time slots following a time division multiple access (TDMA) rule. For simplicity, we assume that in each slot each cluster head allocates one sensor node over one of the available channels. We emphasize that the proposed method can be easily generalized to the cases with multiple sensor nodes per slot. A sensor node is in idle mode unless the current time x x a a

70 International Journal of Distributed Sensor Networks 3 A1 A2 A3 A4 Cluster A B1 B2 B3 B4 B5 Cluster B without ambiguity. Then, the signal-to-interference-andnoise ratio (SINR) at cluster head i canbeexpressedas γ i = P i h i,i 2 N j=1,j =i I j i +σ. (2) 2 C1 C2 C3 C4 Cluster C Consequently, the expected capacity for link i in bits/s/hz is given by R i =θ i,ai log 2 (1+γ i ). (3) ID Sensor node with given ID transmittals Time slots An idle slot Let a =(a 1,...,a N ) be the channel assignment profile of all active clusters. The global objective of the system is to find the optimal channel selection profile a opt that maximizes the sum capacity. Formally, consider Figure 2: Exemplary time slot allocation in a frame. In the first slot, cluster heads A and C assign channels for sensor nodes A1 and C1, respectively. slot is allocated for it. A cluster head is idle if none of its sensor nodes is transmitting data and active otherwise. We therefore introduce an active ratio, which is defined as the percentage of active clusters in a time slot. An exemplary time slot allocation is depicted in Figure 2. The spectrum is divided into C channels, and the channels may be licensed to different macrocells (a.k.a spectrum owners). By utilizing CR, the sensor nodes access the same frequency band as the macrocell does. Since the sensor nodes are in an energy-tight situation and operate with ultralow power, we assume that the transmission power of a macrocell user equipment (MUE) is much higher than that of the sensors. Thus, the uplink transmission of an MUE will block the nearby sensor nodes using the same spectrum. For crosstier interference mitigation, we define an avoiding region for each MUE. A channel is available to a cluster only if the channel is not assigned to an MUE whose avoiding region covers the cluster head. The channel availability for sensor clusters is expressed as a binary matrix X {0, 1} N C,in which the element x i,c equals one if channel c is available to link i and zero otherwise. The elements of X follow the Bernoulli distribution and can be described by a probability matrix Θ [0, 1] N C,wheretheelementθ i,c is the probability that x i,c =1. Assuming perfect synchronization in time and frequency, let P i denote the power of sensor node i,andlet h i,j 2 indicate the link gain between cluster head i and sensor node j. The interference received by cluster head i from sensor node j is given by I j i = 1 {ai (n)=a j (n)}p j h i,j 2, i,j N, (1) where a i (n) is the action (channel selection) of cluster head i in frame n. For notational brevity, we will hereafter discard the timing dependence of the action a i (n) in occasions a opt = argmax a N i=1 θ i,ai log 2 (1+γ i ). (4) To reflect a practical cognitive sensor network, our system model incorporates the following considerations. (1) The uplink resource allocation for MUE is timevarying during the learning period, and the channel availability statistics (i.e., Θ) arefixedbutunknown to any sensor nodes. (2) There is no centralized controller and the channel selection is performed independently by each cluster head. (3)Thenumberofsensorclustersinthesystem,N, is unknown. With these considerations, solving (4) is a challenging task, since the only available information for decision making at each individual player is its own action-reward history. Thus,afullydistributedchannelselectionschemeisproposed. 4. Game-Theoretic Model and Channel Selection In this section, we present the game-theoretic formulation of the self-organized cognitive sensor network channel selection problem. Our objective is to devise for each cluster head a distributed channel assignment strategy that takes into account the effect of both the sensor-tier and cross-tier interference. We summarize our notations related to the game formulation in Table Problem Formulation and Game Model. The channel selection problem described in the previous section can be modeled by a normal-form game with external state, expressed as a 4-tuple: G =(X, N,{A i } i N,{u i } i N ), (5) where X is the external state (channel availability) space, N is the set of players (cluster heads), A i is the set of actions

71 4 International Journal of Distributed Sensor Networks Table 1: Summary of notations in game-theoretic formulation. Symbol X X N A i s i A i a i (n) A i a i (n) A i P i := Δ(A i ) p i (n) P i r i (n) R û i (n) R A i (ε i,λ i ) Meaning External state (channel availability) A realization of external state (channel availability) Set of players Set of actions of player i An element of A i Action (channel selection) of player i at slot n Actions of players except for i at slot n Set of probability distribution over A i Mixed strategy of player i at slot n Observed utility of player i at slot n Estimated utility vector of player i at slot n Learning rates of player i (selections of channels) that player i can take, and {u i } i N is the utility function of player i that depends on his own action as well as on the actions of other players. Inspired by [21], the reward function is designed to consider the interference received (inward) and generated (outward) by each link. In this way, the cluster heads implicitly cooperate to reduce the interference generated toward other sensor nodes. We define the generalized SINR (gsinr) for player i as γ i = P i h i,i 2 N j=1,j =i (I j i +I i j )+σ. (6) 2 Then the instantaneous reward function of cluster head i is designed as r i ={ log 2 (1 + γ i), if x i,ai =1; 0, otherwise. By the definition in (7), when the channel is available, the reward is given by Shannon s capacity formula where both inward and outward interference are accounted for. When the channel is not available, the reward is zero. Notice that thecalculationoftherewardfunctionin(7) reliesonthe knowledge of other players action. This leads to overhead due to the required information. The implementation is possible, anddiscussiononsuchprotocoldesigncanbefoundin[21]. The self-organization claimedinthispaperisbasedonthefact that the action in each time instant is selected by each player independently and simultaneously. For systems with the channel availability as the external state, the utility function is defined as the expected reward of player i over the external state (i.e., channel availability X); that is, (7) u i (a i,a i ) =θ i,ai log 2 (1+ γ i ). (8) Furthermore, if the cluster heads are assumed to be selfish and rational players, they will compete to maximize their own individual utility. In fact, a selfish cluster head will not only maximize the capacity of its own user but also reduce the interference. Formally, the game G is expressed as (G) :max a i A i u i (a i,a i ), i N. (9) 4.2. Analysis of Nash Equilibrium. With the utility function defined in (8), we show the existence of an NE point for the proposed game in the following proposition. Proposition 1. The game G is an ordinal potential game (OPG) which possesses at least one pure strategy NE. Proof. Consider the function Φ: i N A i R + : Φ (a) = log 2 (1 + N k=1 P k h k,k 2 N k=1 N j=1,j =k I k j ). (10) Now consider an improvement step made by cluster head i that changes its action unilaterally from a i to a i,sothat u i ( a i,a i )>u i (a i,a i ).DefiningIi j 1 { a i =a j }P i h j,i 2 and I j i 1 { a i =a j }P j h i,j 2,wehave u i ( a i,a i )>u i (a i,a i ) N [Ii j +I j i N j=1,j =i j=1,j =i [I i j +I j i ] N [Ii j +I j i N N I j k j=1,j =i j=1,j =i k=1,k =i,j < N j=1,j =i [I i j +I j i ]+ N N j=1,j =i k=1,k =i,j I j k. (11) Here we have used the fact that when cluster head i changes its action, the effects are only on the interference that it receives (I j i ) and generates (I i j ). From (10)and(11), we obtain u i ( a i,a i ) u i (a i,a i )>0 Φ( a i,a i ) Φ(a i,a i )>0. (12) Therefore, G is an OPG with potential function Φ,andthe existence of a pure strategy NE is always guaranteed [26]since it coincides with the local maxima of the potential function. This completes the proof. Notice that the term N k=1 N j=1,j =k I k j in the potential functionφ denotes the summation of all mutual interference in the sensor network. Therefore, every NE point is the strategyprofile,thatis,alocalmaximumofthesummed interference. 5. Stochastic Learning Procedure Here, we discuss obtaining the NE via stochastic learning. As the channel state is time-varying and the action is selected

72 International Journal of Distributed Sensor Networks 5 (1)Initially, set n=0. Set the channel assignment probability vector and utility estimation as p i,si (0) = 1/ A i, u i,s i ( 1) = 0, i N, s i A i. (2) At the beginning of the nth slot, each player selects an action a i (n) according to the current channel assignment probability p i (n). (3) In each slot, each BS transmits data. At the end of each slot, each BS receives the instantaneous reward r i (n) specified by (15) depending on the precoding scheme. (4) All players update their channel assignment probability vector and utility estimation according to the rules: u i,si (n) u i,si (n 1) =η i 1 {ai (n)=s i } (r i (n) u i,si (n 1)), p i,si (n)(1 + ε ) u i,s i (n) i p i,si (n + 1) =, ( ) s i A i p i,s (n)(1 + ε ) u i,s i (n) i i where ε i and η i are the learning rates for action probability and utility estimation, respectively. Algorithm 1: Distributed channel assignment (DCA). by each player simultaneously and independently in each play, previous algorithms that require complete information (e.g., better response dynamics [26]) may not be applicable here. Thus, we propose a decentralized stochastic learning- (SL-)basedalgorithmbywhichtheBSslearntowardthe equilibrium strategy profile from their individual actionreward history. To facilitate the development of the SL-based channel selection algorithm, we extend the channel selection game into a mixed strategy form. Let p i (n) = [p i,1 (n),...,p i,c (n)] T, for all i N, be the channel selection probability vector for player i, wherep i,si (n) is the probability that player i selects strategy s i A i at slot n. More precisely, using mixed strategies means that the channel assignment of cluster head i is the outcome of a probabilistic experiment based on the probability vector p i (imagine that each SU rolls a biased dice in each strategy update). The mixed-strategy extension of the utility function is defined upon i N P i,wherep i is the set of probability distributions over the action space of player i. Let P(n) = [p 1 (n),..., p N (n)] be the mixed strategy profile of G,andletψ i (s i, P) be the expected reward function of player i if he employs pure strategy s i while other players j, forall j N,j=i,employamixedstrategyp j ;thatis, ψ i (s i, P) = u i (a 1,...,s i,...,a N ) p j,aj. a l,l =i j =i (13) The proposed distributed channel assignment (DCA) algorithm for cognitive sensor networks is described in Algorithm 1. In each play, the channel selection is based on a probability distribution over the set of channels. After each play, cluster head i obtains the instantaneous reward and updates the mixed strategy (i.e., channel selection vector) p i (n) and utility estimation û i (n). Notably, the utility estimation serves as a reinforcement signal so that higher utility induces higher probability in the next play. Furthermore, the proposed learning algorithm is fully distributed, and the channel selection is solely based on individual action-reward experience without a centralized controller. In fact, the proposed algorithm belongs to the combined fully distributed payoff strategy reinforcement learning (CODIPAS-RL) [27]. The evolution of the mixed strategies is described as follows. Proposition 2. If the learning rates are sufficiently small, the sequence {P(n)} converges to P, which is the solution for the following ordinary differential equation (ODE): dp i,si (t) =p dt i,si (t) [ ψ i (s i, P) ψ i (s i, P)p i,s (t)]. [ si A i i ] (14) Proof. Please refer to [28,Section4]. The ODE in (14) is actually the ODE of the replicator dynamics [29]. An intuitive interpretation is that the probability of taking an action increases if the utility is higher than the average utility over all possible actions and decreases otherwise. The convergence property of the proposed algorithm is discussed in the following proposition. Proposition 3. The SOCA algorithm converges to a pure strategy NE for OPGs if the learning rates are sufficiently small. Proof. First,werewritetheODEin(14) as follows: dp i,si (t) dt =p i,si (t) p i,s i (t) [ψ i (e si, P i ) ψ i (e s i, P i )]. s i A i Let Ψ(P) be the expected potential function; that is, Ψ (P) = Φ(a 1,...,a N ) a i A i N i=1 (15) p i,ai. (16) For OPGs, Ψ(e si, P i ) = Ψ(P)/ p i,si is an increasing function of ψ i (e si, P i ).Let D i,si,s i E i,si,s i =ψ i (e si, P i ) ψ i (e s i, P i ), =Ψ(e si, P i ) Ψ(e s i, P i ), (17)

73 6 International Journal of Distributed Sensor Networks 1 ε = ε = 0.5 Probability 0.5 Probability p i,1 (t) p i,2 (t) Iteration (n) p i,1 (t) p i,2 (t) Iteration (n) Figure 3: Evolution of the mixed strategies (probability of taking different actions) of all players. Each pair of p i,1 (t) and p i,2 (t) shows the behavior of player i. and we may write D i,si,s i >0 E i,si,si >0. (18) By applying (15)and(18), the derivation of Ψ(P) with respect to t is given by dψ (P) dt Ψ (P) = p s i A i,si i i N = i N s i,si A i = 1 2 0, i N s i,s i A i s i <s i dp i,si dt p i,si p i,s i Ψ(e si, P i ) D i,si,s i p i,si p i,s i E i,si,s i D i,si,s i (19) where the last inequality holds since, given the condition in (18), D i,si,si and E i,si,si alwayshavethesamesign. Thus, Ψ is nondecreasing along the trajectories of the ODE, and asymptotically all the trajectories will be in the set {P P :dψ(p)/dt = 0}.From(15)and(19), the following is known: dψ (P) dt =0 p i,si p i,s i [ψ i (e si, P i ) ψ i (e s i, P i )] 2 =0 dp i,s i dt =0, i,s i,s i P is a stationary point of the ODE (14). (20) In other words, when starting from an interior point of the simplex of the mixed strategy space P, thesequence P(n) converges to a stationary point of the ODE in (15). By Proposition 3,we complete the proof. While the SL-based learning algorithm converges to an NE point when the learning rates approach zero, smaller learning rates lead to a slower convergence. Therefore, the Table 2: Simulation parameters. Parameter Value Minimum distance between nodes 3 m Carrier frequency 2 GHz Number of channels 2 Transmission bandwidth of each channel 180 khz Path loss and shadowing Table A [25] Penetration loss Table A [25] Sensor transmission power 1 mw Thermal noise 174 dbm/hz Learning rates (default) (λ i,ε i ) = (0.1, 0.1) choice of the learning rates strikes a trade-off between accuracy and speed and may be determined by training in practice. 6. Numerical Results For system-level simulations, we consider a cognitive sensor network deployed within the coverage of a cellular network. As in Figure1, the simulation environment includes one macrocell covering one dual-stripe apartment block. The apartment block contains 40 single-floor apartments. There is one sensor cluster in each apartment. When a sensor cluster is active, its cluster head assigns one channel to cluster members randomly located in the same apartment. Without loss of generality, we consider the channel assignment in the first slot of each frame, in which for each active cluster there is one cluster member. The simulation parameters are listed in Table Convergence of the Proposed SL-Based Learning Algorithm. We first study the time-evolving behaviors of the proposed stochastic learning method Evolution of Mixed Strategies. Figure 3 shows the evolutionsofthechannelassignmentprobabilities(i.e.,mixed strategy) using the proposed SL-based algorithm. We consider different learning rates and study the convergence behaviors. It is observed that, with equal initial probability,

74 International Journal of Distributed Sensor Networks Utility 4 2 Utility Link ID Link ID NE Deviation NE Deviation (a) ε=0.1 (b) ε=0.5 Figure 4: Test of unilateral deviation from the resulting strategy profile of each of the 10 players. Action Action Action Action a 3 (j) Iteration (n) a 5 (j) Iteration (n) Iteration (n) a 6 (j) a 10 (j) Iteration (n) Figure 5: Evolution of the actions a i (j) for some players. thechannelassignmentprobabilityconvergestoapurestrategy (i.e., the probability of choosing one strategy approaches one) in around 80 and 20 iterations for ε = 0.1 and ε = 0.5, respectively. As expected, larger learning rate results in faster convergence. Figure 4, we verify the NE property by testing the unilateral deviation with a 25% active ratio and different learning rates. AscanbeseenfromFigure 4(a), whenε = 0.1, aunilateral deviation results in lower utility for all players. In other words, the outcome of the learning algorithm is an NE point. On the other hand, when ε = 0.5, asshowninfigure 4(b),links number 4 and number 8 both achieve higher throughput by unilateral deviation, and thus the resulting strategy is no longer an NE point. These results reflect the trade-off between accuracy and convergence speed which we mentioned before Evolution of Actions. During the learning procedure, the channel assignment is based on probabilistic experiments. When the channel assignment changes in the next frame, the switching between different channels brings overhead since the sensor node needs to be reconfigured. The evolution of actions for selected players is shown in Figure 5. Ascan be seen, while Figure 3 (Left) reveals that it takes around 80 iterations for all players to converge to pure strategies, the actions seldom change after about 60 iterations in the learning procedure. This suggests that channel switching, if at all happens, usually happens only in the beginning of the entire learning procedure. Actually, our proposed learning algorithm aims at learning the equilibrium strategy in the long run. The channel switching and the incurred sensor node reconfiguration are manageable overheads compared to the long operation time Different Active Ratios. We further consider different active ratios and investigate the convergence behaviors under different levels of mutual interference. The results for active ratio of 50% and 75% are shown in Figure 6.Weobservethat the convergence toward pure strategy takes around 100 and 150 iterations for active ratio of 50% and 75%, respectively. Comparing the case of 25% active ratio in Figure 3 (Left), we see that it takes fewer iterations for densely active networks to converge than for sparsely active sensor networks Verification of NE. As shown in Figure 3, the convergencetowardpurestrategyisobservedforbothε = 0.1 and ε = 0.5. An intuitive question to ask is as follows: does the resulting strategy profile achieve the Nash equilibrium? In 6.2. Capacity Performance Capacity under Unilateral Deviation. In Figure 4 we have shown that unilateral deviation leads to decreased utility.

75 8 International Journal of Distributed Sensor Networks Probability Active ratio = 50% Iteration (n) Probability Active ratio = 75% Iteration (n) p i,1 (t) p i,2 (t) p i,1 (t) p i,2 (t) Figure 6: Evolution of the mixed strategies (probability of taking different actions) of all players with active ratios of 50% and 75%. Each pair of p i,1 (t) and p i,2 (t) shows the behavior of player i. 8 8 Capacity Capacity Link ID Link ID NE Deviation NE Deviation (a) Link capacity (b) Average capacity Figure 7: Test of unilateral deviation from the resulting strategy profile of each of the 10 players. While the altruistic utility function design reduces the mutual interference, we are also interested in the performance of Nash equilibrium strategy in terms of the throughput of each clusteraswellasthewholesystem.therefore,infigure 7,we test the change in capacity under unilateral deviation from the NE strategy for all players. As depicted in Figure 7(a), there is no significant change in the average capacity per sensor link when only one player unilaterally deviates from its NE strategy. From Figure 7(b) we observe that, for all players, deviation from NE strategy decreases their own capacity Comparison with Other Methods. We further compare the performance of the proposed channel selection scheme with two other approaches, namely, random allocation and exhaustive search, described as follows. (i) In the random allocation scheme, each cluster head randomly selects a channel for its sensor node in each frame. Neither learning algorithm nor centralized controller is implemented. (ii) In the exhaustive search scheme, it is assumed that there exists a centralized controller which knows all system information including the channel gains, the channel availability statistics, and the number of clusters. The channel assignment profile is determined by maximizing the expected sum capacity (i.e., solving (4)). Table 3: Comparison of the capacity and fairness for different channel assignment schemes. Number of SUs Proposed Exhaustive Random Active ratio = 25%, R avg Active ratio = 25%, J Active ratio = 50%, R avg Active ratio = 50%, J The performance of different channel selection schemes is evaluated by the average capacity per sensor node, R avg = (1/N) N i=1 R i, and the fairness among sensor nodes. In the literature, fairness of resource allocation is usually quantified by Jain s fairness index (JFI) [30], which is defined as J= ( N i=1 R i) 2 N N. (21) i=1 R2 i The value of JFI falls in the interval of [1/N, 1],andahigher JFI value indicates better fairness. The simulation results of average capacity and JFI for differentactiveratiosaresummarizedintable 3.Weobserve that the exhaustive search method results in the best average capacity with the worst fairness. The random selection scheme, in contrast, has the lowest average capacity but good fairness due to its randomness nature. The proposed

76 International Journal of Distributed Sensor Networks 9 method shows well-balanced performance in terms of both averagecapacityandfairness.theresultsshowtheadvantages of the proposed method; through the learning procedure toward equilibrium, the capacity of each player is considered and fewer players are sacrificed. If we examine the final channel selection profile, it is observed that, in the progress of convergence toward the NE point, the proposed learning algorithm allocates the mutually interfered users on different channels. 7. Conclusion In this paper, we have studied the problem of distributed channel assignment in self-organized cognitive sensor networks with unknown channel and unknown number of clusters. We have presented a game-theoretic approach to distributively manage interference and enable the coexistence of sensor and macrocell operations in a scenario where sensor nodes operate in the same spectrum as a cellular system. We modeled channel assignment problem by means of an ordinal potential game. A decentralized stochastic learning algorithm has been proposed. Simulation results have demonstrated the convergence of the algorithm toward a pure strategy Nash equilibrium with sufficiently small learning rates. The proposed method outperforms the random selection scheme in terms of average capacity, while the performance loss compared to the exhaustive search is limited. In addition, its fairness level is comparable to that of the random selection and surpasses the exhaustive search scheme. Conflict of Interests The authors declare that there is no conflict of interests regarding the publication of this paper. Acknowledgment This work was supported in part by the National Science Council, Taiwan, under Grant NSC E References [1] I. F. Akyildiz, W. Su, Y. Sankarasubramaniam, and E. Cayirci, A survey on sensor networks, IEEE Communications Magazine, vol. 40, no. 8, pp , [2] B.Wang,Y.Wu,andK.J.R.Liu, Gametheoryforcognitive radio networks: an overview, Computer Networks, vol.54,no. 14, pp , [3]C.Liu,O.Granados,R.Duarte,J.Andrian,andC.Author, Energy efficient architecture using hardware acceleration for software defined radio components, JournalofInformation Processing Systems,vol.8,no.1,pp ,2012. [4] J. Ng, Ubiquitous healthcare: healthcare systems and application enabled by mobile and wireless technologies, Journal of Convergence,vol.3,no.2,pp.31 36,2012. [5] Y. Zhu and Q. Jin, An adaptively emerging mechanism for context-aware service selections regulated by feedback distributions, Human-Centric Computing and Information Sciences, vol.2,no.1,pp.1 15,2012. [6] S.Silas,K.Ezra,andE.BlessingRajsingh, Anovelfaulttolerant service selection framework for pervasive computing, Human- Centric Computing and Information Sciences, vol.2,no.1,pp. 1 14, [7] G. H. Carvalho, A. Anpalagan, I. Woungang, and S. K. Dhurandher, Energy-efficient radio resource management scheme for heterogeneous wireless networks: a queueing theory perspective, Energy, vol. 3, no. 4,2012. [8] O.B.Akan,O.B.Karli,andO.Ergul, Cognitiveradiosensor networks, IEEE Network,vol.23, no.4,pp , [9] K. Sundaresan and S. Rangarajan, Efficient resource management in OFDMA femto cells, in Proceedings of the 10th ACM International Symposium on Mobile Ad Hoc Networking and Computing (MobiHoc 09), pp , New York, NY, USA, May [10] D. López-Pérez, A. Valcarce, G. de la Roche, and J. Zhang, OFDMA femtocells: a roadmap on interference avoidance, IEEE Communications Magazine,vol.47,no.9,pp.41 48,2009. [11] A. Hatoum, N. Aitsaadi, R. Langar, R. Boutaba, and G. Pujolle, FCRA: femtocell cluster-based resource allocation scheme for OFDMA networks, in Proceedings of the IEEE International Conference on Communications (ICC 11), pp. 1 6, Kyoto, Japan, June [12] C. Seneviratne and H. Leung, A game theoretic approach for resource allocation in cognitive wireless sensor networks, in Proceedings of the IEEE International Conference on Systems, Man, and Cybernetics (SMC 11), pp ,Anchorage, Alaska, USA, October [13] K.R.Chowdhury,N.Nandiraju,P.Chanda,D.P.Agrawal,and Q.-A. Zeng, Channel allocation and medium access control for wireless sensor networks, Ad Hoc Networks, vol. 7, no. 2, pp , [14] Q. Yu, J. Chen, Y. Fan, X. Shen, and Y. Sun, Multi-channel assignment in wireless sensor networks: a game theoretic approach, in Proceedings of the IEEE INFOCOM, pp. 1 9, San Diego, Calif, USA, March [15]R.Y.Chang,Z.Tao,J.Zhang,andC.-C.J.Kuo, Multicell OFDMA downlink resource allocation using a graphic framework, IEEE Transactions on Vehicular Technology, vol. 58, no. 7, pp , [16] J. Ansariand P. Mähönen, Channel selection in spectrum agile and cognitive MAC protocols for wireless sensor networks, in Proceedings of the 8th ACM International Workshop on Mobility Management and Wireless Access (MobiWac 10), pp.83 90, Bodrum, Turkey, October [17] Y. Sung and K. Cho, A method for learning macro-actions for virtual characters using programming by demonstration and reinforcement learning, JournalofInformationProcessing Systems,vol.8,no.3,pp ,2012. [18] B. J. Oommen, A. Yazidi, and O.-C. Granmo, An adaptive approach to learning the preferences of users in a social networkusing weak estimators, Journal of Information Processing Systems,vol.8,no.2,pp ,2012. [19] M. Bennis and D. Niyato, A Q-learning based approach to interference avoidance in self-organized femtocell networks, in Proceedings of the IEEE GLOBECOM Workshops, pp , Miami, Fla, USA, December [20] A. Galindo-Serrano and L. Giupponi, Femtocell systems with self organization capabilities, in Proceedings of the 5th International Conference on Network Games, Control and Optimization (NetGCooP 11), pp. 1 7, Paris, France, October 2011.

77 10 International Journal of Distributed Sensor Networks [21] N. Nie and C. Comaniciu, Adaptive channel allocation spectrum etiquette for cognitive radio networks, in Proceedings of the 1st IEEE International Symposium on New Frontiers in Dynamic Spectrum Access Networks (DySPAN 05), pp , Baltimore, Md, USA, November [22] Y.Xu,J.Wang,Q.Wu,A.Anpalagan,andY.-D.Yao, Opportunistic spectrum access in unknown dynamic environment: a game-theoretic stochastic learning solution, IEEE Transactions on Wireless Communications, vol. 11, no. 4, pp , [23] B. Singh and D. Lobiyal, A novel energy-aware cluster head selection based on particle swarm optimization for wireless sensor networks, Human-Centric Computing and Information Sciences,vol.2,no.1,pp.1 18,2012. [24] X. Li, N. Mitton, A. Nayak, and I. Stojmenovic, Achieving load awareness in position-based wireless ad hoc routing, Journal of Convergence,vol.3,no.3,2012. [25] 3GPP, E-UTRA: further advancements for E-UTRA physical layer aspects, Tech. Rep. TR , v 9.0.0, 3GPP, Sophia Antipolis, France, [26] D. Monderer and L. S. Shapley, Potential games, Games and Economic Behavior,vol.14,no.1,pp ,1996. [27] M. A. Khan, H. Tembine, and A. V. Vasilakos, Game dynamics and cost of learning in heterogeneous 4G networks, IEEE Journal on Selected Areas in Communications,vol.30,no.1,pp , [28] H. Tembine, Distributed Strategic Learning for Wireless Engineers, CRC Press, New York, NY, USA, [29] D. Fudenberg and D. Levine, The Theory of Learning in Games, vol. 2, The MIT press, Cambridge, Mass, USA, [30] R. Jain, D. Chiu, and W. Hawe, A quantitative measure of fairness and discrimination for resource allocation in shared computer systems, Tech. Rep. TR-301, DEC, Maynard, Mass, USA, 1984.

78 Hindawi Publishing Corporation International Journal of Distributed Sensor Networks Volume 2014, Article ID , 12 pages Research Article Broadcast Aggregation to Improve Quality of Service in Wireless Sensor Networks Evy Troubleyn, Jeroen Hoebeke, Ingrid Moerman, and Piet Demeester Department of Information Technology (INTEC), Ghent University-iMinds, Gaston Crommenlaan 8, Bus 201, 9050 Ghent, Belgium Correspondence should be addressed to Evy Troubleyn; Received 28 September 2013; Accepted 28 January 2014; Published 9 March 2014 Academic Editor: Luis Javier Garcia Villalba Copyright 2014 Evy Troubleyn et al. This is an open access article distributed under the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited. In-network aggregation is used in wireless sensor networks to reduce energy consumption on sensor nodes with limited capabilities. Typically, only packets with the same destination are aggregated along the routing path and these packets are sent by unicast. While this is adequate for traditional multipoint-to-point sensor network applications, this is not suitable when wireless sensor nodes are connected in a full mesh topology, as is the case in the Internet of Things. In these full mesh topologies, queues will be filled with packets with many different destinations, which limits the aggregation possibilities. Furthermore, the Quality of Service level will decrease since packets have to wait longer to find aggregation candidates. Therefore, in this paper, we propose to use broadcast aggregation that can aggregate packets independent of their destination. Broadcast aggregation is analyzed through simulations against unicast aggregation and no aggregation. Results show that energy can be reduced up to 13% compared with unicast aggregation and up to 27% compared with no aggregation. In terms of reliability, the Quality of Service is improved up to 15% compared with unicast aggregation and up to 23% compared with no aggregation. The delay on its turn is decreased by 52% compared with unicast aggregation. 1. Introduction Reducing energy consumption in wireless sensor networks has been extensively studied in recent years. Wireless sensor networks are networks that contain low-power and low-cost sensor nodes that are communicating through a wireless radio interface. These sensor nodes are often battery powered, and hence energy is a scarce resource [1]. A technique that is often used to reduce energy consumption is in-network aggregation [2]. In traditional in-network aggregation, many packets with the same (next-hop) destination are aggregated into one big packet and the aggregated packet is sent along the routing path by unicast messages. As a consequence, fewer transmissions and receptions are needed, and since radio communication consumes most of the sensor network energy, energy is saved by reducing the radio-on time. However, the impact of in-network aggregation on the Quality of Service (QoS) level is currently not fully addressed and the trade-off between energy and QoS is often ignored. In previous work [3], we have already studied the impact of in-network aggregation on energy, delay, and reliability and we presented a tunable QoS-aware in-network aggregation scheme that makes a trade-off between energy consumption andapredefinedqoslevel.butwehaveobservedthatthis technique works only well when there are many aggregation opportunities, which is true for traditional source-to-sink applications. Today, more and more sensor networks are interconnected with the Internet, forming what is called the Internet of Things (IoT). For example, Libelium [4] lists 57 sensor applications for a smarter world situated in home, environments, industry, healthcare, and many more. These sensor nodes will join the Internet dynamically and use it to collaborate and accomplish their task [5]. A drawback is that there are many different communication paths which can lead to fewer aggregation opportunities and fewer packets that can be aggregated into one single packet (in the following referred to as the degree of aggregation, DoA) since the queues on the nodes become filled with packets with many different destinations. As a consequence, more energy is wasted and the QoS level is decreased since packets will have to wait longer resulting in a higher drop probability.

79 2 International Journal of Distributed Sensor Networks Therefore, in this paper, we propose to use broadcast aggregation to aggregate packets independent of their nexthop destination. We show that applying broadcast aggregation leads to faster aggregation with a higher overall QoS level and lower energy consumption. The remainder of this paper is organized as follows. Section 2 gives an overview of the related work on QoSaware unicast and broadcast in-network aggregation, while in Section 3 the applied broadcast mechanism is explained. A definition and problem statement is given first, followed by a performance analysis. Simulation results are discussed in Section 4.Weendthispaperwithaconclusionandalookon future work in Section Related Work Because energy is a main concern in wireless sensor networks and in-network aggregation is very suited to reduce energy consumption, in-network aggregation has been already extensively studied in recent years [2, 6]. Since packets are often exchanged by unicast messages or by broadcast messages, in-network aggregation protocols can be classified into unicast-based in-network aggregation protocols and broadcast-based in-network aggregation protocols. AswefocusonbothenergyconsumptionandQoS,we limit ourselves to lossless QoS-aware in-network aggregation approaches. With lossless we mean that the initial data can be reconstructed and we do not consider aggregation functions such as minimum, maximum, and average. In unicast-based in-network aggregation, aggregation is performed along the intermediate nodes of the routing path. It goes without saying that the routing algorithm and the routing path have a profound influence on the performance of the in-network aggregation. Examples can be found in [7 10]. In LUMP [7],asimpledataaggregationprotocolwhich enables QoS support for applications is proposed. Therefore, it prioritizes packets for differentiated services and facilitates aggregation decisions. The architecture has a cross-layer design and is a completely independent module residing between data link and network layer. The priority level represents the tolerable end-to-end latency of the packet. The authors of [8] investigate data gathering and aggregation in a distributed, multihop sensor network under specific QoS constraints. Firstly, delay controlled Data Aggregation and Processing (Q-DAP) is performed at the intermediate nodes in a distributed fashion. Each node can decide independently if it performs aggregation. If the delay constraint can be satisfied, the report is deferred for a fixed time interval with a certain probability; otherwise, it is sent to thenexthop.ifitcannotbesatisfiedinanycase,itis discarded. Secondly, a Localized Adaptive Data Collection and Aggregation (LADCA) approach is proposed for the end nodes. This algorithm defines the data sample rate taking into account energy-efficiency, delay, accuracy, and buffer overflow. In [9], Padmanabh and Vuppala present an adaptive data aggregation algorithm with bursty sources in wireless sensor networks.inthispaper,bothlossyandlosslessaggregation schemesaretakenintoaccount.furthermore,thedegreeof aggregation is a controllable parameter and buffer management is used to optimize the QoS by minimizing the packet lossduetobufferoverflow. Finally, Akkaya et al. [10] describe an algorithm for achieving maximal possible energy savings through data aggregation while meeting the desired level of timeliness. In order to perform service differentiation and ensure bounded delay for constrained traffic, a weighted fair queuing based mechanism is employed. Broadcast-based in-network aggregation takes advantage of the density of sensor networks in which one sensor node often has many neighbors. By broadcasting a packet, the packet can be received by many neighboring nodes, and for this reason, this approach is often used to increase reliability of multipath routing. Most broadcast-based in-network aggregation techniques in the literature focus however on lossy aggregation (by using aggregation functions such as minimum, maximum, and average) and sending this aggregated value by multiplepathstothesink.afocusherebyisdevelopingduplicatesensitive aggregation functions since broadcast aggregation can lead to incorrect values at the sink node[11 14]. This is however out of scope of this paper. In AIDA [15], a lossless adaptive application-independent data aggregation mechanism is presented. This solution contains an aggregation module that resides between the data link and network layer. Aggregation decisions are made in accordance with an adaptive feedback-based packetscheduling scheme that dynamically controls the degree of aggregation in accordance with the MAC delay. This dynamic feedback scheme is based on the overall queuing delay imposed on AIDA payloads that are waiting for transmission. The default degree of aggregation is 1, while if the traffic builds up, a greater degree of aggregation is allowed prior to sending. If the packets that are ready to be aggregated are targeting the same next-hop node, AIDA sends a manycast packet(oraunicastpacketinthecasethatthereisonlyone packet) with the target node specified. However, when there arenetworkpacketsthathavetobeaggregatedwithadifferent next-hop address, these packets are aggregated into a single packet and the MAC broadcast address is used as destination. A drawback of AIDA is that it only tries to reduce end-to-end delay, and energy consumption is more considered as benefit than as trade-off value. Our focus is on energy reduction within the QoS constraints. 3. Broadcast Aggregation 3.1. Definition and Problem Statement. When performing innetwork aggregation, there will always be a trade-off between energy and QoS. For instance, a higher degree of aggregation will lead to less energy consumption, but packets will have to wait longer in the queue and, as a consequence, the delay increases. Since in this paper the focus is both on energy reduction and QoS improvement, we try to send as much packets with the highest possible degree of aggregation, within the limits of the given QoS constraints. So packets will be kept in the queue as long as the maximum degree

80 International Journal of Distributed Sensor Networks 3 of aggregation (or the maximum number of packets that physically can be aggregated into one single packet) is not reached. However, since packets will have a limit on their maximum end-to-end delay, a timeout time is introduced. This is the time that a packet maximally may reside inside a node before it should be sent to the next-hop node. The number of packets that can be aggregated into a single packet strongly depends on the protocols being used for transferring application data. Highly compact proprietary solutions can be used leading to a high DoA. However, even when using IETF-based IoT protocols (e.g., 6LoWPAN, UDP, and CoAP [16])upto5packetscanbeaggregatedascan be seen in Figure 1. Including the PHY header, the MAC header and footer, and the compressed 6LoWPAN/UDP header for link-local addresses, this leaves us up to 8 bytes for the application header and data, which should be sufficient for simple sensor network transactions such as temperature, humidity, or light responses. When using CoAP as the application protocol, a typical CoAP response without any options consumes 4 bytes for the CoAP header and 1 byte for the payload marker, leaving 3 bytes available for the actual payload. In traditional unicast aggregation, many small packets with the same (next-hop) destination are aggregated into one big packet and sent by unicast to the next-hop node along the routing path. In typical source-to-sink applications such as temperature monitoring, there are many aggregation possibilities since many packets are routed to the sink. A high degree of aggregation is possible and, as a consequence, much energy can be saved and QoS is only marginally affected. However, in the IoT, a huge amount of devices may be interconnected with many bindings between individual devices (e.g., sensor-actuator interactions), and as a consequence, many different nodes can send packets to many different destination. Each intermediate routing node may contain many packets that have to be routed to different destinations. This leads to fewer aggregation candidates, so packets will have to be routed without being aggregated, or with a lower degree of aggregation. However, when both the timeout time and the predefined DoA max value (= the maximum number of packets that can be aggregated into one single packet) are not yet reached, packets will be dropped becausethequeueisfullyoccupiedwithpacketsthatare waiting for the required number of aggregate candidates or for their timeout time. This is illustrated in Figure 2. Alternatively, instead of dropping packets, we could also send packets before their timeout time, but this will again increase the energy consumption. Furthermore, more packets that are not aggregated will lead to more packets in the air and a higher drop probability due to channel issues (collisions, etc.). To overcome these issues, we propose to use broadcast aggregation. Instead of aggregating packets with the same (next-hop) destination and sending them by unicast, packets are aggregated independent of their (next-hop) destination and sent by broadcast. This decouples aggregation from the routing path. In unicast communication, packets are sent to a single destination on the routing path, while by broadcast communication, a transmitted packet is received by every node within the coverage area. The receiving nodes can then extract the packet and retrieve the packet parts that are destined for them. This broadcast aggregation mechanism is visualized in Figure Performance Analysis. In the following, we compare broadcast aggregation with unicast aggregation. As already explained in Section 3.1, broadcast aggregation is performed as soon as the number of available packets reaches the predefined DoA max value or when the maximum timeout time has reached. To summarize, broadcast aggregation is performed. (i) When there are DoA max packets in the system, DoA max packets are aggregated and sent by broadcast B=DoA max (ii) When fewer than DoA max packets are in the system and the maximum timeout time has reached for at least one of the packets B<DoA max. With B thenumberofpacketsthatareaggregatedina single packet that is broadcasted to all 1-hop neighbors. For unicast aggregation, it is not possible to select the first DoA max packets since these packets can have different nexthop destinations. Unicast aggregation is therefore performed. (i) When there are DoA max packets with the same nexthop destination in the system U=DoA max (ii) When fewer than DoA max packets with the same nexthop destination areinthesystemandthemaximum timeout time has reached for at least one of the packets with the same next-hop destination U < DoA max. With U the number of packets that are aggregated in a single packet that is sent to a single next-hop node using unicast. To compare the broadcast aggregation with unicast aggregation, we can use the M/M/1/GD/K/ queuing system. This is a queuing system in which the interarrival times and service time are exponentially distributed with rate λ and rate μ, respectively. There is only one serving unit and there is a general queue discipline. However, only K packets can enter the system. This queuing system is shown in Figure 4. For λ =μand with ρ = λ/μ the steady-state probabilities are given by: p 0 = 1 ρ 1 ρ K+1, p i =ρ i p 0 (i=1,2,...,k), p i =0 (i = K + 1, K + 2,...) Reliability. In a M/M/1/GD/K/ queuing system, the drop probability equals the probability that a new packet arrives when there are already K packets in the queue. Substituting p 0 in p K, the reliability can be expressed as (1) R=1 p K =1 ρk (1 ρ). (2) 1 ρk+1

81 4 International Journal of Distributed Sensor Networks Bytes: 5 1 SYNC header PHY header 127 PHY payload MHR MAC payload MFR 7 MHR MAC payload MFR MHR MAC payload 2 MFR 6 Compressed 6LoWPAN/UDP HR 8 Application HR + data Figure 1: Minimal aggregated packet structure for Internet of Things packets based on packets with SYNC (synchronization) header, PHY (physical) header, MHR (MAC header), MFR (MAC footer), and compressed 6LoWPAN/UDP header for link-local addresses where up to 5 packets can be aggregated. DoA =5,K=10 New incoming packet Drop packet Figure 2: Queue overflow with predefined DoA max = 5 and maximum queue size K = 10.A packet drop occurs since a new packet enters the system while the queue is fully occupied with packets that are waiting for the required number of aggregate candidates or for their timeout time. In Figure 5, the reliability (or the chance that a packet is not dropped from the queue) is shown for different μ values. The evaluation of (2) is done for K=15. This is a realistic assumption of the available buffer space on today s sensor nodes. For instance, the Zolertia Z1 sensor node [17] has 8kB of RAM, but 4 6 kb is often used for code, which leaves only 2 4 kb for buffering. We can see from the figure that doubling the service rate can significantly improve the reliability, or in other words, doubling the service rate leads to an increased load capacity before a packet will be dropped. So when broadcast aggregation is applied and the maximum timeout time has not passed, packets can be aggregated as soon as there are enough packets to meet the DoA max value which will prevent from queue overflows. In unicast aggregation, the network topology determines the service rate since when there are many different streams with different next-hop destinations; it can take some time before there are enough packets with the same next-hop destination before aggregation can occur. So while B and U are the same, B will be reached earlier, which releases the queue earlier and increases the reliability. X Y Z t 1 t 2 t 3 4 X Y Z t 4 t 4 t X Y Z Figure 3: Broadcast aggregation mechanism: nodes 1, 2, and 3 send data packets (resp., X, Y,andZ)sequentially(t 1 <t 2 <t 3 )towards, respectively, nodes 7, 6, and 5. These packets wait in the intermediate node (node 4) until the requested DoA is reached and then they are aggregated independent of their next-hop destination. This aggregated packet is then sent by broadcast on t 4 to the neighboring nodes. Destination nodes 5, 6, and 7 can then extract the data part that is destined for them (resp., Z, Y,andX). In this section, we mentioned the node reliability caused by queue overflows. The end-to-end reliability will also be influenced by specific protocol implementations and channel related issues (e.g., collisions) Delay. The delay or the average time that a packet resides in a node can be calculated from Little s queuing formula: D avg = N λ eff (3) with N the average number of packets in the node and λ eff the average number of packets that actually enter the node. Recall that some packets may arrive in a fully occupied queue X X Y Y Z Z

82 International Journal of Distributed Sensor Networks 5 λ λ λ λ λ λ K 1 K Reliability (%) μ μ μ Figure 4: M/M/1/GD/K/ queuing system μ=10 μ=20 μ=30 μ λ (packets/min) Figure 5: Reliability analysis for different μ values (with μ the service rate, λ the interarrival rate, and K = 15the maximum queue occupation). and will hence be dropped. These packets are not considered in λ eff. λ eff canbecalculatedas λ eff = K k=0 and N canbecalculatedas N= μ λ k p k =λ(1 p K ) (4) K k=0 k p k. (5) Combining (3) with(4), (5), and (1), the average delay D avg that a packet resides inside the system can be expressed as follows: D avg = ρ[1 (K+1) ρ K +Kρ K+1 ] (1 ρ K+1 )(1 ρ)λ(1 (ρ K (1 ρ)/(1 ρ K+1 ))). In Figure 6, the average delay is given for different μ values that are expressed in terms of the initial arrival rate λ. The evaluation of (6) is done for K=15.Wecanseefromthe figure that doubling the service rate can significantly reduce the average delay that a packet resides inside the system. So again, while B and U are the same, B will be reached earlier, which decreases the average (end-to-end) delay. μ (6) Delay (min) λ (packets/min) μ=10 μ=20 μ=30 Figure 6: Delay analysis for different μ values (with μ the service rate, λ the interarrival rate, and K = 15 the maximum queue occupation) Throughput. The end-to-end throughput of the network is expressed as the average rate of successfully delivered data from source to destination. This end-to-end throughput is influenced by two parts: a node throughput and a channel throughput.thenodethroughputequalstheservicerate μ. The channel throughput depends on channel overhead, transmission errors, and so forth. Aggregation in general is beneficial for channel utilization since fewer packets have to contend for the medium which leads to fewer packet drops Energy. In [3], we have shown that energy consumption can be reduced when aggregation is applied together with a sleep-wake-up scheme. Aggregation leads to fewer transmissions and receptions which results in more sleep opportunities and a reduced energy consumption level. The transmission energy consumption can be calculated as E reduction tx =1 E A E NA (7) in which E A is the total transmission energy consumed when aggregation is performed and E NA is the total energy consumed when no aggregation is performed. Expressing E A in terms of E NA,thisleadsto E reduction tx 1 E TO +E A H + DoA avge NA DATA E TO +E NA H +ENA DATA 1 DoA avg (8) in which E TO is the energy consumption caused by transmission overhead per packet transmission. This contains the Clear Channel Assessment (CCA) time, the RX-TX turnaround time but also the overhead created by the MAC protocol, for example, by beacons. E NA H is the energy consumedtosendanotaggregatedpacketheader,e A H is the

83 6 International Journal of Distributed Sensor Networks energy consumed to send a header of an aggregated packet, and E NA DATA is the energy consumed to send individual data parts. From (8), we can see that the energy reduction depends on the average DoA level (DoA avg ). In general, the more packets that are aggregated (= a higher DoA avg ), the more energy that will be saved. Because in broadcast aggregation more packets are faster aggregated, more energy will be saved. However, the total energy consumption will also depend ontheappliedsleep-wake-upschemeandtheusedmac protocol. Since the impact is very protocol specific, this will be further discussed in the simulation results. 4. Simulation Results 4.1. Reliability. Simulations are done using Castalia, an OMNeT++ based network simulator for simulating wireless sensor networks [18]. We use a network scenario with 256 nodes, deployed in a square grid with size 75 m. This setup was chosen according to the RFC which describes the requirements for home automation routing in low power and lossy networks [19]. An own implementation of the DYMO protocol is chosen as routing protocol and the low power listening tunable MAC protocol that is available in the Castalia simulator is chosen as MAC protocol. The protocol operation is shown in Figure 7. BeforenodeAtransmitsits data,itsendsaccamessagetocheckifthechannelisclear.if the medium is free, it starts periodically sending small beacon messages during a whole sleep interval. Afterwards, the actual data is sent. When node B receives a beacon message, it stays awake until the data transmission has finished. After this reception, it waits a listen interval before going back to sleep because other packets can follow. This behavior is the same for each node that receives beacon messages, regardless of the fact that this node is the destination for the actual data or not. Simulations are performed with DoA =5and with maximum queue size K=15. In our scenario, each node (except the destination nodes) is transmitting packets to one of 20 randomly chosen destination nodes. A simulation lasts 3600 seconds, but statistics are generated in steady state between 300 and 3000 seconds. After 100 seconds, the network is up and running and the DYMO routes remain active during the entire simulation, as canbeseeninfigure 8. This is, for instance, the case when there is a fixed communication between the sensor and actuator with regular traffic. The large amount of network setup traffic can be explained by the broadcast nature of the RREQ messages sent by the DYMO routing protocol. Simulation statistics later than 3000 seconds are not counted to ensure that all generated packets can reach their destination within the evaluated time. Furthermore, simulations are performed for different average traffic rates (from 1 to 30 packets/min) as given on the x-axis of the figures. For instance, when the average traffic rate is 5 packets/min, nodes choose a random traffic rate between 1 and 10 packets/min and send their packets according to this traffic rate without jitter. A higher average packet rate will lead to a higher network load. To assess the realism of these traffic rates, we can consider ascenariowherearesourceonasensornode(e.g.,the temperature) is being monitored using the CoAP observe option.inthiscase,thesensornodewillsendanotification every time the resource state changes (e.g., the temperature) or when its max-age value, which indicates the freshness of the resource state, expires. The default max-age value is 60 seconds, resulting in a minimum traffic rate of 1 packet/min. When values change more frequently, more messages will be sent. The maximum packet timeout was set to 10 seconds. Thisvaluegivesnodesatlowtrafficratesenoughtimetofind aggregate candidates before their timeout time passes. The usedsimulationsettingscanbefoundintable 1. Figure 9 shows the overall reliability as the number of application packets received compared to the number of applicationpackets sent. Figures10 and 11 show, respectively, thenumberofpacketsdroppedinthequeueandthenumber of packets dropped due to the lower network layers (Radio, MAC, Physical channel). The packet error rate (PER) is expressed as the number of application packets dropped compared to the number of application packets sent. Packets in the queue are dropped due to queue overflows, while packets dropped by the lower layers are caused by the fact that a sensor node cannot simultaneously send and receive, and packets can collide. From Figure 10, wecanseethatforunicastaggregation, starting from an average traffic rate of 3 packets/min, packet drops occur in the queue. This is not the case in the no aggregation and broadcast aggregation scenario, since in the no aggregation scenario, packets do not have to wait for aggregation candidates and can be sent immediately. In the broadcast aggregation scenario, aggregated packets can be sent as soon as there are 5 packets in the queue. This effect can be seen infigure 12that shows the average queue occupation. Wecanseethattheaveragequeueoccupationisonaverage 2 packets lower for broadcast aggregation than for unicast aggregation. Figure 11 shows on its turn the impact of the lower network layers on the reliability. The figure shows that in the beginning the packet error rate (PER) is relative high, then drops,andfinallyslowlyincreasesagain.thehighperatlow traffic rates can be explained by the fact that each node will have approximately the same packet rate. Remember that the packet rate on the x-axis is the average packet rate of all the nodes. So at low packet rates, most sensor nodes will have thesamepacketrateandwilltrytoenterthemediumon the same time. As a consequence, more packets will be lost. From an average traffic rate of 5 packets/min, the PER starts increasing since more packets are in the air, which leads to a higher PER. Furthermore, we can see that the PER is higher in the no aggregation scenario, since when packets are not aggregated, more packets are in the air and more packets can be lost. Since both unicast and broadcast aggregation combine packets into one big packet, we should expect that the packet loss due to the lower network layers should be approximately thesame.thisishowevernotthecaseatlowtrafficrates.this canbeexplainedbyafacttowhichwereferinthefollowing as partial aggregation.

84 International Journal of Distributed Sensor Networks 7 Listen/sleep scheme Listen 100ms Sleep 400ms Node A Data CCA Node B Listen Listen RX Listen Sleep Node C Listen Sleep Listen Sleep Figure 7: Operation of the tunable MAC protocol. Node A sends a data packet to node B by first performing CCA, afterwards sending beacon messages, and finally sending the data packet. Node B stays awake when it receives one or more beacon messages until the data transfer is completed. Node C cannot receive these beacon messages and will again go to sleep after the listen interval. Table 1: Simulation settings. Parameter Value Simulation time 2700 s Simulation field 75 m 75 m Number of nodes 256 Node deployment grid Routing protocol DYMO MAC protocol Tunable CSMA based MAC (Castalia) with duty cycle 0.2 (100 ms listening/400 ms sleeping) Radio CC2420 with no transmission errors Wireless channel No interference Collision domain ±16 m, depends on the receiver signal strength (±2hops) Queue size (K) 15 DoA 5 A partly aggregated packet is a packet that contains fewer than DoA aggregated packet parts. For instance, when the DoA was set to 5, a partly aggregated packet will contain fewer than 5 aggregated packet parts. This effect is mainly caused by the timeout time that was introduced. At low traffic rates, the timeout time will pass before DoA packet parts canbeaggregated.atthismoment,theaggregatedpacket will be sent with as much packet parts that are available. This partial aggregation has a cascading effect. When a packet with DoA packet parts is sent, on the following node, thereareobviouslyenoughpacketspartstosendanew aggregated packet to the next-hop node. However, when a partly aggregated packet was received, this node will on its turn wait again on DoA packet parts, and again, the timeout time can pass. Figure 13 shows this actual average DoA at which aggregation is performed. We can see that, at lower traffic rates, this DoA value is indeed much lower due to the partial aggregation. Partly aggregated packets however also mean more packets in the air and an increased chance on packet drops. The higher the load in the network, the fewer partly aggregated packets that exist, and the fewer packet drops that occur. We indeed observe in Figure 11 thatforhigherloadstheper issimilarforunicastandbroadcastaggregation.figure 14 shows the ratio of MAC data packets sent to application packets sent. This ratio can be higher than 1 since the MAC data packets are measured through the network, and thus a MAC packet is measured on each intermediate node, while the application packets are only measured on the sending node. The figure clearly shows the effect of partial aggregation at low traffic rates. For the no aggregation scenario, the ratio equals the average number of hops, which is 4. When packets are aggregated, this ratio decreases and for high traffic rates, we can see that this ratio drops below 1 because there maximal aggregation (up to 5 packets) occurs. We can see that with broadcast aggregation, partial aggregation is reduced since more packets will be aggregated faster.

85 8 International Journal of Distributed Sensor Networks Total number of route setup messages Reliability (%) Minimum packet rate Maximum packet rate Simulation time (s) Figure 8: Steady state calculation No aggregation Unicast aggregation Broadcast aggregation Traffic rate (packets/min) Figure 9: End-to-end reliability. PER ( %) Traffic rate (packets/min) No aggregation Unicast aggregation Broadcast aggregation Figure 10: Packet error rate (PER): number of dropped packets in the queue compared with the number of packets sent. PER (%) Traffic rate (packets/min) No aggregation Unicast aggregation Broadcast aggregation Looking back on Figure 9, wecanseethattheendto-end reliability of broadcast aggregation is increased up to 23% compared with the no aggregation scenario and up to 15% compared with the unicast aggregation scenario. Although there are no packet drops in the queue in the no aggregation scenario, we see that the reliability is significantly lower than the broadcast aggregation scenario. This has been explained earlier by the fact that with broadcast aggregation, fewer transmissions occur which is beneficial for the channel occupation and as a consequence, the reliability increases Delay. Figure 15 displays the average end-to-end packet delay. This packet delay is measured on the individual application packets, not on the aggregated packets. We can Figure 11: Packet error rate (PER): number of dropped packets due to the lower network layers compared with the number of packets sent. see that with broadcast aggregation, the delay can be reduced up to 52% compared with unicast aggregation. The delay is of course higher than with no aggregation since with aggregation packets are waiting in the queue for a certain period in order to have fewer transmissions and save energy Throughput. In Figure 16, the end-to-end throughput of the network is demonstrated. This throughput is expressed as the average number of successful received packets per secondpernode.wecanseeatanaveragetrafficrateof10 packets/min, up to 1.88 packets/min are more received with

86 International Journal of Distributed Sensor Networks 9 Average queue occupation (packets) No aggregation Unicast aggregation Broadcast aggregation Traffic rate (packets/min) Figure 12: Average queue occupation. Sent MAC packets versus sent application packets No aggregation Unicast aggregation Broadcast aggregation Traffic rate (packets/min) Figure 14: Ratio of MAC data packets sent to application packets sent. Average DoA Average end-to-end packet delay (s) No aggregation Unicast aggregation Broadcast aggregation Traffic rate (packets/min) Figure 13: Average degree of aggregation (DoA) level at which aggregation occurs. broadcast aggregation compared with no aggregation and at traffic rate of 12 packets/min, up to 1.59 packets/min are more received with broadcast aggregation compared with unicast aggregation. This is mainly caused by faster transmission and less packet loss Energy. Figure 17 shows that the total transmit energy consumption with broadcast aggregation can be reduced up to 29% compared with unicast aggregation and up to 71% compared with no aggregation. From (8) and the values in Table 2,we could calculate that theoretically up to 74% of energy could be saved when broadcast aggregation with DoA avg =5is applied compared Traffic rate (packets/min) No aggregation Unicast aggregation Broadcast aggregation Figure 15: Average end-to-end packet delay. Table2:Energysettings. Parameter Value Transmit power mw Listen power 62 mw Sleep power 1.4 mw Receive power 62 mw Beacon size 16 bytes Beacons per transmission 8 CCA time ms Rx-Tx turnaround time ms with no aggregation. The difference between the theoretical and the actual energy reduction could be explained by the

87 10 International Journal of Distributed Sensor Networks Throughput (packets/min) No aggregation Unicast aggregation Broadcast aggregation Traffic rate (packets/min) Figure 16: End-to-end data throughput per node. Total energy consumption (J) Traffic rate (packets/min) No aggregation Unicast aggregation Broadcast aggregation Figure 18: Total energy consumption Transmit energy consumption (J) Energy consumption (J) No aggregation Unicast aggregation Broadcast aggregation Traffic rate (packets/min) Figure 17: Total transmit energy consumption. fact that in our simulation DoA avg = 5 is not reached. Furthermore, we can see that starting from an average traffic rate of 21 packets/min, broadcast aggregation consumes more energy than unicast aggregation. This is because lost packets arenottakenintoaccount.indeed,withunicastaggregation, more packets are lost and since lost packets are not further routed, less energy is consumed than expected. Figures 18 and 19 show, respectively, the total energy consumption and the average energy consumption per successfully delivered packet. Taking into account the total energy consumption (Figure 18), 27% of energy can be saved compared with no aggregation and up to 13% compared with unicast aggregation. For the no aggregation scenario, we can also see that the total energy consumption stagnates and reaches a maximum Traffic rate (packets/min) No aggregation Unicast aggregation Broadcast aggregation Figure 19: Average energy consumption per successful delivered data packet. because the network becomes congested. For the unicast and broadcast aggregation scenario, this maximum is expected somewhat later. When we take a look at the energy per successfully delivered packet (Figure 19), we can see that up to 38% of energy can be saved compared with no aggregation and up to 19% compared with unicast aggregation. From the figures, we see that the total energy reduction is much lower than the total transmit energy reduction. The reason can be found in the used MAC protocol and can be seen in Figure 20 that shows the energy distribution of broadcast aggregation in terms of transmit energy, idle

88 International Journal of Distributed Sensor Networks 11 Energy consumption broadcast aggregation (J) Transmit energy Receive energy Traffic rate (packets/min) Idle listen energy Sleep energy Figure 20: Total broadcast aggregation energy distribution. Receive energy consumption (J) Traffic rate (packets/min) No aggregation Unicast aggregation Broadcast aggregation Figure 22: Total receive energy consumption. Idle listening energy consumption (J) No aggregation Unicast aggregation Broadcast aggregation Traffic rate (packets/min) Figure 21: Total idle listening energy consumption. listening energy, receive energy, and sleep energy. We can see that much energy is lost due to idle listening. A detailed overview of the energy contributions in Figure 20 can be found in Figures 21 to 23. In Figure 21, we can see that the idle listening time reaches a maximum in the no aggregation scenario. This can be explained by the used MAC protocol and the load. First, thenumberofsentpacketsincreases,sothenumberofsent beacons also increases and nodes are longer awake. As a consequence, the time spent on idle listening, and thus the resultingenergyconsumption,increases.however,fromacertain point, the time spent on idle listening will decrease since part of this time will be used to transmit/forward/receive the increasing amount of packets. This behavior is also expected for unicast and broadcast aggregation, but since the amount of traffic increases slower since more packets are aggregated, this point will occur at bigger traffic rates. Figure 22 shows the total receive energy consumption. This is the energy consumed by receiving both beacons and data, even the data that is not destined for this node. Finally, Figure 23 shows the total amount of energy that is consumed by sleeping. Both from Figures 22 and 23, we can see that broadcast aggregation seems to perform worse than unicast aggregation starting from an average traffic rate of 21 packets/min. This is however not true since lost packets are not taken into account. Unicast aggregation leads to more lost packets which are not routed. This results in less received energy consumption and more sleep energy consumption. 5. Conclusion and Future Work In-network aggregation is a very efficient way to reduce energy consumption in wireless sensor networks. However, the traditional unicast-based in-network aggregation only works well for source-to-sink traffic. When there are multiple destinations, as is the case in the Internet of Things, aggregation becomes slower, delay increases, reliability drops, and energy consumption increases. In this paper, we propose to use broadcast aggregation as a solution to overcome these drawbacks. We have shown that broadcast aggregation reduces the average queue occupation with 2 (of the 15 available) places, which leads to fewer packet drops.thisleadsonitsturntoathroughputandreliability increase up to 23% compared with no aggregation and up to 15% compared with unicast aggregation. Moreover, we have shown that packets become less dependent of the individual timeouts per destination which reduces the drawbacks of partial aggregation. Furthermorewehaveshownthattheaveragequeuedelay is decreased by 52% compared with unicast aggregation becauseaggregationcanbeperformedfaster.

89 12 International Journal of Distributed Sensor Networks Sleep energy consumption (J) Traffic rate (packets/min) No aggregation Unicast aggregation Broadcast aggregation Figure 23: Total sleep energy consumption. Finally, the average DoA is higher, which leads to fewer packets, less overhead, and as a consequence, an energy reduction up to 27% compared with no aggregation and up to 13% compared with unicast aggregation. The overall energy reduction is mainly realized by reducing the transmit energy. Further gains are to be expected by reducing the idle listening energy as well. To this end, several MAC optimizations will be investigated in order to further reduce the energy consumption. For instance, we will investigate the use of destination addresses in the beacon messages, so that nodes that are not the intended receiver can immediately go back to sleep. Conflict of Interests The authors declare that there is no conflict of interests regarding the publication of this paper. Acknowledgments The research of Evy Troubleyn is funded by a Ph.D. grant of The Institute for the Promotion of Innovation through Science and Technology in Flanders (IWT-Vlaanderen). This research is also partially funded by the FWO Flanders through projects 3G and 3G References [1] I. F. Akyildiz, T. Melodia, and K. R. Chowdhury, A survey on wireless multimedia sensor networks, Computer Networks, vol. 51,no.4,pp ,2007. [2] E. Fasolo, M. Rossi, J. Widmer, and M. Zorzi, In-network aggregation techniques for wireless sensor networks: a survey, IEEE Wireless Communications,vol.14,no.2,pp.70 87,2007. [3] E. Troubleyn, I. Moerman, and P. Demeester, Protocol-independent qos-aware in-network aggregation in wireless sensor networks, Wireless Networks, [4] Libelium, Top 50 iot sensor applications ranking, 2013, 50 iot sensor applications ranking/. [5] D. Christin, A. Reinhardt, P. S. Mogre, and R. Steinmetz, Wireless sensor networks and the internet of things: selected challenges, in Proceedings of the 8th GI/ITG KuVS Fachgespräch Sensor Networks,pp.31 34,2009. [6] R. Rajagopalan and P. K. Varshney, Data-aggregation techniques in sensor networks: a survey, IEEE Communications Surveys and Tutorials,vol.8,no.4,pp ,2006. [7] J. Jeong, J. Kim, W. Cha, H. Kim, S. Kim, and P. Mah, A QoS-aware data aggregation in wireless sensor networks, in Proceedings of the 12th International Conference on Advanced Communication Technology: ICT for Green Growth and Sustainable Development (ICACT 10), pp , February [8] J.Zhu,S.Papavassiliou,andJ.Yang, AdaptivelocalizedQoSconstrained data aggregation and processing in distributed sensor networks, IEEE Transactions on Parallel and Distributed Systems,vol.17,no.9,pp ,2006. [9] K. Padmanabh and S. Vuppala, An adaptive data aggregation algorithm in wireless sensor network with bursty source, Wireless Sensor Network,vol.1,no.3,pp ,2009. [10] K. Akkaya, M. Younis, and M. Youssef, Efficient aggregation of delay-constrained data in wireless sensor networks, in Proceedings of the 3rd ACS/IEEE International Conference on Computer Systems and Applications, pp , January [11] S. Nath, P. B. Gibbons, S. Seshan, and Z. R. Anderson, Synopsis diffusion for robust aggregation in sensor networks, in Proceedings of the Second International Conference on Embedded Networked Sensor Systems (SenSys 04), pp , November [12] S. Motegi, K. Yoshihara, and H. Horiuchi, DAG based innetwork aggregation for sensor network monitoring, in Proceedings of the International Symposium on Applications and the Internet (SAINT 06), pp , January [13] S. Madden, M. J. Franklin, J. M. Hellerstein, and W. Hong, Tag: a tiny aggregation service for ad-hoc sensor networks, SIGOPS Operating System Review,vol.36,pp ,2002. [14] S. Gobriel, S. Khattab, D. Mossé, J. Brustoloni, and R. Melhem, RideSharing: fault tolerant aggregation in sensor networks using corrective actions, in Proceedings of the 3rd Annual IEEE Communications Society on Sensor and Ad hoc Communications and Networks, pp , Reston, Va, USA, September [15] T. He, B. M. Blum, J. A. Stankovic, and T. Abdelzaher, Aida: adaptive application independent data aggregation in wireless sensor networks, ACMTransactionsonEmbeddedComputing System,vol.3,no.2,pp ,2004. [16] I. Ishaq, D. Carels, and G. K. Teklemariam, Ietf standardization in the field of the internet of things (iot): a survey, Journal of Sensor and Actuator Networks,vol.2,no.2,pp ,2013. [17] Zolertia Z1, Wireless Sensor Node, 2014, RevC Datasheet.pdf. [18] NICTA, Castalia wireless sensor network simulator, 2013, [19] IETF, Rfc home automation, 2013, draft-ietf-roll-home-routing-reqs-11/.

90 Hindawi Publishing Corporation International Journal of Distributed Sensor Networks Volume 2014, Article ID , 6 pages Research Article A Zone-Based Media Independent Information Service for IEEE Networks Fábio Buiati, 1,2 Luis Javier Garcia Villalba, 1 Delfín Rupérez Cañas, 1 Ana Lucila Sandoval Orozco, 1 and Tai-hoon Kim 3 1 Group of Analysis, Security and Systems (GASS) and Department of Software Engineering and Artificial Intelligence (DISIA), School of Computer Science, Office 431 Complutense University of Madrid (UCM), Madrid, Spain 2 Electrical Engineering Department, University of Brasilia, Brasilia, DF, Brazil 3 Department of Convergence Security, Sungshin Women s University, Dongseon-dong 3-ga, Seoul , Republic of Korea Correspondence should be addressed to Luis Javier Garcia Villalba; javiergv@fdi.ucm.es Received 4 September 2013; Accepted 27 November 2013; Published 9 March 2014 Academic Editor: Naveen Chilamkurti Copyright 2014 Fábio Buiati et al. This is an open access article distributed under the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited. Next generation networks integrate different wireless technologies, including Wi-Fi, Wi-Max, and 3GPP (UMTS, HSPA, and/or LTE), in which the mobile node (MN) has the opportunity to switch from one network to another, under an always best connected scheme. In such heterogeneous environment, discovering which types of network connectivity and services are available is a critical challenge. The IEEE standard specifies a network information server entity providing network information within a geographical area by which the MN can discover a service or a network. In this paper, we propose a zone-based media independent information service using the IEEE standard to accelerate the neighbor discovery procedure. In the proposed scheme, the access networks are associated and grouped in mobility zones, through an efficient set of rules, to minimize the amount of control messages flowing in the core network. Through a NS-2 based simulation, the results demonstrate that the proposed scheme reduces the neighbor discovery delay as well as the signaling overhead if compared with the standard MIIS deployment. 1. Introduction In the heterogeneous wireless environments, discovering available networks is one of the most challenging issues. Typically, the available information about candidate networks provided by advertisements messages (e.g., beacon frames, DCD) is rather minimal. To improve the MN experience, the IEEE media independent handover standard [1] specifies a media independent information service (MIIS) supporting various information elements which provide network information within a geographical area. Based on that information (such as available bandwidth, cost per use, and security) from several access networks, the MN can qualitatively choose a better handover candidate amongst the possible ones, taking a more accurate handover decision. To access this information, the MN queries a MIIS server using specific query IEEE information messages, which transport the information, and the MIIS server replies with information about the availability of access networks in a determined geographic area. To exploit geographic location, the IEEE standard allows some parameters to be introduced into the query message that can be used by the MIIS to refine its response. Between the available values, we can mention the querier location parameter which enables themntosendthequeryrequestwithitscurrentlocation information. The value field contains either the MNs current location measurement (the MN can use global positioning system (GPS) or other kind of service location equipment) or, when the MN does not have its current location information, an observed link-layer address (e.g., beacon frame or some broadcast mechanism for other technologies) that the MIIS server will be able to use as a hint to establish an estimate of the clients current location. For many applications, this is sufficient but others require much more specific, granular, and precise location data. Moving between networks requires an accurate location map of users and networks. In a crowded and populated city (Manhattan use case), tens or hundreds of meters may strongly impact the number of networks in

91 2 International Journal of Distributed Sensor Networks the response message from a MIIS server, leading to the fact that the user experiences delay and overhead. Looking at the standard [1]aswellastherelatedwork[2 14], it is typically assumed that the MIIS server is aware of the MNs location, sending a response message with information about candidate networks. The representation of the MIIS role and usability is very lacking in terms of usage details, architecture, and characteristics used. Moreover, the existing MIIS deployments usually consider a network-centric approach not being scalable systems. In this way, we propose in this paper a zone-based MIIS architecture which exploits the geographical location of the networks by splitting the coverage into different mobility zones, enhancing the neighbor discovery task. Using such a scheme, the MN is provided with a complete and consistent view of detailed handover possibilities information. The aim of this architecture is threefold. First, MN without GPS built can use the MIIS server and obtain network-related information within a geographical area. Second, MN will experience less neighbor discovery delay if compared with the standard MIIS architecture. Finally, the proposed architecture reduces the signaling load in the operators core network. The paper is organized as follows. In Section 2,therelated work focused on the IEEE services is presented. Then, we briefly describe the IEEE standard. We then move on to present the zone-based MIIS architecture deployment. Then, we evaluate the performance of our proposed MIIS system in Section 5,throughaNS-2basedsimulation.Finally, we conclude this work in Section 6 with some final considerations. 2. Related Work A number of mobility mechanisms in heterogeneous environments have been proposed [2 14] that employ models relying on the IEEE MIIS service. In [2], a vertical handover scheme between UMTS and WiMAX employing the IEEE framework is proposed. They use the MIIS service in order to obtain relevant information from neighbourhood networks. Seol and Chung [3] propose an interesting vertical handover solution for WiMAX and 3GPP networks based on IEEE services taking a networkbased mobility management approach using proxy mobile IP. Stevens Navarro et al. [4] use the IEEE MIIS service to determine the conditions under which vertical handoff should be performed. The problem is formulated as a Markov decision process with the objective of maximizing the total expected reward per connection. The proposal by Liu et al. [5] istoobtainnetworkmaps from MIIS service and provide this information to the elimination-based cost function to enable energy efficient handover. Christakos et al. [6] explore the MIIS service to improve mobility performance for FMIPv6 by providing authentication information allowing the MN to authenticate with the target network while connected elsewhere on the network. They focus especially on information that aids the authentication process, providing MNs with authentication information that they would not normally have until they connect to a new PoA. Mussabbir et al. [7] define a heterogeneous network information container for facilitating in thestoreandretrievalofthel2andl3staticinformation of neighboring networks obtained through the IEEE MIIS. In [8], the authors present a timely effective handover architecture based on the neighbor network information. In the proposed architecture, they estimate the exact required handover time based using the network information obtained by the MIIS server. Lim et al. [9] makeuseofthemiis serviceprovidingthemnwithavalidchannelliststored on the MIIS server. Upon receiving a response message, themnperformsaselectivescanningprocedure,reducing the network discovery time. In [10], the authors propose integrating the MIH architecture into an IP multimedia subsystem (IMS) in order to optimize the quality of end-to-end service. Their cross-technology architecture considers a MIIS infrastructure, where MIIS servers exchange information such as QoS and cost related parameter. Some authors have started to work in a more detailed MIIS framework and architecture specifications. The authors in [11, 12] introduce an enhanced information server in which the MN periodically reports dynamic information to the MIIS server. Their main contribution is that the MIIS server is able to store, manage, and deliver realtime dynamic information, such as the user preferences, running services, mobile device characteristics, and available network resources. In [13] an enhanced media independent handover framework and mobility management mechanism are proposed. The MIIS service is used to collect linklayer and application layer information from the networks. Finally, in [14] a decentralized MIIS approach is specified. The envisioned architecture is based on a hierarchical distributed hash table (DHT), where the MIIS information database is also maintained by the mobile users. The previously mentioned works typically assume that the MIIS server is aware of the MNs location, sending related network information from a particular geographical area. No attention is given to how the MN can obtain information if GPS equipment is not available. Our framework contributes to the MN neighboring network information acquisition even without using any GPS equipment, by specifying a zonebased MIIS architecture. 3. IEEE Standard The IEEE standard [1] specifies a media independent handover (MIH) framework that facilitates handover in heterogeneous access networks by exchanging information and defining commands and event triggers to assist in the handover decision making process. Specifically, the standard consists of a framework that enables service continuity while a MN transitions between heterogeneous link-layer technologies. Also, it defines a new logical entity created therein called the media independent handover function (MIHF). The MIHF also provides three primary services: event services, command services, and information services.

92 International Journal of Distributed Sensor Networks 3 ZMIIS i Operator s core network ZMIIS i+1 ZMIIS l Mobility zone i Mobility zone i+1 Mobility zone m MN Figure 1: Zone-based MIIS architecture. The media independent event service (MIES) is responsible for detecting events at lower layers and reporting them from both local and remote interfaces to the upper layers (the MIH users). A transport protocol is needed for supporting remote events. These events may indicate changes in state and transmission behavior of the physical data link and logical link layers or predict state changes of these layers. The media independent command service (MICS) refers to the commands sent from MIH users to the lower (physical data link, and logical link) layers in order to control it. The commands generally carry the upper layer decisions to the lower layers on the local device entity or at the remote entity. MIH users may utilize command services to determine the status of links and/or control the multimode device for optimal performance. The media independent information service (MIIS) provides a framework and corresponding mechanisms by means of which a MIHF entity may discover and obtain network information existing within a geographical area to facilitate the handovers. MIIS includes support for various information elements which provide information that is essential for a network selector to make intelligent handover decisions. The information may be present in some MIIS server, where themihfinthemnmayaccessit.moreover,themiis provides capability for obtaining information about lower layers like neighbor maps and other link-layer parameters, as well as information about available higher layer services such as internet connectivity, for instance, knowledge of whether security, supported channels, cost per use, networks categories (such as public, enterprise, and home), and QoS supportedmayinfluencethedecisiontoselectsuchanaccess networkduringhandoverprocess. The information supplied by the MIIS is provided in information elements (IE) which can relate to higher layer services such as availability of IP mobility schemes at a certain operator or to lower layer such as link neighbor maps and link configuration parameters. More concretely, information availableviathemiiscanbecategorizedasfollows. (i) General and access network specific information: general overview of different networks, providing coverage within a specific area such as network type, operator, and service identifier. Information including QoS, security, technology revision, and cost is also available. (ii) Link connection point information: information about points of attachment for each access network available, comprising aspects such as MAC address of the access point, geographical location, and channel configuration. (iii) Other information: network, service, or vendor specific information. Detailed information about the IEEE standard, its services, and characteristics can be found in [15 17]. 4. Zone-Based MIIS Architecture In this section we describe the zone-based MIIS architecture and its support for an optimized MN neighbor discovery performance. We propose the splitting of the network coverage area into mobility areas or zones (MZ) as illustrated in Figure 1. Each MZ is composed of several access networks or point of attachments (PoA). A zone MIIS server (ZMIIS) is specified to manage the information details of each one of these MZs. The ZMIIS is able to interchange information with different MZ, with an awareness of which ZMIIS servers are related to which specific access network. Algorithm 1 summarizes the neighbor discovery scheme using the zone-based MIIS architecture. In the initialization phase,themnchecksalltheavailablenetworksandselects one for the current PoA. Upon connecting, the operator also supplies detailed network information about endpoints in the particular MZ. In the movement from one network coverage to another, the MN receives a link detected trigger. As long as the MN detects a new PoA, it looks inside the zone information to check if the detected PoA has better characteristics than the PoA that the MN is actually

93 4 International Journal of Distributed Sensor Networks A Initialization (); B Table T=MZ i information; C while MN movement and detects a new PoA i do D inzone = check (T,PoA i ); E if inzone then F trigger handover decision; else G send query (PoA i ) to MNs ZMIIS server; H ZMIIS contacts the target zones ZMIIS; 0 target zones ZMIIS builds a optimized response; 1 MN receives ZMIIS response; 2 trigger handover decision; Algorithm 1: Neighbor discovery scheme. connected. If the detected PoA belongs to the same MZ, no MIIS query is sent to the ZMIIS server, because the MN holds enough information to take an optimized handover decision. In the case of the detected PoA belonging to a different MZ, the MN sends a MIH get information request message to the ZMIIS server containing the detected PoA identifier (PoA i ). Upon receiving the request, the MNs ZMIIS is able to contact the ZMIIS server from the target zone and obtains the required information. In this way, it knows which ZMIIS server holds the desired information that replies with a MIH get information response message. When moving to a new MZ, the MN automatically obtains information regarding neighboring PoAs within that MZ. In the case of the MN moving between multiples zones, it is useful to maintain the information of each of the MZs for a certain time. Since the MN often moves back and forth between a small set of PoAs, an internal cache can be helpful. This prevents the MN from querying the MIIS server unnecessarily. An important feature in the specification of a MIIS server architecture is the scalability support. As can be seen in Figure 1,wespecifyl ZMIIS i servers, (i = 1, 2, l)withi being the identifier of each ZMIIS and l the number of ZMIIS servers. Also, we define m MZ i,(i = 1,2,...,m)inwhich each MZ can be composed by n PoA i,(i = 1,2,...,n). With this conceptual representation, the proposed architecture is flexible and scalable enough to support different mobility scenarios, even multiple operator environments, a common drawback in the related work. As such, a collaboration agreement should be set between operators for information services availability. Using the zone-based MIIS architecture, the MN receives detailed information only related to its general neighborhood environment, even without any location service equipment. This architecture also distributes the queries over several ZMIIS servers with the objective of reducing operators burdens, evolving into a cheaper and more efficient architecture. 5. Performance Evaluation This section presents the performance evaluation of the proposed MIIS architecture. The simulations were made in Table 1: Simulation parameters. Parameter Default values Other values Topology area Variable ZMIIS, MZ, PoA 4 16, 36 PoA transmission range 100 m MN 1 Mobility model RWP MN speed 2 m/s 4, 6, 8, 10 Pause time 3 s Hop count (MN MIIS) 5 2, 4, 6, 8, 10 Wired delay 5 ms 10, 15, 20, 25 Wireless delay 8 ms Simulation time 1 hour the NS-2 simulator [18]. We have modified the original module, adding the MIIS functionality in a decentralized way. Table 1 shows the network parameters considered for the simulation. The number of ZMIIS servers, MZs, and Wi-Fi PoAs are variable (4, 16, and 36). We have chosen these values to always keep a topological square area, obtaining more reliable results. Finally, a MN is moving using the random way point mobility (RWP) model. MN speed varies from 2 m/s to 10 m/s. The simulation time is 1 hour. In order to compare the results with the existing MIIS implementations, the standard MIIS server (std. MIIS) is located inside the operator (core network side), since it is a network-centric deployment. In particular, three performance metrics are evaluated: (1) the average number of MIIS queries triggered by the MN, (2) the total MIIS query delay, and (3) the signaling overhead (in bytes) Average Number of MIIS Queries Triggered by the MN. From Figure 2, the effect of the MN speed on the number of MIIS queries can be observed. Higher velocity indicates that more PoAs are detected and discovered by the MN. Consequently, it crosses more MZs and triggers more MIIS queries. The results show that the std. MIIS has less performance because it triggers more MIIS queries than the proposed scheme. Also, it can be noted that the more PoAs per zone

94 International Journal of Distributed Sensor Networks 5 Table 2: Simulation parameters. MIIS 4 zones 16 zones 36 zones Queries Bytes Queries Bytes Queries Bytes Std. MIIS ZMIIS Average number of MIIS queries Total MIIS query delay (sec) MN speed ZMIIS (4 PoA per zone) ZMIIS (16 PoA per zone) ZMIIS (36 PoA per zone) Std. MIIS (4 PoA per zone) Std. MIIS (16 PoA per zone) Std. MIIS (36 PoA per zone) Figure 2: Average number of MIIS queries ZMIIS Hop count 4 2 Std. MIIS against ZMIIS Std. MIIS Wired one hop delay (ms) Figure 3: Total MIIS query delay. exist the fewer MIIS queries are generated using the ZMIIS server. Hence, most PoAs in the scenario will increase the average number of std. MIIS queries, causing overhead in the backbone Total MIIS Query Delay. In Figure 3,theeffectofdelayon the wired link and the effect of hop count between MN and MIIS server on the total MIIS query time are shown. The MIIS query time is the time from the instant the MN sends a query message, up to the time it receives the response message from any MIIS server. The total MIIS query delay is the delay for one MIIS query multiplied by the number of triggered queries during the simulation time. The hop count between the MN and the MIIS server varies from 2 to 10 and the wired delay varies from 5 ms to 25 ms. Itcanbeseenthat,aswireddelayincreases,theperformance of the std. MIIS degrades. Also, the MIIS query time using the std. MIIS is considerably affected with an increase of the hop count. The MN experiences a total MIIS query delay from 1.03 s up to 7.56 s using the ZMIIS server and 2.24 s up to s using the std. MIIS server, due to the fact that the ZMIIS servers are installed closer to the MN (at the access router) and the std. MIIS is located in the operators core network. The results clearly show that the zone-based MIIS architecture drastically reduces the neighbor discovery delay Signaling Overhead (in Bytes). We also evaluate the effect of the number of MZs in the signaling overhead. Each MIIS message carries about 40 bytes of length for the std. MIIS and 52 bytes for the ZMIIS (4 PoAs information). It is expected that, increasing the number of MZs, more queries are generated using the ZMIIS server, causing more overhead. However, Table 2 shows that the difference in the number of transferred bytes when there are 4 MZs (1.5 KB) and 36 MZs (1.67 KB) is minimal. Moreover, the ZMIIS architecture always presents a mean overhead reduction of almost37%inrelationtothestd.miis. 6. Conclusion We presented a zone-based MIIS architecture, in which the access networks are grouped into mobility zones, managed by different MIIS servers. The decentralized MIIS deployment provides higher resilience and scalability with regard to the mobility information distribution. The results show that the proposed scheme outperforms the std. MIIS in terms of discovery delay and signaling overhead. Future work includes the study of security mechanisms and interoperator service agreement models. Conflict of Interests The authors declare that there is no conflict of interests regarding the publication of this paper. Acknowledgments Part of the computations of this work was performed in EOLO,theHPCofClimateChangeoftheInternational Campus of Excellence of Moncloa, funded by MECD and MICINN. This is a contribution to CEI Moncloa. This work was supported by the PNPD/CAPES-Programa Nacional de Pós-Doutorado/CAPES program.

95 6 International Journal of Distributed Sensor Networks References [1] IEEE Standard, IEEE standard for local and metropolitan area networks part 21: media independent handover, IEEEStd ,January2009. [2] J.-Y. Baek, D.-J. Kim, Y.-J. Suh, E.-S. Hwang, and Y.-D. Chung, Network-initiated handover based on IEEE framework forqosservicecontinuityinumts/802.16enetworks, in Proceedings of the IEEE 67th Vehicular Technology Conference- Spring (VTC 08),pp ,Singapore,May2008. [3] J.-H. Seol and J.-M. Chung, IEEE MIH based handover for next generation mobile communication systems, in Proceedings of the 4th International Conference on Innovations in Information Technology (IIT 07), pp , Dubai, United Arab Emirates, November [4] E.Stevens-Navarro,Y.Lin,andV.W.S.Wong, AnMDP-based vertical handoff decision algorithm for heterogeneous wireless networks, IEEE Transactions on Vehicular Technology, vol. 57, no. 2, pp , [5] H. Liu, C. Maciocco, V. Kesavan, and A. L. Y. Low, IEEE assisted seamless and energy efficient handovers in mixed networks, in Proceedings of the 2nd International Conference Mobile Wireless Middleware, Operating Systems, and Applications, J.-M. Bonnin, C. Giannelli, and T. Magedanz, Eds., vol. 7ofLecture Notes of the Institute for Computer Sciences, Social Informatics and Telecommunications Engineering, pp.27 42, Springer, Berlin, Germany, [6] C. K. Christakos, A. Izquierdo, R. Rouil, and N. Golmie, Using the media independent information service to support mobile authentication in fast mobile IPv6, in Proceedings of the IEEE Wireless Communications and Networking Conference (WCNC 09), pp. 1 6, Budapest, Hungary, April [7] Q. B. Mussabbir, W. Yao, Z. Niu, and X. Fu, Optimized FMIPv6 using IEEE MIH services in vehicular networks, IEEE Transactions on Vehicular Technology, vol.56,no.6,pp , [8] S.-J. Yoo, D. Cypher, and N. Golmie, Timely effective handover mechanism in heterogeneous wireless networks, Wireless Personal Communications,vol.52,no.3,pp ,2010. [9] W.-S. Lim, D.-W. Kim, Y.-J. Suh, and J.-J. Won, Implementation and performance study of IEEE in integrated IEEE /802.16e networks, Computer Communications, vol.32, no.1,pp ,2009. [10] J. W. Floroiu, M. Corici, B.-J. Lee, S. Lee, S. Arbanowski, and T. Magedanz, A vertical handover architecture for end-to-end service optimization, in Proceedings of the 16th IST Mobile and Wireless Communications Summit, pp. 1 5, Budapest, Hungary, July [11] Y. Kim, S. Pack, C. G. Kang, and S. Park, An enhanced information server for seamless vertical handover in IEEE MIH networks, Computer Networks,vol.55,no.1,pp , [12] P. Neves, J. Soares, S. Sargento, H. Pires, and F. Fontes, Contextaware media independent information server for optimized seamless handover procedures, Computer Networks, vol. 55,no. 7, pp , [13] W. Ying, Z. Yun, Y. Jun, and Z. Ping, An enhanced media independent handover framework for heterogeneous networks, in Proceedings of the IEEE 67th Vehicular Technology Conference- Spring (VTC 08), pp , Singapore, May [14] K. de Vogeleer, S. Ickin, and D. Erman, A decentralized information service for IEEE media independent handover (MIH), IEEE Groups, [15]A.delaOliva,A.Banchs,I.Soto,T.Melia,andA.Vidal, An overview of IEEE : media-independent handover services, IEEE Wireless Communications,vol.15, no.4,pp , [16] G. Lampropoulos, A. K. Salkintzis, and N. Passas, Mediaindependent handover for seamless service provision in heterogeneous networks, IEEE Communications Magazine, vol. 46, no. 1, pp , [17]F.Buiati,L.J.G.Villalba,D.Corujo,S.Sargento,andR.L. Aguiar, IEEE information services deployment for heterogeneous mobile environments, IET Communications, vol. 5, no.18,pp ,2011. [18] NIST Mobility Package for Network Simulator-2, May 2007.

96 Hindawi Publishing Corporation International Journal of Distributed Sensor Networks Volume 2014, Article ID , 7 pages Research Article Implementation of Intelligent Electronic Acupuncture System Using Sensor Module You-Sik Hong, 1 Baek-Ki Kim, 2 and Bong-Hwa Hong 3 1 Department of Computer Science, Sangji University, Wonju , Republic of Korea 2 Department of Information & Telecommunication Engineering, Gangneung-Wonju National University, Wonju , Republic of Korea 3 Department of Digital Media Engineering, Kyung Hee Cyber University, Seoul , Republic of Korea Correspondence should be addressed to Baek-Ki Kim; bkkim@gwnu.ac.kr and Bong-Hwa Hong; bhhong@khcu.ac.kr Received 30 August 2013; Accepted 2 February 2014; Published 9 March 2014 Academic Editor: Young-Sik Jeong Copyright 2014 You-Sik Hong et al. This is an open access article distributed under the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited. Conventional electronic acupuncture can stimulate only one acupuncture point, and patients have to decide the time and the strength by themselves. In order to solve these problems, intelligent electronic acupuncture using biometric sensors and fuzzy technology was developed in this paper. And wireless electronic acupuncture system using sensor modules was developed in this paper. We used the sensor modules to obtain a patient s diagnosis signals. These sensor modules consist of 5 parts. The signals were analyzed to make instructions for the treatment, and the sensing pad for electronic acupuncture was designed. In addition, adaptive wireless acupuncture system was developed to adjust strength and time of acupuncture and several acupuncture points of patients by using fuzzy technology. We implemented efficient wireless electronic acupuncture system to get acupuncture easily using intelligent diagnosis system. 1. Introduction The electronic acupuncture is different from the traditional acupuncture in their shape and treatment method. But its basic principle of treatment is the same. More than 60 percent of the electronic acupunctures developed in the country use low frequency and the rest is developed using instantaneous electro stimulation. Existing low-frequency therapeutic apparatuses are simple frequency generator (16 32 Hz) which attaches electrodes to patient s diseased area. Patient cannot be treated effectively because it does not provide detailed frequency. Furthermore, it cannot find acupuncture points since it has no consideration of the patients sex, age, weight, illness, and so forth. And it causes a problem that some children and elderly people are bruised or wounded after getting electronic acupuncture due to inappropriate acupuncture time and strength [1]. Intelligent electronic acupuncture means that the acupuncture system can treat a patient automatically with acupuncture adapted voltage, current, and frequency. To perform this electronic acupuncture the system has function of sensing and treatment simultaneously. And the system requires an accurate analysis and processing technique of logical and statistical data using fuzzy [2, 3]. The pulse is considered an important factor in oriental medicine because a person s pulse rate may reflect his or her health condition. For example, if a patient s heart stops, it is a very serious situation and this situation can be judged by pulse. Oriental doctors have considered pulse rates as important data in diagnosis. But the existing blood pressure pulse analyzers have some problems. It is uncertain whether the blood pressure pulse analyzing sensor is located precisely on the radial artery and it is also difficult to diagnose pulse exactly depending on the thickness of forearm. Furthermore, the analogue type of blood pressure pulse analyzers has problems with quantification of the blood pressure pulse. Although some people may have the same forearm length but the thickness of their blood vessel may differ. Therefore there is no set of data that is considered reliable enough to judge the accuracy of blood pressure pulse rates. Oriental doctors should not only judge the basic biological signals suchasthepulse ssize,strength,andspeed,butshouldalso

97 2 International Journal of Distributed Sensor Networks Body signal parts Blood pressure sensing Skin conductivity sensing ECG signal sensing Oxygen saturation signal sensing Body temperature sensing Signal analyzing and treatment system Monitoring and DB generating Electronic acupuncture part Figure 1: Whole system diagram of the intelligent electronic acupuncture system. considerthebasicandquantitativeanalysisofthepulsein order to make an accurate diagnosis. Also, the doctors should consider physical characteristics, such as the thickness of theskinandbloodvessels,inordertoreachanaccurate conclusion. Therefore, measurement of the blood flow rate is a vital indicator in understanding the blood pressure rate and how the substances in the blood are transported [4 6]. The method of exiting diagnosis has a problem which cannot diagnose the old and the infirm exactly because the patient s condition including gender, age, skin is not taken into consideration. To solve this problem, we analyzed the fine distinction considering thickness of skin and blood vessels and pulse, whether they are big or small, strong or weak and fast or slow. We proposed the algorithm that diagnoses the condition of a patient optimally using intelligent fuzzy technique [7, 8]. Adaptive wireless acupuncture system was developed in this paper by using pulse diagnosis system to adjust strength and time of acupuncture and several acupuncture points of patients to whom intellectual fuzzy technology is applied. Conventional electronic acupuncture cannot find the acupuncture points at once. However, SW which can stimulate multiple acupuncture points and calculate the time of the electronic acupuncture was developed in this paper. Conventional electronic acupuncture only stimulates the acupuncture point, but the electronic acupuncture with KIT (SW + HW) developed in this paper made remote or self-diagnosis possible, using the conditions of the patients and disease reasoning function. Doctors help is needed to find the acupuncture point with conventional electronic acupuncture. Intelligent electronic acupuncture that easily calculates optimal acupuncture time considering the patients health condition with smart phones was developed in this paper. Figure 1 showsthewholesystemdiagramoftheintelligent electronic acupuncture system. It consists of 4 parts. The first partisasensormodule,thesecondpartisamainpartwhich analyzes the transferred signals and generates the treatment signals. The third part is an electronic acupuncture part which applies electronic acupuncture according to treatment signals

98 International Journal of Distributed Sensor Networks 3 from the main part. The last part is a program for monitoring and generating DB. The composition of this paper is as follows. Section 2 is about a sensor module for electronic acupuncture, and Section 3 is about intelligent pulse diagnosis algorithm. Section 4 deals with implementation of the electronic acupuncture system. Finally the conclusion is made in Section Sensor Module for Electronic Acupuncture We used several sensor modules to obtain a patient s diagnosis signals. These sensor modules consist of 5 parts, and they detect and analyze the abnormal signals from human body. Figure 2 shows the sensor modules for electronic acupuncture system. The measured signals from the each sensor of modules are transferred to main the part. (1) Pulsimeter module: It measures pulse rate. It measures the data from the finger connected to the finger sensor. (2) EGC module: It measures electrocardiogram. (3) SPO2 module: It measures oxygen saturation of blood. (4) Skin conduction module: It measures conductivity of palmar skin. (5) Body temperature module: It measures temperature of human body. 3. Intelligent Pulse Diagnosis Algorithm The intelligent pulse diagnosis system is composed of three parts. The first part is composed of the sensor to detect the conductance which is appropriate for injured part of human body and reference signal generator to adjust the signal generatedfromthepatients.thesecondpartiscomposedof DSP (Digital Signal Processor) board in which the signals are measured and classified using fuzzy algorithm. The last part is composed of a computer system that displays the signal from DSP board to the monitor and analysis software to diagnose the patients. Figure 3 shows the whole diagram of physical signal data network for electronic acupuncture. The algorithm consists of 3 parts. First step is sensing methods, the second step is indexing methods and the third step is classification methods. Pulse is beat-wave pattern of chest wall and great arteries accordingtoheartbeat.themainpurposeofpulseisto observe cardiomotility and blood movement. Recently study using physical characteristics shows that pulse wave pattern can change depending on condition of blood vessels and blood circulation. The pulse wave pattern can be obtained by second differentiation of digital plethysmogram using physical specific status such as uncertain inflection points. In this paper, we classified a patient s physical condition into three categories, as dangerous, ordinary and normal condition adapting pulse diagnosis algorithm using acceleration pulse wave pattern [9]. Fuzzy rules are generally presented with IF-THEN format. Fuzzy inference is procedures that infer new relations or facts from the given rules, and max-min reference is used. Input: x is A AND y is B R1:IFxis A1 AND y is B1,THENzis C1 OR R2:IFxis A2 AND y is B2,THENzis C2.. OR Rn:IFxis An AND y is Bn,THENzis Cn Conclusion: z is C Combination Function of Trust Value. 1 and 2 type of fuzzy creation rule reduced from type of 5 and 6 can come to the same node and conclusion through different inference path to infer fuzzy. In this node the same conclusion reached two of more different trust value. In this case combination function of trust value is used to recalculate trust value of conclusion [2, 8]. β c =β comb (β c,β old c )=max (β c,β old c ). (1) Here β old c is trust value of the conclusion reached through inference path already, β c is trust value of other conclusion reached through another inference path. If the 4 patients (a, b, c, d) illness condition is end-stage, the value is displayed as shownintheleft,incaseofthemiddlestagethevalue is and in case of the first stage the value is displayed as The value in the middle shows patient s physical condition. For example, if the patient s height is 150 cm and weight is lower than 45 kg the value is displayed as When the patient s height is between 151 and 170 cm and the weight is between 46 kg and 70 kg, the value is displayed as , and when the height is 171 cm 200 cm and weight is 71 kg 130 kg, the value is displayed as In Figure 3 the process to calculate fuzzy correction factor according to patient s physical condition is shown. 4. Implementation of the Electronic Acupuncture System Electronic acupuncture system with built in multi pad which can find out the condition of the patients automatically and treat the patients simultaneously. The system includes the function that can treat the patients with acupuncture and adjust voltage, current, frequency oscillation automatically according to their physical conditions. To perform the function, the system senses and treats acupuncture simultaneously, and requires logical and statistical data processing technique using fuzzy and exact analysis. Installing the 5 round pads underneath the palm, we can change the signal, andthenadaptiveacupuncturetreatmentcanbegiven. At this point, measurement of the signal uses the wireless type instead of cable type. Because the wireless type has advantage of convenience to get acupuncture, reduction of noise by using cable connected to a computer system and prevention of electric shock depending on abrupt hightension electricity [10, 11].

99 4 International Journal of Distributed Sensor Networks Pulse sensor Blood pressure and sugar sensor Hmote 2420 ECG sensor Infrared temperature (a) Sensor module parts considering patient s physical conditions Blood pressure sensor REF PGA AMP CC bit A/D USB to serial BAT and DC/DC Photo diode LED PGA AMP AMP A/D D/A CC2430 USB to serial BAT and DC/DC EGC #1 EGC #2 EGC #3 Notch filter CC bit A/D USB to serial BAT and DC/DC Blood pressure #1. #2 PGA AMP MOS FET bit A/D TR array PIC18F85 USB to serial BAT and DC/DC TFT LCD UART SD slop 2 G DC/DC SSPV210 (cortex-a8) SDRAM 512 MB Nand flash 256 MB CPLD decoder Audio ALC5622 USB host 2.0 ZigBee Ethernet 10 M WiFi Bluetooth (b) Block diagram of Sensor modules Figure 2: Sensor modules for electronic acupuncture system. In order to treat acupuncture, it is important not only to get information from the human body but also to learn ages, sexes, height and weight of the patients. To do this, control variables using fuzzy algorithm are made before treatment of acupuncture. Figure 4 shows Circuit of the acupuncture signal. The part of sensing pad and contact point of the fingertip made of stripe array type to distribute contact point area evenlyafterbeingplatedwithgoldtoreduceelectricresistance.

100 1 International Journal of Distributed Sensor Networks Mhz Hmoto2420: sender RF channel: 11 Group ID:0 05 Hmoto2420: receiver 2410 Mhz Hmoto2420: sender RF channel: 12 Group ID: 0 01 Hmoto2420: receiver Hmoto2420: sender Hmoto2420: sender RF channel: 11 Group ID: 0 01 RF channel: 11 Group ID: 0 03 Hmoto2420: receiver 2415 Mhz Hmoto2420: receiver Hmoto2420: sender RF channel: 13 Group ID: 0 01 Hmoto2420: receiver Figure 3: Whole diagram of physical signal data network for electronic acupuncture. Data A Data B GND +5 V C4 10 uf/25 V + C8 10 uf/25 V + C5 10 uf/25 V C10 10 uf/25 V R1IN R2IN T1IN T2IN C+ C1 C2+ C2 V+ V GND +5 V 16 VCC 15 R1OUT R2OUT T1OUT T2OUT U3 MAX232 TX RX C6 18P(CH) C1 104 X-TAL1 +5 V m C7 18P(CH) + C9 10 uf/25 V R2 8.2 K U1 AT89C2051 P3.0/RXD P1.0/AIN0 P3.1/TXD P1.1/AIN1 P3.2/INTO P1.2 P3.3/INT1 P1.3 P3.4/T0 P1.4 P3.5/T1 P1.5 P3.7 P1.6 P1.7 XTAL1 XTAL2 RST/VPP VCC 10 GND D LS1 Buzzer 1N PWM-1 PWM-2 LOAD A_DIR-1 A_DIR-2 B_DIR-1 B_DIR-2 C_DIR-1 C_DIR-2 D_DIR-1 D_DIR-2 +5 V + 2 Q1 C101 1 C3 1 uf/16 V +5 V + R1 220 D1 1N4007 U2 2 3 VOUT VIN ADJ C2 104 VCC C A_DIR-1 A_DIR-2 PWM C17 22 uf/16 V U4 VIN VOUT LIM +SENSE REF CUR.LIM L6203 +CUR.LIM BOOST DIM +BOOST +VIN C C R3 1 K Output CH1 Output CH1 C R5 R6 R7 10 K 1 K 20 K 2 3 VCC U5 6 R8 20 K C R4 2 K C Load C Output module-ch1 Figure 4: Circuit of the acupuncture signal. In this paper, we designed the optimal algorithm which couldjudgetheremotemedicaldiagnosisusingfuzzylogic andfuzzyinferencerules,andwesimulatedtheprocessto calculate the optimal acupuncture time of the body condition ofpatients.weproducedthewirelesscommunicationpart to transmit condition of patients pulse, skin conductance and oxygen saturation data to user s terminal or remote medical terminal, and to receive the control signal from user s terminal or remote medical terminal. To do this, we made the sensing pad, the circuit of AMP andacupuncturesignal,wirelesscommunicationmoduleand charging circuit for storage battery. And also we proposed the software including algorithm of analysis and control using fuzzy technique. Existing acupuncture system using DSP has a complex structure, uses up a lot of electricity and it s big and expensive. But the adaptive wireless acupuncture system proposed in this paper is simple, inexpensive and safe. Figure 5 shows simulation of the glove type electronic acupuncture. To implement wireless system, we used the way of RF data modem for wireless communication using Narrowband FSK. The feature of this way is robust to noise and it can

101 6 International Journal of Distributed Sensor Networks Figure 7: Transmit/receive system for ubiquitous network. Figure 5: Simulation of the glove type electronic acupuncture. Figure 8: Analysis of electro stimulation to fingertips. Figure 6: Data transmitter and receiver using RF communication. Figure 9: Output of electronic acupuncture needle time simulation using FIS matlab. transmit data easily by simple communication protocol. And this system is adapt to design multi type data communication system and can be designed by low power, one 3 V battery, in case of short distance. We considered not only resistance measurement but capacitive component to reduce error depending on several conditions of human body. To do this, we applied the pulse wave DC 50 V 200 V, 500 ua 1,500 ua, intermittent stimulation of 5 Hz 5 KHz to the main pad and fingertip and measured the voltage peak and phase frequency [12, 13]. We used 470 MHz band frequency and designed the system to change 21 physical frequency. And logical address of a channel corresponding to each adaptive acupuncture was assigned using polling technique and then called. The system supports half duplex communication. This way is suitable for the system because the system requires low data and uses relatively low speed communication. The output power of wireless signal using button type battery is 1 mw, and it is adequate to transmit data without noise. The speed of transmission is bps and wireless encoding uses a way of Biphase Manchester code. Communication between notebook computer and wireless modem uses RS232C. Figure 6 shows the data transmitter and receiver using RF communication. For remote medical treatment, the transmitter acquires data from 4 sensors, and then transmit the data to receiver using RF communication. In Figure 7 the system consists of transmit and receive system parts for ubiquitous network. It is made of MSP240CPU and CC2420 RF chip. Figure 8 shows analysis of electro stimulation to fingertips using pads. To obtain signal, we send a reference signal to palm, and then decide body condition of patients on the basis of data obtained from pre-investigation using sensing pads and MCU attached to fingertips. As soon as signal processing is completed, electric stimulation signal generated by fuzzy algorithm is transmitted to sensing pads. Table 1 explains fuzzy inference of a variety of patients with the same disease according to varying blood pressure condition, Heart rate condition, and vascular aging condition. In other words, Table 1 clearly shows that the system calculate varying time of acupuncture for different patients physical conditions. Figure 9 shows the Output of electronic acupuncture needle time simulation using Fuzzy Inference System Matlab. It explains how the system calculates the output condition

102 International Journal of Distributed Sensor Networks 7 Table 1: Electronic acupuncture needle time simulation. Patient biometric information Optimal acupuncture needle time Input data (minutes) Blood pressure condition Heart rate condition Vascular aging condition Conventional Intelligence Medium Medium Small Big Big Big Big Big Medium Medium Medium Medium Medium Big Big Medium Medium Small Small Big Big Small Medium Big Small Small Small of the time for acupuncture from the input data of the 3 conditions of patient physical conditions. 5. Conclusion In this paper, we implemented intelligent electronic acupuncture system using sensor modules. We used the sensor modules to obtain a patient s diagnosis signals. These sensor modules consist of 5 parts. These sensor modules detect and analyze the abnormal signals from human body. We analyzed the signals to make instructions for the treatment. And then we designed the sensing pads for electronic acupuncture. And we also developed adaptive wireless acupuncture system to adjuststrengthandtimeofacupunctureandseveralacupuncture points of patients by using fuzzy technology. We made the sensing pads, the circuit of AMP and acupuncture signal. We implemented efficient electronic acupuncture system to get acupuncture easily using intelligent diagnosis system. The intelligent acupuncture system proposed in this paper is simple, inexpensive and safe compared with conventional acupuncture systems. Conflict of Interests The authors declare that there is no conflict of interests regarding the publication of this paper. Acknowledgment This research was supported by the MSIP (Ministry of Science, ICT and Future Planning), Korea, under the IT/SW Creative research program supervised by the NIPA (National IT Industry Promotion Agency) (NIPA-2013-H ). [2] H. K. Baruah, The theory of fuzzy sets: beliefs and realities, International Journal of Energy, Information and Communications,vol.2,no.2,pp.1 22,2011. [3] J. Jeong, The development of web-based decision tree program for the analysis of clinical information, ideas constitution, Korea Institute of Oriental Medicine,vol.12,pp.81 87,2008. [4] Y.J.Lee,J.Lee,H.J.Lee,H.H.Yoo,E.J.Choi,andJ.Y.Kim, Study on the characteristics of blood vessel pulse area using ultrasonic, Korea Institute of Oriental Medicine Researches,vol. 13, no. 3, pp , [5]P.A.Shaltis,A.T.Reisner,andH.H.Asada, Cufflessblood pressure monitoring using hydrostatic pressure changes, IEEE Transactions on Biomedical Engineering, vol. 55, pp , [6] National College of Oriental Medicine, Diagnostics, Saint Functionality of Medicine, St Functional Medicine, 2008, Gunja Publisher, [7] Department of Medical Sciences, College of Oriental Medicine, Kyung Hee University, Medical practice materials, [8] O. P. Verma and S. Singh, A fuzzy impulse noise filter based on boundary discriminative noise detection, Journal of Information Processing System,vol.9,no.1,2013. [9] S.-S. Lee, M.-C. An, and S.-H. Ahn, A new measurement method of a radial pulse wave using multiple hall array devices, Journal of Magnetics,vol.14,no.3,pp ,2009. [10] S. Haykin, Modem Wireless Communication, Prentice-Hall, [11] A. Swami and H. Ya, Wireless Sensor Networks: Signal Processing and Communications,JohnWiley&Sons,2007. [12] J. K.-Y. Ng, Ubiquitous healthcare: healthcare systems and applications enabled by mobile and wireless, Journal of Convergence,vol.3,no.2,2012. [13]A.SinhaandD.K.Lobiyal, Performanceevaluationofdata aggregation for cluster-based wireless sensor network, Human- Centric Computing and Information Sciences, vol.3,article13, References [1] Y.S.Hong,H.K.Kim,andB.K.Kim, Implementationofadaptive electronic acupuncture system using intelligent diagnosis system, International Journal of Control and Automation, vol. 5, no. 3, pp. 141 l52, 2012.

103 Hindawi Publishing Corporation International Journal of Distributed Sensor Networks Volume 2014, Article ID , 12 pages Research Article Node Placement Analysis for Overlay Networks in IoT Applications Yuxin Wan, 1 Junwei Cao, 1 Kang He, 1 Huaying Zhang, 2 Peng Yu, 2 Senjing Yao, 2 and Keqin Li 3 1 Department of Automation, Research Institute of Information Technology, Tsinghua National Laboratory for Information Science and Technology, Tsinghua University, Beijing , China 2 Shenzhen Power Supply Co. Ltd., China Southern Power Grid, Shenzhen , China 3 Department of Computer Science, State University of New York, New Paltz, NY 12561, USA Correspondence should be addressed to Junwei Cao; jcao@tsinghua.edu.cn Received 11 October 2013; Revised 22 January 2014; Accepted 22 January 2014; Published 6 March 2014 Academic Editor: Luis Javier Garcia Villalba Copyright 2014 Yuxin Wan et al. This is an open access article distributed under the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited. The Internet of Things (IoT), which combines identification, sensing, computing, and communication technologies, is considered one of the major trends in information and communication technologies. Communication performance is critical for IoT applications. According to previous research, an internet-based overlay model is feasible for the implementation of the IoT. One important issue in the overlay routing model is the overlay node placement problem (ONPP). Once the size of overlay node set is fixed to a particular number k, the ONPP changes to k-onpp. In this work, the IoT-based overlay node placement problem is formulized and analyzed. The major contributions of the paper include providing the time complexity of multi hop k-onpp and its theoretical limit boundary of approximation ratio and proposing a local search algorithm. Furthermore, the time complexity and approximation ratio boundary of the local search algorithm are given. The proposed local search algorithm is evaluated by both time and efficiency where efficiency refers to the degree of approximation of algorithm results with optimal solutions. Another algorithm, TAG, is used for comparison. Finally, a simulation experiment based on network simulator EstiNet is provided. The experimental results show network delay benefits from the proposed method. 1. Introduction The Internet of Things (IoT) has been regarded as the future of internet and one of the major trends in information and communication technologies [1]. The key idea of IoT is combining identification, sensing, computing, and communication technologies to provide a better description of physical processes. IoT technologies can be applied in a wide variety of applications such as smart homes, smart cities, environmental monitoring, and health care [2]. Many IoT-based applications require timely interaction between users and physical objects. Therefore, communication performance is very important in IoT implementation. There are three options for the implementation of the IoT: using the current internet, building a new network, and buildinga dual-layer network [3]. Based on the consideration of both performance and ease of implementation, an internet based dual-layer network is suitable for the IoT. Here, the dual-layer network refers to the overlay network. Currently, many IoT applications are implemented using an overlay network. Take the smart grid, for example. One typical example of a smart grid is the wide area management system (WAMS). The WAMS uses the phasor measurement unit (PMU) as sensor and data collector. The collected data need to be transferred to a control center for analysis. The current WAMS is built on an IP-based network, and many studieshavebeenconductedontheinfluenceofnetwork performance on WAMS [4 6]. However, as the internet only provides a best-effort service, internet-based overlay networks should add additional methods to improve network performance. Such methods

104 2 International Journal of Distributed Sensor Networks include admission control and overlay routing. Admission control guarantees the worst-case delay boundary, but it may deny a connection [7] and requires special network devices. Overlay routing has been proved useful in reducing end-toend delay [8], and no further devices are needed. The overlay routing method can be used to reduce the communication delay between sensors and the data center where the data areanalyzed.oneimportantissueintheoverlayrouting model is the overlay node placement problem (ONPP) [9]. TheobjectiveoftheONPPistofindtheoptimaloverlaynode set with minimum total data transfer cost. However, the size of overlay node set may be fixed to a given number k due to cost and efficiency considerations. This modified ONPP is called k-onpp. In this work, the overlay node placement problem (ONPP) in IoT applications is formulized and a local search algorithm is proposed. The time complexity of k-onpp is analyzed. Furthermore, we give the theoretical limit boundary of the approximation ratio for k-onpp. Additionally, the approximation ratio boundary of the proposed local search algorithm is provided. A genetic algorithm and a greedy algorithm are introduced for performance comparison. All algorithms are evaluated by time cost and efficiency with MATLAB tools. Here, efficiency refers to the degree of approximation of algorithm results with optimal solutions. Finally, a simulation experiment based on the network simulator EstiNet [10] is provided to test the efficiency of theproposedoverlaynetworkmodelandalgorithm.the experimental results show network delay benefits from the proposed method. The rest of the paper is organized as follows. Section 2 introduces the research background and related work in overlay node placement. Section 3 presents the model and formulation of k-onpp in the IoT and provides a theoretical analysis for this problem. Section 4 proposes a local search algorithm and provides its time complexity and approximation ratio boundary. Section 5 evaluates the algorithms based on time cost and algorithm approximation using MATLAB tools. A genetic algorithm and a greedy algorithm, TAG, are introduced for comparison. The local search algorithm is tested in a simulation scenario with the network simulator EstiNet in Section 6. Some additional factors that impact the algorithm are discussed. Finally, a general conclusion is provided in Section Research Background and Related Work Overlay routing has been proved to be a feasible method to improve network performance with unreliable internet infrastructure [8]. The basic concept of overlay routing is choosing one or more nodes in the overlay network as hop nodes for data transfer. Overlay routing will use an additional routing algorithm, separate from the underlying internet routing algorithm. There are two types of overlay networks: peer-to-peer networks and infrastructure networks [9]. Network nodes of peer-to-peer networks may change rapidly, while the nodes of infrastructure networks have higher persistence. In most infrastructure overlay networks, overlay nodes belong to a single entity, so it is feasible to apply the routing algorithm in these overlay nodes. Currently, most networks of IoT applications are similar to infrastructure overlay networks, so this paper focuses on the overlay node placement problem in such networks. Previous work in [8] has proved that with the overlay routing method, the RTT of end-to-end packet may be reduced. Andersen et al. found that network performance would improve with only one hop node [11]. A random-k algorithm is proposed in [12]. The basic idea of this algorithm is randomly selecting k nodes out of M. Then,asource sends a test packet through these k nodes to the destination. The intermediate node whose response comes back first will be chosen as the overlay node. This algorithm aims to improve the network reliability after path failure occurs, so other performance metrics such as communication delay are not considered. Additionally, this algorithm is based on experimental experience, no theoretical analysis is given. In [13], Zhu et al. further studied the node placement problem with one hop node. Their scenario uses overlay routing and multihoming to improve network performance and availability. They proved the overlay node placement problem with one hop node which is an NP-hard problem based on a reduction from the set covering problem. The number of overlay nodes is not fixed in their scenario. In contrast to previous work in [12], network performance is considered in [13]. Four heuristic methods, Random, Customer-driven, Traffic-driven, and Performance-driven, are introduced and tested. Their results show an improved RTT from sources to destinations with overlay routing. A measurement study on the benefits from overlay routing is made in[14]. Their scenario uses overlay routing to improve end-to-end network performance. The intermediate nodeisalsosettobeone.incontrasttopreviousworkin[11 13], the number of overlay nodes is fixed to a given number k in this work. The problem is also proved to be an NPhard one. Four greedy heuristic algorithms are introduced and tested. A generic description of k-onpp is given in [9] asfollows. Given M possible overlay nodes and a number k,choose k nodes out of M to optimize the application-specialized performance metric. Moreover, the number of intermediate nodes in an overlay path is unfixed. In [15], Yang et al. considered this generalized problem with the performance metric group delays from sources to destinations. In this work, the node placement problem is described by a linear programming formula and solved with ILOG CPLEX, an optimization software package designed by IBM. Although in this work the generalized problem is considered, it is unfeasible to implement the solution in a real network as the results are calculated by another software. The latest work on overlay node placement problem can be found in [16 18]. In [16], Roy et al. introduced a greedy algorithm called Traffic Aware Greedy (TAG) and compared this algorithm with the node degree-based algorithm. Another greedy algorithm is proposed in [17]. Cohen and Raz made progress on this problem by providing a theoretical limit boundary of the approximation ratio that canbeachievedbyapolynomialalgorithmbasedonthe

105 International Journal of Distributed Sensor Networks 3 Application layer IOT network layer Sensor/actuator layer Data analysis Dual layer network Current internet Different sensors/actuators Energy Traffic management management Smart grid sensor/actuator Distributed IoT data server Intelligent transportation sensor/actuator Meteorological disaster early-warning Environment monitor sensor Figure 1: IoT architectural scheme. set cover problem [18]. The overlay node placement problem also has been discussed in other contexts such as web cache placement in [19, 20], but the motivation and objective are quite different. As described above, most of the current works on the overlay node placement algorithm are greedy algorithms and lack a theoretical analysis. Although [18]analyzedthisproblem in a theoretical manner and provided a limit boundary of approximation ratio, their analysis is based on a onehop overlay. In addition, the approximation ratio of their proposed algorithm is denoted by another parameter m, where m is the size of the maximum minimal overlay vertex cut.as written in [18], finding the minimal overlay vertex cut is not easy. In this work, the overlay node placement model for IoT applications is proposed, and a different analysis approach is made based on a reduction from the k-median problem. We further analysis the k-onpp problem with a multihop overlay and provide its time complexity and the limit boundary of the approximation ratio. A local search algorithm is proposed, and the approximation ratio boundary has been provided in a more calculable way. 3. Problem Formulation and Analysis According to previous research in IoT [21, 22], the architectural of IoT can be divided into three parts: sensor/actuator layer; information transmission layer (network layer), and application layer. As mentioned in the introduction, currently the internet based dual-layer network is feasible for an IoT application, so an abstract IoT architectural scheme can be described as in Figure 1. Consider the above IoT architectural scheme, the communication network of IoT is used for data gathering from distributed sensors to analysis centers. Some data concentrators, such as PMU in smart grid, are deployed so that an internet-based dual layer network is already there. These concentrators and sensors are fully constructed and controlled by the same entity or group, so they can be used as an overlay node to gain benefits. As these concentrators are almost persistent, this overlay network can be regarded as an infrastructure overlay network. Thus, the problem is how to place these concentrators to maximize the benefits. Figure 2 inthefollowingpageprovidesasketchmapofsuchadual layer network Problem Formulation. We consider the performance metric of group communication delay, which is the total communication delay from each sensor to the analysis center. Group communication delay is used because data generated by sensors in IoT application is predictable and generally periodic. This means that if group communication delay drops, system performance may become better. Then, the overlay node placement problem can be formulized as follows. Consider a physical network represented as a graph G(V, E), wherev denotes the networking devices and E denotes links between V.Theweightoflinkein E is defined by a metric such as network bandwidth or communication delay, denoted by l ij >0,wherei, j indicates the vertexes of link e. We use communication delay as metric. A group of source vertexes denoted by S needs to send data to destination vertexes denoted by T. A candidate set of vertexes B V may be suitable locations to deploy concentrators. Let O B denote the chosen overlay node set. The destination of each s Sis fixed to t T. We define the function t(s), which denotes the t that s is connected to. Once the overlay node set O is chosen, each of the source vertexes s S can use indicate the weight of the direct path from vertex s to t(s) and l s,t(s) O indicate the weight with overlay nodes. l s,t(s) O is the weight of shortest path with overlay set O. Suppose that the shortest overlay path from s to t(s) is (s, o 1,o 2,...,o k, t(s)) then l s,t(s) O =l so1 +l o1 o 2 + +l ok t(s). If overlay set O is not helpful to reduce the original weight, O to transfer data. Let l s,t(s) p then s will link directly to t(s).then,l s,t(s) O l t(s),t(s) p =l t(s),t(s) O =0;then,l s,t(s) O. We define can be defined as follows: =l s,t(s) p l s,t(s) O = min o {O+{t(s)}} (ls,o p +lo,t(s) O ). ( ) Clearly, l s,t(s) O l s,t(s) p. This is different from other discussions, as in others each source must connect to one overlay node. Then, the objective function can be written as finding O B to minimize s S l s,t(s) O,wheret(s) is defined above and denotes the destination t Ttowhich s connects. Considering the cost of deploying these overlay nodes and the cost of maintaining communication delay information between the overlay nodes, the size of set O should be limited. Suppose that a given number k is used. We define this problem as the

106 4 International Journal of Distributed Sensor Networks Data destination Underlay internet channel Data destination Underlay internet channel Data relay node Data relay node Data relay node Underlay internet channel Underlay internet channel Underlay internet channel Data generating node Data generating node Data generating node Data generating node Data generating node Data generating node Figure 2: Sketch map of a dual layer network of the IoT. k-onpp problem. Then, the problem is modified to finding O Btominimize s S l s,t(s) O,andthesizeofOis k Problem Analysis. In this section, we will analysis the time complexity for k-onpp and discuss the theoretical limit boundary of approximation ratio. We give the following theorems. Theorem 1. k-onpp is an NP-hard problem. Proof. First, we consider another NP-complete problem. The k-median decision problem is a typical NP-complete problem which can be described as follows. Given a client set C and a candidate position set B,theweightfromeach c C to b B is denoted by l c b >0. Determine whether there exists a set O out of B where the size of O is k such that c C, o O min ( l c o ) U. We define this problem as P 1. Then, we consider a modified problem from k-onpp as follows. Consider a graph G(V, E), asourcesets, a destination set T, andapossiblesetb. Find + lo,t(s) p ), where the size of O is k. DefinethisproblemasP 2.Thisobjective function means that at most only one overlay hop can be used in an overlay path. Consider the decision problem P 2, determining whether there exists a set O out of B where the size of O is k such that O B to minimize s S,o O min(l s,t(s) p s S,o O,l s,o p (1) min (l s,t(s) p,l s,o p +lo,t(s) p ) U. (2) We define this problem as P 3. Next, we modify problem P 3 into a different version. Let B =B, C =S,and k =k. Consider a client point c C and a candidate overlay node b B.Define l c, b = min (l s,t(s) p,l s,b p +lb,t(s) p ). (3) Then, problem P 2 changes to determining whether there exists a set O out of B where Size( O) = k such that c C, o O min( l c o ) U. It is obvious that modified problem P 3 isthesameasproblemp 1.BecauseP 1 is NP-complete, then P 3 is NP-complete as well. Additionally, this proves that problem P 2 is the same as the k-median problem. Now, we consider the original k-onpp. Given a graph G(V, E), a source set S, a destination set T,andapossibleset B, findo B to minimize s S l s,t(s),wherethesizeofo is k. Suppose there is a polynomial algorithm for k-onpp. Construct a special case of k-onpp. Let l b 1,b 2 p O =T l s,t(s) p, where b 1,b 2 Band s S.Obviously,inthisconstructedk- ONPP,onlyonehopnodeatmostmaybeusedintheoverlay path. If there is a polynomial algorithm for k-onpp, then the constructed k-onpp can be solved, then problem P 3 can be solved. The algorithm for P 3 can be designed as follows. (1) Use the polynomial algorithm for k-onpp to find the result R of constructed k-onpp. (2) Test if R U. Because P 3 k-onpp and P 3 is NP-complete, k-onpp is an NP-hard problem. This proves Theorem 1. Next, we provide the theoretical limit boundary of an approximation ratio for k-onpp. In k-onpp, define following parameters: d max = max {min (l s,t(s) p,l s,b p +lb,t(s) p ) t(s) T,s S,b B}, d min = min {min (l s,t(s) p,l s,b p +lb,t(s) p ) α=max { { { l s,b p l s,b p +lb,t(s) B +lp b,t(s) t(s) T,s S,b B}, ω= d max d min, t(s) T,s S,b B } }. } Theorem 2. There is no polynomial algorithm for k-onpp with an approximation ratio less than α (1 + ((ω 1)/e)). Proof. As proved above, problem P 2 isthesameask-median problem. With the above-defined d max and d min, ω also denotes max l c, b/ min l c, b, where b B in problem P 1. (4)

107 International Journal of Distributed Sensor Networks 5 Algorithm: LocalSearchAlgorithm Input: Candidate set B; Cost function Cost(N);Neighborhood structure F(N); Delay graph G(V, E) Output: Sub-optimal overlay node set O (1) Random select a set N which Size(N) = k (2) Constructing a new graph G (V,E ) with V = {N, {t(s)}} (3) Apply Dijkstra algorithm in G to get the shortest path from N to {t(s)} (4) Calculating the Cost(N) = s S min n N (l s,n p +ln,t(s) N ) (5) If N F(N) that Cost(N ) < Cost(N) then N=N, return to Step 1 (6) Return N. Algorithm1:Proposedlocalsearchalgorithm. Pan et al. proved that with a so-defined ω, thereareno polynomial algorithms with an approximation ratio less than 1 + ((ω 1)/e) unless NP DTIME(n O(log log n) ) for a general distance space k-median problem [23]. Let the optimal result for P 2 be O 1 and the optimal result for k-onpp be O 2.Letthe best result which can be obtained with polynomial algorithm for P 2 be R 1 and the best result which can be obtained with a polynomial algorithm for k-onpp be R 2.Obviously,wehave O 2 O 1, R 2 R 1.BecauseR 1 /O 1 (1 + ((ω 1)/e)), R 2 /O 2 R 2 /O 1 R 2 /R 1 R 1 /O 1. However, for each overlay path for client s S,min(l s,t(s) p,l s,o p +lo,t(s) O ) min(ls,t(s) p,l s,b p + l b,t(s) B ) α min(l s,t(s) p,l s,b p +lb,t(s) p ). Therefore, R 2 /R 1 αand R 2 /O 2 α (1 + ((ω 1)/e)).ThisprovesTheorem 2. Both α and ω are easy to calculate in this formula. It is obvious that the time complexity to obtain ω is O(Size(S) Size(B)). The time complexity to obtain α is the time complexity of shortest path algorithm. Because l b,t(s) B means using whole candidate set B as overlay node set, the shortest path algorithm can be applied. 4. Proposed Algorithm and Analysis As discussed above, the k-onpp problem is similar to the k-median problem, so the algorithm for the k-median problem may also be applied in k-onpp. The proposed local search algorithm is modified from the local search algorithm developed by Arya et al. in [24]. However, the discussion in [24] is based on the metric space, which means that the distance definition satisfies the symmetrical characteristic and triangle inequality. However, neither of these two properties is consistent with network delay. In fact, if network delay satisfies triangle inequality, there is no need ls,a p +lat p.themodifiedlocal search algorithm works as follows. We define the cost function as Cost(N) with given set N, which indicates the group communication delay from the client set S to destination T with a given overlay set N. A neighborhood structure for the set N is defined as F(N) = {N n + m n N,m B,m N}. We define a local optimum as Cost(N) < Cost(N ) for all N F(N). Then, to optimize the delay as l s,t p the steps of proposed local search algorithm are described in Algorithm 1. Next, we discuss the time complexity of the proposed algorithm. We state the following theorem. Theorem 3. The time complexity of the proposed local search algorithm is polynomial. Proof. Let Size(B) = M, where p indicates the number of iterations. Suppose that k N. The time cost for the Dijkstra algorithm is O(k 2 ). The maximum replacement in each iteration for set N is Size(B-N). So, the total time complexity for local search algorithm is O(k 2 Mp).Asdiscussedin[24], p can be defined as p=log(cost(s 0 )/Cost(O))/ log(1/(1 ε/q)). Here,S 0 is the initial value, O is the optimal result, ε>0is constant, and Q isthesizeofθ G(S),whereS is thesetofallfeasiblesolutionsandg(s) is the neighborhood of S. With proposed local search algorithm G(S) = F(S) = {S n+m n S, m B, m S}; therefore, Q Size(S) Size(B S),whichispolynomialbecauseQ,log(Cost(S 0 )) and log(cost(o)) are polynomial with the input size. So, the time complexity of the local search algorithm is polynomial. This proves Theorem 3. Finally, we provided the approximation ratio boundary of the local search algorithmin Theorem 4. Theorem 4. The approximation ratio boundary of the proposed local search algorithm is ω/α. Proof. Suppose that the optimal set for k-onpp is O and local optimal set is N. LetO = (o 1,o 2,...,o k ) and N = (n 1,n 2,...,n k ). d max and d min are defined the same as before, ω = d max /d min.asn isthelocaloptimalset,thenforall N F(N),weobtain Cost (N ) Cost (N) >0. (5) From inequality (5), we replace the overlay node n 1 in N with node o 1 in O,thenweobtain Cost (N n 1 +o 1 ) Cost (N) >0. (6) Define N 1 =(o 1,n 2,...,n k ), N 2 =(n 1,o 2,...,n k ),..., N k = (n 1,n 2,...,o k ).DefineD C (o 1 ) as the clients which connect o 1

108 6 International Journal of Distributed Sensor Networks as first overlay node. For N 1,letc D C (o 1 ) connect to o 1 as first overlay node, and let c C D C (o 1 ) connect to n {N 1 +{t(c )}} with minimum min(l c,n p +l n,t(c ) N 1 ). Therefore, inequality (6) can be expanded as follows: Cost (N n 1 +o 1 ) Cost (N) = (l c,o 1 p +l o 1,t(c) N 1 l c,t(c) N )+ (l c,t(c) N 1 l c,t(c) N ) c D C(o 1) c C D C(o 1) >0. For the first portion before the plus sign in (7), l c,o 1 p d min 1 α 1 α min (ls,t(s) p,l s,b p +lb,t(s) B ) min (lc,t(c) p,l c,o p +lo,t(c) O )= 1 α lc,t(c) O +l o 1,t(c) N 1 l c,o 1 p +l o 1,t(c) p ωd min ω α lc,t(c) O. For the second portion in (7), l c,t(c) N 1 l c,t(c) p (7) (8) ωd min ω α lc,t(c) O. (9) So, inequality (7) can be expanded as follows: 0< (l c,o 1 p +l o 1,t(c) N 1 l c,t(c) N )+ (l c,t(c) N 1 l c,t(c) N ) c D C(o 1) c C D C(o 1) ( ω α lc,t(c) O l c,t(c) N )+ ( ω α lc,t(c) O l c,t(c) N ) c D C(o 1) c C D C(o 1) = ( ω c C α lc,t(c) O l c,t(c) N )=ω α Cost (O) Cost (S). (10) Then, we have obtained the approximation ratio with defined ω and α as follows: This proves Theorem 4. Cost (S) < ω Cost (O). α ( ) 5. Algorithm Evaluation In this paper, the proposed local search algorithm is tested with both Matlab and the network simulator EstiNet. EstiNet is used to test the algorithm s performance in a network environment. However, as the amount of network nodes is limited in a simulator, in this section, Matlab is used to evaluate the algorithm based on time cost and effectiveness. A genetic algorithm is introduced to approach the optimal result, while the TAG algorithm proposed in [16] isusedfor comparison Experiment Design and Implementation. As described above, a physical network can be represented as a graph G(V, E),whereV denotes networking devices and E denotes the time delay between V. In the actual network, networking devices are connected in two modes: directly connecting with links and indirectly connecting with routers or switchers. So, the first step of experiment is generating an M M matrix to record the graph, where M is the number of network devices. After that, direct connections with time delay between graph vertexes are randomly generated. In the third step, each pair of two vertexes in the graph is connected through the directly connecting vertexes. The time delay between indirectly connecting vertexes is the sum of the time delay between directly connecting vertexes in the path. The Matlab test is implemented on a laptop with two Intel i5 core processors and 3 GB memory. For each experiment, we test the algorithm 500 times. The mean value of the proposed algorithm is then calculated to better expose the performance. In addition, the 95% quartile of 500 tests is calculated and the 95% confidence interval of the mean value is obtained. For the genetic algorithm, the best result of the 500 tests will be recorded, as it is used to generate the optimal result Two Other Algorithms for Comparison Genetic Algorithm. To evaluate the time cost and algorithm approximation, the global optimal result is needed for comparison. However, as previously proved, k-onpp is NP-hard; we cannot calculate the global optimal solution with an increasing problem scale. So, the result of a genetic algorithm is used to approximate the global optimal result. It is unnecessary to provide the detailed steps of the genetic algorithm, so only the key definitions are described here as follows. (1) Genetic representation.thetargetofk-onpp is finding k nodes out of possible set B to minimize group delay. A natural thought is using the tag of these nodes to represent the solution. So, we mark each node in set B with a number and represent each individual as a set of numbers. The size of the individual set is fixed to k. (2) Fitness function. The property of the fitness function is that the better the solution is, the larger the fitness will be. However, the cost function defined above is the group delay. Suppose that N denotes current generations and C denotes the fitness of calculated individual. We define fitness function as follows: 1 Cost (C) min (Cost (N)) max (Cost (N)) min (Cost (N)). (11) (3) Crossover. Randomly choose the crossover point in an individual set. Then, the tag number before or after the crossover point is swapped. However, the same point may appear twice in a single individual set after crossover. For example, this occurs if individual A is (3, 4, 7, 9); individual B is (1, 3, 5, 7), and the crossover has happened at the second point; one of the results is

109 International Journal of Distributed Sensor Networks 7 (3, 3, 5, 7). In case of such a scenario, we extract the set ofthesamepointsfromindividualsandcrosstheleft part. After the crossover, this set is added into both results. During the experiment, an adaptive crossover probability is used based on the work of Srinivas and Patnaik in [25]. (4) Mutation. Given a mutation probability, for each point in the individual set, randomly generate a number between 0 and 1. If the random number is bigger than mutation probability, select another point in candidate set B to replace this point. An adaptive mutationprobabilityisalsousedbasedon[25] TAG Algorithm. The TAG algorithm is a greedy algorithm proposed by Roy et al. in [16]. The TAG algorithm works as follows. It selects overlay nodes from candidate set B based on a greedy strategy. During each step, the algorithm chooses the node that gives the best value of the cost function. Suppose that there are already m nodes in the overlay node set O. Then,the(m + 1)th node is selected among the rest of the B-O nodes. There is no replacement strategy in TAG, so once the node is chosen, it cannot be modified. The time complexity of TAG is O(k 3 M), wherek is the given size of overlay set and M isthesizeofcandidatesetb. Algorithm results (ms) Size of candidate set B Average result of local search algorithm with K=5 Best result of genetic algorithm with K=5 Average result of TAG algorithm with K=5 Figure 3: Algorithm results comparison Experimental Results Comparison between Optimal Solutions and Genetic Algorithm Solutions. As a genetic algorithm is used to acquire the global optimal result, the efficiency of genetic algorithm must first be tested. Table 1 presents the comparison of the genetic algorithm with the global optimal algorithm in a small scale problem. The global optimal algorithm is achieved by the traversing method, so it is limited by the problem scale. For example, the time cost with M = 50 and k = 5 is s, which is almost 8 hours. For genetic algorithm, as described above, the best result out of 500 experiments is used to increase the possibility of finding the global optimal result. In the following, k denotes the size of overlay node set and M denotes the size of candidate set B. In order to eliminate randomness, these experiments are carried out with different network topology. So, experiment results of different parameter M are not comparable. As mentioned above, the adaptive crossover and mutation probability are used in the implementation of genetic algorithm. Adaptive function and parameters are the same as described in [25] except a minimum crossover probability is set to 0.1 and a minimum mutation probability is set to Population size is set to 300. The iteration stop condition is that the fitness value remains unchanged for 100 iterations. As Table 1 in the following page presents, the genetic algorithm works efficiently with the above definitions and parameters. The obtained results of genetic algorithm are very close to the results calculated by traversing method. Also, the best result of genetic algorithm can be the same as optimalresult as illustratedin Table 1. These results illustrate the efficiency of the proposed genetic algorithm. During the experiment, we also found that increased population size Time costs (s) Size of candidate set B Time cost of local search algorithm with K=5 Time cost of genetic algorithm with K=5 Time cost of TAG algorithm with K=5 Figure 4: Time cost comparison. would lead to increased probability of getting optimal results but with increased time cost too Comparison of Different Algorithms. Figure 3 compares the results of the local search algorithm, the TAG algorithm, and the best result achieved by the genetic algorithm. Clearly, in Figure 3, the results of local search algorithm are almost as good as the best result of the genetic algorithm, whiletheresultsofthetagalgorithmareveryunstable. Although in some cases the TAG algorithm can obtain a result close to local search algorithm, the result of TAG is not monotonic, as shown in Figure 3. This occurs when the candidate set B increases but the result decreases. This is

110 8 International Journal of Distributed Sensor Networks M=30 M=40 M=50 M/k Table 1: Comparison of genetic algorithm results with global optimal results. Average result of 500 experiments acquired by the genetic algorithm (ms) Best result of 500 experiments acquired by the genetic algorithm (ms) Global optimal results (ms) k= k= k= k= k= k= M TAG algorithm results Table 2: Stability of proposed algorithm. Mean value of proposed algorithm 95% quartile of proposed algorithm 95% confidence interval of mean value ( , ) ( , ) ( , ) ( , ) ( , ) ( , ) ( , ) ( , ) ( , ) ( , ) Algorithm results (ms) Number of overlay node K Algorithm results with M = 100 Algorithm results with M = 200 Algorithm results with M=300 Algorithm results with M=400 Algorithm results with M=500 Figure 5: Algorithm results with an increasing k. because TAG selects a node that gives the best cost value according to the current node set. However, a good node in onestepmaynotbepartofthebestoverlaynodeset.and, oncethenodeisselected,thereisnostrategyforreplacement. Figure 4 illustrates the time cost of different algorithms. Time cost of local search algorithm and TAG algorithm is the mean time cost of 500 tests. Time cost of genetic algorithm iscalculatedbasedonthemeanvalueofiterationstepwhich returns the best solution. As Figure 4 presents, the genetic algorithm is more timeconsuming than the local search algorithm, and the time cost of TAG is minimal. This result is consistent with the theoretical analysis. Additionally, these results clearly show that the time cost of the local search algorithm is linear with the size of candidate set B Stability of the Proposed Algorithm. As there is no randomness in the TAG algorithm, each of the 500 tests obtains the same result. For the proposed local search algorithm, there are random steps, so the corresponding result may fluctuate. However, during the tests, we discovered that the result of the proposed algorithm only varies over a small interval; therefore, the proposed algorithm works stably. Table 2 presents the 95% quartile of 500 tests and the 95% confidence interval of the mean value. Regardless of the distribution of the results, the confidence interval of the mean value is still reasonable because of the central limit theorem. Each of the experiments can be treated as an independent random variable, and these independent random variables have the same mean and variance. Thus, the mean value of these independent random variables follows the normal distribution. As Table 2 presents, the proposed algorithm works quite stably.

111 International Journal of Distributed Sensor Networks 9 Time cost (s) Number of overlay node K Time cost with M = 100 Time cost with M = 200 Time cost with M=300 Time cost with M=400 Time cost with M=500 Figure 6: Time cost with an increasing k Impact of the Number of Overlay Nodes. In Section 3, we noted that because of the cost of deploying overlay nodes and the cost of maintaining communication delay information between overlay nodes, the size of set O should be limited. Figures 5 and 6 present the algorithm results and time cost with k increasing under the same M. As Figures 5 and 6 show, the algorithm result and the number of overlay nodes are not linearly correlated. When k increases to a certain range, the algorithm results may remain unchanged. In contrast, the time cost will keep increasing. Thus, considering both the algorithm efficiency and cost of deployment, the size of the overlay set should be limited. 6. Experiment with Network Simulation In this section, the proposed local search algorithm is tested in a network simulation environment. This section is used to represent the algorithm performance and some factors that impact the performance. The network simulator EstiNet is applied EstiNet Network Simulation. EstiNet is a novel network simulator developed by Wang et al. since 1999 [26]. The current version is EstiNet 8.0, which can be found in [10]. A novel simulation methodology called kernel reentering methodology is implemented in EstiNet. It combines the advantages of both simulation and emulation. In contrast to existing network simulators NS2 or OPNET, EstiNet uses the real-life UNIX TCP/IP protocol stack during the simulation, and thus all the real-life network application programs can readilyrunonanynodeinasimulatednetworkwithoutany modification. This approach can only be performed in emulation mode with other simulators. However, with emulation mode, the performance of switches, links, and so forth, is controlled by the operating system, which makes the results unpredictable; thus, the simulation result cannot be repeated precisely. There is a useful tool in EstiNet called Generate Large Internet-like Network. It automatically generates a large network that is similar to the Internet. We use this tool to implement the following experiments Experiment Design. Figure 7 in the following page shows the schematic diagram of the proposed experiment. The red points denote the sensors that are used to collect data. The black points denote the data analyzer used to gather data from sensors. The blue points denote the overlay node, andthereddottedlinesdenotethelogicpathfromsensor to analyzer through the overlay node. A data-generating program is implemented in each sensor which periodically generates data packets with size L. Adataserverprogram is implemented in the analyzer. A data transfer program is implemented in the overlay node, and it retransfers each packet from the sensor to the next overlay node or data server. The goal is to find the optimal overlay node set with given size k that minimizes the sum of time delay from each sensor to the analyzer. In a realistic environment, a delay testing program and transfer program should be placed at each candidate node. Once the optimal overlay set has been found, the transfer program in overlay node is set to continue while the other transfer program suspends. The recalculation should be triggered by time or by network events such as increasing network delay from sensors. For simplicity of analysis, we decompose the progress in the simulation. The simulation steps are as follows. (1) First, we add network traffic in the simulation environment. (2) Then, we find the original time delay from the sensors to the analyzer. In this step, data-generating programs implemented in sensors are directly sending data to the server with the server IP address. (3) To obtain the optimal overlay node set, a delay matrix that indicates the time delay of each of the two nodes in candidate set B is needed. Thus, a program to measure time delay from each of the two nodes is required. Here, we use the ping method to acquire the RTTofeachofthetwonodes.Wemodifiedtheping programtocalculatethemeanrttandrecorditina file. (4)Weconstructthedelaymatrixoutofmeasuredtime delay and find the optimal overlay node set. (5) We implement the node transfer programs in the overlay nodes and retest the group time delay from the sensors to the analyzer. The simulation network includes 167 nodes; we randomly select 11 nodes as sensors and 1 node as the analyzer. The overlay node set size is set to 5, and all the other nodes are added into candidate node set. 5 overlay nodes are used because system performance improvement becomes trivial with more

112 10 International Journal of Distributed Sensor Networks Overlay path 4 Sensor Overlay path Overlay path Overlay path 1 5 Analyzer Overlay path Overlay path 2 Overlay path 3 Overlay path Overlay path Overlay node Data generator Data packet Overlay path Overlay path Data Data Data transfer Data transfer Data server packet packet Figure 7: Schematic diagram of the experiments. Table 3: Comparison of overlay network performance with increasing data size. D (Byte)/P (Byte) Original cost (ms) Overlay network cost (ms) Delay improvement (percentage) P = % D = 100 P = % P = % P = % D=1K P = % P = % P = % D=5K P = % P = % P = % D=10K P = % P = % P = % D = 100 K P = % P = % overlay nodes. To eliminate the effect of randomness of the network, each of the following experiments has been repeated five times Experimental Results and Analysis. Network communication delay consists of four parts: nodal processing delay, queuing delay, transmission delay, and propagation delay. Queuing delay, nodal processing delay, and propagation delay can be measured by RTT, while transmission delay is related to network bandwidth and data size. To test the overlay network performance with different conditions, different data sizes are used in the experiment. In addition, all the network programs used in the experiment are implemented based on the Linux socket. Within socket programming, a large fileshouldbedecomposedintosmallpacketfortransfer; otherwise, the delay would increase. Consequently, packet size is also used as a parameter in the experiment. Some interesting phenomena are revealed. Table 3 illustrates the overlay network performance with increasing data size. Here, D denotes the size of data sent from sensors and P denotes the program transfer packet size. The original cost means the group delay from sensors to analyzer without the overlay. The overlay network cost means the group delay after applying the overlay. The total improvement is calculated with a percentage. Table 3 illustrates that overlay network with the proposed algorithm can improve network delay under all conditions. Additionally, the transmission delay would affect system performance. When the sensor data size is small enough, the delay improvement is extremely good. This performance would decrease with the increase of sensor data size. However, after data size achieves a certain range, system performance would be stable. In addition, Table 2 also reveals that the transferring packet size should be set in agreement with MTU. When P is 1430, the system performance is the best. (Although when P is 1800, the delay improvement seems better and the actual delay cost increases.) 6.4. Further Discussion. Although the above experiments show good results of overlay network with the proposed

113 International Journal of Distributed Sensor Networks 11 Time costs (ms) Test times Figure 8: Time cost with random large files. algorithm, there is still work to be done. As described above, the weight of an edge in a network is represented by the average RTT. This weight matrix is the basic requirement for optimization. However, the mean value may not be sufficient to denote this network edge weight matrix. According to current research in internet performance, both internet traffic and delay show the self-similarity characteristic [27, 28].In a self-similarity time-series, mean value and variance are not appropriate for describing the property because they may not exist [29]. To simulate the effect of hop phenomenon of RTT in real network, a large file is randomly added into network traffic. According to [30], this may be one cause of self-similarity. In Figure 8, the effect of such a phenomenon is shown. Figure 8 shows the group delay of 30 former experiments with D=5KandP = Itrevealsthatnetworkdelay burst will affect overlay network performance, especially with an IoT application, as the data of an IoT application are generated periodically. As the mean value of RTT is not sufficient for overlay node placement optimization in IoT applications, other parameters such as the Hurst parameter should be imported to define the weight of edges. 7. Conclusions Network delay is one of the critical issues in IoT applications. Basedonbothperformanceandeaseofimplementation, an internet based dual-layer network, which refers to the overlay network, is suitable for the IoT. Overlay routing hasbeenprovedtobeafeasiblesolutiontooptimizethe end-to-end delay. In this work, one of the most important issues in overlay routing, the overlay node placement problem (ONPP), has been discussed. The NP-hardness of multihop k-onpp has been proven, and a theoretical boundary for k- OPNN optimization is provided. With defined parameters ω and α, thereisnopolynomial algorithm with an approximation ratio that is less than α (1 + ((ω 1)/e)). A local search algorithm has been proposed and a theoretical approximation ratio bound has been provided. The approximation ratio of the local search algorithm is less than ω/α. The proposed local search algorithm has been tested and compared with a genetic algorithm and the TAG algorithm with MATLAB tools. The results illustrate that the local search algorithm obtains better performance than the TAG algorithm, and the time cost is linear with the problem scale. The local search algorithm is finally tested with the network simulator EstiNet. The experimental results show a stable benefit from the proposed method. Moreover, an additional discussion about measuring network edge weight is provided, which reveals a future research direction for ONPP. Conflict of Interests The authors declare that there is no conflict of interests regarding the publication of this paper. Acknowledgments This work is supported in part by the Ministry of Science and Technology of China under the National 973 Basic Research Program (Grants no. 2013CB and no. 2011CB302505) and the National 863 Science and Technology Support Program (Grant no. 2013BAH19F01), the National Natural Science Foundation of China (Grants no ), thechinasouthernpowergridscienceandtechnology project (K-SZ ), and the Tsinghua National Laboratory for Information Science and Technology (TNList) crossdisciplinary research program. The authors thank the EstiNet Company for providing a free license for the EstiNet software and technical support. References [1] L. Atzori, A. Iera, and G. Morabito, The internet of things: a survey, Computer Networks,vol.54, no.15,pp ,2010. [2] D.Miorandi,S.Sicari,F.DePellegrini,andI.Chlamtac, Internet of things: vision, applications and research challenges, Ad Hoc Networks,vol.10,no.7,pp ,2012. [3] J. J. Wu and W. Zhao, WInternet: from net of things to internet of things, Journal of Computer Research and Development,vol. 50, no. 6, pp , [4] M. Chenine, I. Al Khatib, J. Ivanovski, V. Maden, and L. Nordström, PMU traffic shaping in IP-based wide area communication, in Proceedings of the 5th International Conference on Critical Infrastructure (CRIS 10), pp. 1 6, September [5] A. G. Phadke and J. S. Thorp, Communication needs for wide area measurement applications, in Proceedings of the 5th International Conference on Critical Infrastructure (CRIS 10), pp.1 7,September2010. [6] M. Chenine, E. Karam, and L. Nordström, Modeling and simulation of wide area monitoring and control systems in IPbased networks, in Proceedings of the IEEE Power and Energy Society General Meeting (PES 09),pp.1 8,July2009.

114 12 International Journal of Distributed Sensor Networks [7] A. Raha, S. Kamat, X. Jia, and W. Zhao, Using traffic regulation to meet end-to-end deadlines in ATM networks, IEEE Transactions on Computers,vol.48,no.9,pp ,1999. [8] S. Savage, A. Collins, E. Hoffman, J. Snell, and T. E. Anderson, The end-to-end effects of Internet path selection, ACM SIG- COMM Computer Communication Review, vol.29,no.4,pp , [9] S.Roy,H.Pucha,Z.Zhang,Y.C.Hu,andL.Qiu, Overlaynode placement: analysis, algorithms and impact on applications, in Proceedings of the 27th International Conference on Distributed Computing Systems (ICDCS 07),pp.53 63,June2007. [10] EstiNet, [11] D. G. Andersen, H. Balakrishnan, M. F. Kaashoek, and R. Morris, Resilient overlay networks, in Proceedings of the 19th ACM Symposium on Operating Systems Principles (SOSP 01),pp , [12] K. P. Gummadi, H. V. Madhyastha, S. D. Gribble, H. M. Levy, and D. Wetherall, Improving the reliability of internet paths with one-hop source routing, in Proceedings of the 6th Symposium on Operating System Design and Implementation (OSDI 04),pp ,2004. [13] Y. Zhu, C. Dovrolis, and M. Ammar, Combining multihoming with overlay routing, in Proceedings of the 26th IEEE International Conference on Computer Communications (INFOCOM 07),pp ,May2007. [14]L.Tang,Y.Huai,J.Zhou,H.Yin,Z.Chen,andJ.Li, A measurement study on the benefits of open routers for overlay routing, Journal of Communications,vol.4,no.9,pp , [15] S. Yang, Y.-A. Kim, and B. Wang, Designing infrastructurebased overlay networks for delay-sensitive group communications, in Proceedings of the 50th Annual IEEE Global Telecommunications Conference (GLOBECOM 07), pp , November [16] S. Roy, H. Pucha, Z. Zhang, Y. C. Hu, and L. Qiu, On the placement of infrastructure overlay nodes, IEEE/ACM Transactions on Networking, vol. 17, no. 4, pp , [17] N. Cleju, N. Thomos, and P. Frossard, Network coding node placement for delay minimization in streaming overlays, in Proceedings of the IEEE International Conference on Communications (ICC 10),pp.1 5,May2010. [18] R. Cohen and D. Raz, Cost effective resource allocation of overlay routing relay nodes, in Proceedings of the IEEE International Conference on Computer Communications (INFOCOM 11), pp , April [19] S. Jamin, C. Jin, A. R. Kurc, D. Raz, and Y. Shavitt, Constrained mirror placement on the Internet, in Proceedings of the IEEE International Conference on Computer Communications (INFO- COM 01),pp.31 40,April2001. [20] M.-Y. Wu, Y. Zhu, and W. Shu, Optimal multicast overlay placement for realtime streaming media, in Proceedings of the IEEE International Conference on Multimedia and Expo (ICME 04),pp ,June2004. [21] L. V. Á.Caraguay,A.B.Peral,L.I.B.López, and L. J. G. Villalba, Software-defined networking: evolution and opportunities in the development IoT applications, International Journal of Distributed Sensor Networks.Inpress. [22] European Lighthouse Integrated Project, Internet of Things Architecture, [23] R. Pan, D.-M. Zhu, S.-H. Ma, and J.-J. Xiao, Approximated computational hardness and local search approximated algorithm analysis for k-median problem, Journal of Software, vol. 16, no. 3, pp , [24] V. Arya, N. Garg, R. Khandekar, A. Meyerson, K. Munagala, and V. Pandit, Local search heuristics for k-median and facility location problems, SIAM Journal on Computing, vol.33,no.3, pp , [25] M. Srinivas and L. M. Patnaik, Adaptive probabilities of crossover and mutation in genetic algorithms, IEEE Transactions on Systems, Man and Cybernetics, vol.24,no.4,pp , [26] S. Y. Wang, C. L. Chou, C. H. Huang et al., The design and implementation of the NCTUns 1.0 network simulator, Computer Networks,vol.42,no.2,pp ,2003. [27]W.E.Leland,M.S.Taqqu,W.Willinger,andD.V.Wilson, On the self-similar nature of Ethernet traffic, ACM SIGCOMM Computer Communication Review, vol.24,no.3,pp , [28]M.S.Borella,S.Uludag,G.B.Brewster,andI.Sidhu, Selfsimilarity of Internet packet delay, in Proceedings of the IEEE International Conference on Communications (ICC 97),pp , June [29] M. Li, Fractal time series-a tutorial review, Mathematical Problems in Engineering, vol. 2010, Article ID , 26 pages, [30] M. E. Crovella and A. Bestavros, Self-similarity in world wide web traffic: evidence and possible causes, IEEE/ACM Transactions on Networking,vol.5,no.6,pp ,1997.

115 Hindawi Publishing Corporation International Journal of Distributed Sensor Networks Volume 2014, Article ID , 8 pages Research Article Constructing the Green Campus within the Internet of Things Architecture Hsing-I Wang Department of Information Management, Overseas Chinese University, Taichung 407, Taiwan Correspondence should be addressed to Hsing-I Wang; hsing@ocu.edu.tw Received 2 October 2013; Revised 24 January 2014; Accepted 27 January 2014; Published 5 March 2014 Academic Editor: Luis Javier Garcia Villalba Copyright 2014 Hsing-I Wang. This is an open access article distributed under the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited. The Internet of Things radically changes the view of the Internet by embracing every physical object into network. The vision of IoT promisestoenhancethecapabilitiesofobjectsandformsasmartenvironmentsothatpeoplewillbenefitfromtheiot revolution. As the global population grows, the resources on earth are depleted quickly. In order to have a sustainable earth, governments around the world put a lot of efforts to advocate the reduction of carbon production as well as to emphasize the benefits of reducing the consumption of energy. The proposition has been promoted on campus of educational institutions as well. Smart campus is a trendy application in the paradigm of the IoT. This research adopts the concept of the Internet of Things to construct a green campus environment which will realize the idea of energy saving. The architecture of the construction of green campus is established and three application systems have been developed as well. The efforts of this work allow the campus to manage the computer labs and the air conditioners more efficiently. The sensor network will save more energy since data are reported periodically and the analysis will be carried out in time to locate the problems. 1. Introduction The advances of emerging technologies have broadened the meaning as well as the applications of the Internet. In other words, almost every object can be part of a network. With smart connectivity, physical objects are networked and will gaintheabilitytocommunicatewitheachother.thevision of The Internet of Things (IoT) promises to enhance the capabilities of objects and forms a smart environment so that people can benefit from the IoT revolution [1, 2]. The IoT applications cover the building of smart cities, the setup of smart environment, the provision of smart public services, the plan of ehealth, and the building of smart home/office, and so forth [1, 3]. As the global population grows, the resources on earth are depleted quickly. In order to have a sustainable earth, governments around the world put a lot of efforts to advocate the importance of the reduction of carbon production as well as to emphasize the benefits of reducing the consumption of energy. The proposition has been promoted on campuses of educational institutions as well. Smart campus is a trendy application in the paradigm of the IoT. The concept of constructing a Smart campus implies that the institution will adopt advanced ICTs (Information Communication Technologies) to automatically monitor and control every facility on campus. To use the facilities more efficiently and to minimize the energy consumed are believed the most important advantages of building a smart campus. Such effortsarealsorecognizedasconstructinga greencampus. Two major ICTs which make the realization of IoT possible are the emergence of cloud computing and the network of wireless sensors. In fact, cloud computing and wireless-sensor network together can provide the most reliable, scalable, dynamic,andcomposableresourcesthattheiotsrequired [4 7]. Wireless sensor networks are, particularly, adopted in manyurbancitiestoprovidesmarterandadvancedlives[8]. In order to construct a green campus with the utilization of the Internet of Things, this research reviews the cores of IoT, cloud computing, and wireless sensors network. Thereafter, the definition, the architecture, and the steps for the development of green campus are proposed. This paper also demonstrates our work toward constructing a green campus and the system we have developed. The ultimate goal of this work is the implementation of a cloud-based monitoring

116 2 International Journal of Distributed Sensor Networks system built upon wireless sensor network architecture so that data are gathered and stored on cloud database and the analysis can be carried out periodically. The paper is organized as follows. In Section 2,thispaper discusses the development of green campus in depth. Some well-knowngreencampusprojectswillbeintroduced.cloud computingaswellasthedefinition,thearchitecture,and the applications of the Internet of Things are reviewed in Section 2 as well. This paper then discusses the construction of green campus within the IoT architecture. In Section 4, this paper shows the lab management system that we have developed within the IoT architecture. And finally, the conclusion and further works are given in Section Literature Review 2.1. The Development of Green Campus. New emerging technologies have changed human life styles dramatically. As people enjoy advanced and smart lives, ironically, our earth is facing a major crisis that may bring disasters to human lives as well. Figure1 shows the carbon emission records from January 1955 to January The concentration of atmospheric CO 2 was below 320 ppm in By June 2013, the number has increased by 25%. The data indicate how serious the earth has been polluted. In addition, more environmental crises such as global warming and climate disturbance; acid rain and soil erosion; ecosystem damage have got the attention across the world [9, 10]. Information technologies have been introduced to campus to, hopefully, yield new levels of institutional and instructional productivity and to reduce instructional costs at the same time [12]. However, the research revealed that educational institutions might have benefited from information technologies in the areas of content, curriculum, and pedagogy; the costs saved had not shown obvious achievement [12]. On the contrary, the budgets for the investments on information technologies have been increased. More computers, printers, and ICT equipment have been purchased. More environmental pollution issues have been raised as a consequence of the use of information technologies on campus. In other words, instead of investing more on physical facilities, the universities should search for other inexpensive solutions. Scholars and experts have agreed that the knowledge of protecting the earth should be cultivated by educations. Universities should provide leadership for broader society [9]and institutions of higher learning have a special responsibility to address the continuing environmental crisis [9, 10]. In [9], the author specifically points out that one of the greatest opportunities and abilities to conserve energy was through facilities management on campus. Educational institutions across the world, especially the higher education, have recognized that they are in a unique position to prevent the crisis from getting worse. Not only are the faculties realizing that they possess the intellectual capacity to address these issues, but also the institutions are putting a lot of efforts in the integration of all resources and effectively adopting new technologies to their missions to create a green environment. Figure 1: Carbon emission from 1955 to 2015 [11] Some Well-Known Examples. There are a number of wellknown green campus examples. Harvard University also believes universities should play the environmental stewardship. The way Harvard University contributes to protect the earth is by including the commitment to a greenhouse gas (GHG) reduction goal of a 30 percent by In addition, Harvard University also established green building standard to ensure all sustainable design [13]. The mission that the University of Pennsylvania promises is to develop plans to reduce the emissions of greenhouse gases. The climate action plan was launched in In the 2011 progress report, the record showed that total energy usage decreased by 9.5%. The overall carbon emissions per capita on campus remained the same despite the growth of campus [14]. MacquarieUniversityinAustraliaisanotherexample. Macquarie University was one of the greenest universities in Australia. In order to make Macquarie University greener, the goals are set to reduce the total energy consumption per year per EFTP (Equivalent Full Time Persons) by 15% from 2005 by 2014 and to reduce the total GHG emissions produced peryearforcampusoperationspereftpby30%of2005 emissions, and so forth [15]. The goals for green campus at the university of Copenhagen are to reduce energy consumption to a level in 2013 that is 20% below that of 2006; and the CO 2 emissions from energy consumption shall be reduced in 2013 to a level that is 20% below that of 2006; and at least 75% of all purchases via purchase agreements shall require sustainability [16]. InTaiwan,Y.S.SunGreenBuildingResearchCenter Located at the NCKU Li-Hsing Campus is Taiwan s first zero-carbon, energy-saving building. The building is very famous to people in Taiwan as The Magic School of Green Technology. Embedded within The Magic School is the hope that its design principles can eventually be scaled to Taiwan s metropolitan centers [17]. The building was designed to use

117 International Journal of Distributed Sensor Networks 3 adequate techniques, instead of expensive techniques, to achieve quadruple benefit. The aims are estimated to save 50%energy,toconserve30%water,andtoreduce30%carbon emission. It is also expected that the building will be utilized for one hundred years [18]. The building started operation in January 2011, and in six months, the accumulated Energy Usage Intensity (EUI) was 19.3 kwh/m 2.Thefigurewasfar less than Taiwan s medium and low intensity office buildings, which consume 125 kwh/m 2 per year in average [18]. The existence of The Magic School of Green Technology will be amodelforallotheruniversitiesintaiwan. However, in the review of the outstanding green campus examples, we noticed that most action plans that the universities initiated focused on the design of green buildings, environment, or the purchase of energy-saving facilities. Information technologies were rarely, if not entirely, applied or considered in the plans. In fact, contemporary information technologies may contribute a lot in energy saving or in the protection of the environment if they are used smartly. The IoT is one of the smart solutions The Internet of Things. The concept and the realization of the Internet of Things make the world truly ubiquitous since the IoT radically changes the view of the Internet by embracing every physical object into network [4, 19]. The communications could take place not only between things but also between people and their environment in the IoT [6]. With the combination of the Internet, the cloud services, the near-field communications, real time localization, and embedded sensors, we can transform all objects into smart objects so that all components can understand and react to their environment [5]. The new concept of IoT will tie the Internet of information and services together [6] and as a result, more data, more information, and the knowledge will be generated and used. The term Internet of Things has become very popular in recent years. There are books to teach or to discuss various subjects about the IoT. International conferences open up sessions for scholars and specialists to exchange their ideas, opinions, and experiences regarding the development or the applications of the IoTs. And finally in 2009, even the EU Commission realized the importance of the revolution of the Internet and initiated an IoT action plan [20]. In [21], it is suggested that an IoT must be Internetoriented(middleware),thingsoriented(sensors),andsemantic oriented (knowledge). Based on the assertions, [4] proposed that the architecture of an IoT actually contains three segments which are the hardware segment, the middleware segment, and the presentation segment. The hardware segment mainly refers to the connection of sensors or any embedded communication hardware. The middleware segment usually refers to cloud environment which is responsible for data storage, computation, and data analytics. The presentation segment, on the other hand, visualizes the result of data analytics or interprets the data in an easy and understandable format. Moreover, an IoT must possess the capabilities of communication and cooperation, addressability, identification, sensing, actuation, embedded information processing, localization, and user interfaces [19]. At the hardware segment, wireless sensor network is expected to be a key technology for various IoT applications such as home automation [22] and energy saving [23]. The sensor devices in the wireless sensor network work as the communicate node and will communicate to other devices wirelessly [24]. The sensor device also carries out its designateddutytocollectdataandsenddatatodatacenter.therefore, communication and measurement are the two major functions of a wireless sensor network [24]. Sensors can be deployed randomly and densely with much less costs in a wireless sensor network environment. All the conditions are monitored at all times. Therefore, it is believed that the construction of a green campus based on the IoT concept is more advanced than merely purchasing the energy-saving facilities. ZigBee is the name of a standard that specifies the application layer of a wireless network in a small area with a low communication rate [25]. Previous researches and projects have shown that ZigBee sensor networks are suitable for applications in many different areas Cloud Computing. Earlier sensing network applications for environmental monitoring were mostly event-driven [26]. Thedatawerecollectedupontheoccurrenceoftheinstance. Wireless sensor network provides real time monitoring opportunities. As a result, more space is required to store the data and smart tools are mandatory in order to analyze the data.cloudcomputingisrecognizedasthebestsolution[27]. The major function of cloud computing is the delivery of services. It is not new to consider the pursuit of service as the entire and sole philosophy in the adoption of new technology. Clustering computing, grid computing, and service oriented architectures are the three famous examples that have seamlessly combined technologies with business flow. Cloud computing is similar to the aforementioned concepts but with three unique characteristics, which include virtual, dynamic provision on demand, and negotiation. Therefore, in the literature, cloud computing is defined as offering hardware and software resources as services across a parallel and distributed system consisting of a collection of interconnected and virtualized computers that are dynamically provisioned [28, 29]. Cloud services are classified into four categories. (i) Infrastructure as a service (IaaS): main services include the provision of virtual hardware, network, storage, computing power, and so forth. The clients include IT managers or software developers. Amazon S3 (simple storage service) and EC2 (elastic compute) arethetwowell-knownservicesinthecategory. (ii) Software as a service (SaaS): SaaS represents a new concept of software on demand. The software refers to application systems that can be activated directly on the internet. For example, the customer relationship management system provided by Saleforce.com is commonly adopted by businesses. (iii) Platform as a service (PaaS): PaaS delivers a service orientedplatform.thewholeprocessinthesoftware development life cycle (i.e., design, test, execution,

118 4 International Journal of Distributed Sensor Networks and deployment) would be provisioned as an integrated service over the Internet. Services in this category include the APP engine from Google and Azure from Microsoft. The software developers are possibly the major clients in this category. (iv) Database as a service (DaaS): DaaS, such as the SSDS of Amazon, moves the traditional database features, including the definition of data and the storage and retrieving of data, over to the network. The services protect clients from tracking long timing transactions or assist with the maintenance of the integrity of the data. Software developers may be the major clients. Cloud services promise users no longer being confined to limited space, time, or the compatibility of computers [30]. To adopt existing services from the Internet will minimize the expenditures spent on the information facilities and management. The pay-by-use mechanism will rationalize the market of intelligent property. Furthermore, with all services in cyber space, businesses or users will have more flexibility and options in search of the best alternatives and hence save more preprocess time. The new technology may guarantee the leverage of acceptance of users and smooth the introduction of new systems to the users as well. The users will always be updated with the new version and will not be bothered by copyright issues. For the enterprises, the management of documents, the cooperation, and the coordination within the organization will be easier. According to the definition, in the cloud paradigm, there are many distributed systems. In many cases, the distributed architecture consists of wireless sensor networks which are responsible for sensing data. Cloud computing usually plays importantroleinthiskindofarchitecturesincewireless sensor networks are limited in their processing power, battery life, and communication speed, while cloud computing is known for having powerful computational and processing capacity and the communication speed is much faster as well [31].Cloudcomputingisalsobelievedtheparadigmfor delivering services in the realization of IoT environment [4]. Previous researches have proved that the merits and the performances of wireless sensor network will be doubled when the architecture is combined with cloud environment [27, 32, 33]. Cloud environment is also more flexible to be migrated once the university wants to expand the object network and move toward building a smart campus with more smart applications. 3. Constructing Green Campus within IoT Architecture To construct green campus within IoT architecture is exactly the same as running a business. The goals must be clear and a set of objectives should be established. The missions that will be achieved toward the vision are then to be carried out. Figure 2 shows the procedures of the construction of green campus within IoT architecture. In addition to define the vision and objectives, [6] suggested that the issues that need to be addressed in IoT environment also include the transformation of everyday Step 1 Step 2 Vision Objective Objective Objective Action plans overall campus redesign Step 6 Integrated management System architecture Transform objects into smart objects Human involvement System design and development For each plan Figure 2: The procedure of the construction of green campus within IoT architecture. objects into smart objects, the plan of system architecture, the design and development of the systems, the integrated management, and human involvement. In Figure 2, action plans that are related to the overallcampusredesignshouldbeproposedaftervisionand objectives are determined. For each action plan, all the related objects are transformed into smart objects. The system architecture is drawn and the system within the IoT is developed. All projects should be managed in an integrated view and the entire plan should be supported by all people in the organization. The involvement of the executive administration level is especially essential to sustain the project. 4. The Application of IoT on Campus-Lab Management This research takes lab management to realize the architecture we have planned (as shown in Figure 2). The details are described below The Architecture. The first phase of our project is to set up an IoT in our computer labs. Following the steps to construct green campus within IoT architecture given in Section 2, the vision of this project is to efficiently control the use of the computers and air conditioners. Although the definite figures were not determined, the objectives of the project were to reduce the idle time of computers and to reduce the electricity costs. During the second step, every air conditioner wasassignedanip;therfidaswellasthezigbeesensors were installed as well. Based on the definition and the required elements defined before, Figure 3 shows the proposed architecture of part of our green campus within IoT. The system architecture consists of three major segments which are the hardware segment, the middleware segment, and the presentation segment. The hardware segment mainly uses RFID to induce the students who are going to enter the computer labs. The system reads and saves the student ID. The IoT is setup to connect the computers and the air conditioners in the lab. Every computer in the lab has its own IP, so does the conditioner. Steps 3 5

119 International Journal of Distributed Sensor Networks User applications 5 Green campus data and applicants center Ubiquitous sensors RFID Student ZigBee General affair controller Analysis The IOT Control Computation Visualization Figure 3: The architecture of the green campus within IoT proposed in this research. Figure 4: The emitter which equips with the temperature sensor and sends out the temperature reading continuously. The temperature sensor module of ZigBee is used to monitor the temperatures in the lab. In our work, a ZigBee network is constructed with ZB devices from Dmatek Limited Taiwan. The specifications of the devices are as follows: (i) radio frequency: 2.4 GHz band; (ii) data rate: bit/s (max to ); (iii) distances: 10 meters; (iv) number of channels: the device is able to search up to 32 satellite channels; (v) 10 I/O ports. The emitter of the temperature sensor is shown in Figure 4. The emitter device is in the lab and connects to the IoT. The emitter senses the temperature of a lab and sends out the signal continuously. The receiver device of the temperature sensor is shown in Figure 5. The receiver connects to a PC via a USB interface. The receiver device will collect all the data sent by the emitter. Figure 5: The receiver that reads the signal from emitter. There is a cloud server in the middleware segment. The server owns the database, all the applications, and the tools. All the data collected, including the data read by RFID, the status of each of the computers in the lab, and the temperatures of the computer room, are sent to the cloud server. The data then are computed, analyzed, and controlled. At the presentation segment, two major systems are provided to students and controller of the general affairs office. The students may use computers or any mobile devices to connect to the system and retrieve the usage status of the selected computer lab. This will allow the students to make proper decisions if they still want to go to the labs which might not have seats available. The second system is at the general affairs office site. The status of the usage of computer labs as well as the changes of the temperatures of each lab are analyzed and updated every 30 minutes. The results on the screen allow the controller to control the air conditioners in the lab. In addition, a network alert system will track the usage of each computer so that

120 6 International Journal of Distributed Sensor Networks Figure 6: The system shows the usage status of every computer lab. Figure 8: The system indicates the number of seats that are occupied, the seats that are still available, or the number of computers that are not in function. Figure 9: The warning message will pop up on the screen either when the computer has been occupied for more than an hour or the computer was assigned to a student but has been detected idle for some time. Figure 7: RFID reader senses a tag and assigns a seat to the student. thecomputerwillbeshutdownonceithasbeenidlefora designated time. 4.2.TheIntroductionoftheSystemtheInternetofThings. The prototype of the computer labs control system has been developed in this research. Figure 6 through Figure 12 demonstrate how the system operates. On the lab side, the system tracks the usage of every computer lab at all times (Figure 6). The system gives the information of computers that is occupied, available, or malfunctioning. Once a student enters a lab, the RFID reader reads his or herid,thesystemwillassignanavailableseattothestudent (Figure 7), and the status of that seat will be marked with green color to indicate that the seat is in use (Figure 8). Eachstudentisallowedonehourtousethecomputer.A warning message will be given and the computer will be shut down automatically by the system (Figure 9)ifthecomputer hasbeenoccupiedformorethanonehourorifthesystem detects that the computer has been idle for some time. A system control dashboard is provided to the controller in the general affairs office. Four functions are available at the present system. The first tab shows the same labs information asthestudentscansee.thesecondtab(figure 10) givesthe current temperature of a selected lab. By clicking the on/off button, the controller is able to turn on or turn off the air conditioners in the lab. The third page shows the real time average temperatures of all computer labs (Figure 11). In every 30 minutes, this system records the average temperatures of all computer labs. The records are shown on the fourth page of the dashboard (Figure 12). The temperatures that are below 26 are marked with green. If the temperatures are higher than 30, red colors appear to show the warning. Yellow colors are shown if the temperatures are in between. Together with the information of the status of computer labs, air conditioners, and the changes of the temperatures as well as the statistics of the temperatures in the labs, the controller can make decisions easily. The decisions such as how many labs should open to students, when and which air conditioner should be turned on, and finally, the controller can also monitor if the computers are used properly and efficiently. 5. Conclusion This research appeals to the responsibilities the universities should bear in the issues of environmental protection. The performance that information technologies may contribute to the sustainability of universities is emphasized in the

121 International Journal of Distributed Sensor Networks 7 Figure 10: The status of each of the air conditions in the computer lab. Figure 12: The changes of the temperatures in the computer labs. Figure 11: The average temperature measured by ZigBee temperature sensor in the computer labs. paper. This research also proposes the steps as well as the architecture of how to construct a green campus by utilizing theadvancedtechnologiessmartly. Furthermore, this research adopts the concept of the Internet of Things to construct the green campus which will realize the idea of energy saving. The objects of our work include the computers and air conditioners. RFIDs and the ZigBee device with temperature module are used to build up the wireless sensor network. Thecontributionsdeliveredbythesystemwehavedeveloped include the following. (i) The computer labs can be managed efficiently. More labs will be open only when the demand is increasing. (ii) The use of the computers will be monitored at all times. This mechanism decreases the number of idle power-on computers. (iii) The air conditioners will be turned on only when the temperatures reach a preset level. As a result, more energy will be saved. (iv) Combing the wireless sensor network and cloud computing, the architecture we proposed will collect real time data from sensor. The results of the analysis of data will be sent to the appropriate party so that properactionscanbetakenintime. (v) The architecture we proposed allows users to connect to the system with any mobile device in any place. The idea of constructing a green campus is just the first step in our institution. This research shows how to build up the IoT to manage computer labs. The performance of the current project will be examined continuously. The next phase is to build the IoT around the whole campusand,thereafter,theintegrationofallthesubsystemswill be carried out. The energy-saving program has full support from the administration office. However, the idea of green and sustainability has not yet been planted in everybody s mind. In other words, more educating programs need to be arranged to broadcast the concept of green campus. Hopefully, as a higher educational institution, we can show some leadership and demonstrate our responsibilities to the society. Conflict of Interests There is no conflict of interests regarding the publication of this paper. Acknowledgment This paper was financially supported by National Science Council, Taiwan, NSC I A1. References [1] A. Gluhak, S. Krco, M. Nati, D. Pfisterer, N. Mitton, and T. Razafindralambo, A survey on facilities for experimental internet of things research, IEEE Communications Magazine, vol. 49, no. 11, pp , [2]M.Zorzi,A.Gluhak,S.Lange,andA.Bassi, Fromtoday s intranet of things to a future internet of things: a wireless- and mobility-related view, IEEE Wireless Communications, vol. 17, no. 6, pp , [3] Libelium, 50 Sensor Applications for a Smarter World, 2013, 50 iot sensor applications ranking.

122 8 International Journal of Distributed Sensor Networks [4] J.Gubbi,R.Buyya,S.Marusic,andM.Palaniswami, Internet of Things (IoT): A Vision, Architectural Elements, and Future Directions, FGCS, [5] G. Kortuem, F. Kawsar, D. Fitton, and V. Sundramoorthy, Smart objects as building blocks for the internet of things, IEEE Internet Computing,vol.14,no.1,pp.44 51,2010. [6] O. Vermesan, P. Friess, P. Guillemin et al., Internet of things strategic research roadmap, in IoT Cluster Strategic Research Agenda, chapter 2, pp , [7] B. Singh and D. K. Lobiya, A novel energy-aware cluster head selection based on prticle swarm optimization for wireless sensor networks, Human-Centric Computing and Information Sciences,vol.2,no.13,18pages,2012. [8] M.S.Obaidat,S.K.Dhurandher,D.Gupta,N.Gupta,andA. Asthana, Dynamic energy efficient and secure routing protocol for wireless sensor networks in urban environments, Journal of Information Processing Systems,vol.6,no.3,pp ,2010. [9]L.Sharp, Greencampuses:theroadfromlittlevictoriesto systemic transformation, International Journal of Sustainability in Higher Education,vol.3,no.2,pp ,2002. [10] W. Simpson, Energy sustainability and the green campus, Planning for Higher Education, vol. 31, no. 3, pp , [11] Earth s, 2013, [12] K. C. Green and S. W. Gilbert, Great expectations, Change, vol. 27,no.2,pp.8 19,1995. [13] Sustainability at Harvard, Harvard s Commitments to Sustaionability, 2013, [14] University of Pennsylvania, Climate Action Plan Progress Report 2011, default/files/pdf/penn 2011 Climate Action Plan Progress Report.pdf. [15] Macquarie University, Sustainability strategy, target 2014, 2013, us/strategy and initiatives/ sustainability. [16] Green Campus at the University of Copenhagen, 2013, climate.ku.dk/green campus/goals/. [17] B. Fox and J. D Angola, Taiwan s Magic Green School, 2011, [18] The Magic School of Green Technologies, 2009, class&pic dir list=1. [19] F. Mattern and C. Floerkemeier, From the internet of computers to the internet of things, in From Active Data Management to Event-Based Systems and More, K.Sachs,I.Petrov,andP. Guerrero, Eds., vol of Lecture Notes in Computer Science, pp , Buchmann Festschrift, [20] European Commission, Internet of Things An action plan for Europe, COM, 278, 2009, LexUriServ/site/en/com/2009/com en01.pdf. [21] L. Atzori, A. Iera, and G. Morabito, The internet of things: a survey, Computer Networks, vol. 54, no. 15, pp , [22] ZigBee Alliance, ZigBee Home Automation Public Application Profile, 2007, [23] ZigBee Alliance, The Choice for Energy Management and Efficiency, ZigBee White Paper, 2007, [24] M. Terada, Application of zigbee sensor network to data acquisition and monitoring, Measurement Science Review,vol. 9, no. 6, pp , [25] ZigBee Alliance, ZigBee Specification, 2006, [26] P. Morreale, F. Qi, and P. Croft, A green wireless sensor network for environmental monitoring and risk identification, International Journal of Sensor Networks,vol.10,no.1-2,pp.73 82, [27]S.K.Dash,S.Mohapatra,andP.K.Pattnaik, Asurveyon applications of wireless sensor network using cloud computing, International Journal of Computer Science & Emerging Technologies,vol.1,no.4,pp.50 55,2010. [28] Y. S. Tsai, The Issues to Think about Before Entering the Cloud Services Market, 2009, [29] H. R. Motahari Nezhad, B. Stephenson, S. Singhal, and M. Castellanos, Virtual business operating environment in the cloud: conceptual architecture and challenges, Lecture Notes in Computer Science,vol.5829,pp ,2009. [30] Y. Pan and J. Zhang, Parallel programming on cloud computing platforms challenges and solutions, Journal of Convergence,vol.3,no.4,pp.23 28,2012. [31] W. Kurschl and W. Beer, Combining cloud computing and wireless sensor networks, in Proceedings of the 11th International Conference on Information Integration and Web-based Applications & Services (iiwas 09), pp , [32] R. Piyare, P. Sun, M. Se Yeong et al., Integrating wireless sensor network into cloud services for real-time data collection, in Proceedings of the International Conference on ICT Convergenc (ICTC 13), pp , Jeju, Republic of Korea, [33] P. Zhang, Z. Yan, and H. Sun, A novel architecture based on cloud computing for wireless sensor network, in Proceedings of the 2nd International Conference on Computer Science and Electronics Engineering (ICCSEE 13), pp ,Hangzhou, China, 2013.

123 Hindawi Publishing Corporation International Journal of Distributed Sensor Networks Volume 2014, Article ID , 12 pages Research Article Ensuring Healthcare Services Provision: An Integrated Approach of Resident Contexts Extraction and Analysis via Smart Objects Nan-Chen Hsieh, 1 Lun-Ping Hung, 1 Jong Hyuk Park, 2 and Neil Y. Yen 3 1 Department of Information Management, National Taipei University of Nursing and Health Sciences, Taipei 10845, Taiwan 2 Department of Computer Science and Engineering, Seoul National University of Science and Technology, Seoul , Republic of Korea 3 School of Computer Science and Engineering, University of Aizu, Aizu-Wakamatsu , Japan Correspondence should be addressed to Lun-Ping Hung; kiel.hung@gmail.com Received 30 September 2013; Accepted 17 December 2013; Published 2 March 2014 Academic Editor: Luis Javier Garcia Villalba Copyright 2014 Nan-Chen Hsieh et al. This is an open access article distributed under the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited. Advances in healthcare applications benefit the scenario of medical service provision for both care staff and medical institution than ever. Following such success, a healthcare system, named RCTC (Resident Classification by Types of Care), which grades individuals in terms of the severity of five functional status assessment aspects: mental, movement, eating, toilet (urination and defecation functions), and medical treatment, was developed in this study. This system is designed based on a conceptual model that follows the assessment/classification/placement sequence for long-term care institutions. It is implemented to deliver appropriate services according to individual needs based on its preprocessing of classification and further reduces the costs of manpower and loading of care staff through the analysis of assessment logs. With this systematic appraisal, this system can not only help care staff determine the needs of residents but also produce personalized health plans (i.e., weekly schedule towards comprehensive assessment and personalized care services). Results of implemented (and in use as well) system have demonstrated the feasibility that it can enhance the quality of care services to residents, working load of care staff, and efficiency of care-related information management for medical institution. 1. Introduction Health information technology (HIT) can reduce mistakes of medication or diagnosis, help medical professionals to obtain patient information in a timely manner, shorten their waiting time, and improve care quality and efficiency [1, 2]. Currently, HIT has been widely applied in health care industries. For example, the common HIT systems can help practitioners to input medical advice, track patient status, and record prescription and medications. Electronic medical records can keep patients personal health records, including medical records, inspection reports, and medical images [3], and a standard transmission format has been established to enable cross-hospital delivery. For patient safety, computeraided diagnosis and treatment planning have been used to avoid failure to diagnose and improper treatment. One study explored the debate and initiatives concerning the use of HIT in health care in developing countries and discussed the main advantages, limitations, and perspectives [4]. Long-term care (LTC)institutionshavelaggedbehindothersettingsinadopting HIT. Two early studies for resident classification were developed to improve placement decisions and to provide information for resource placement in LTC institutions [2, 5], but these systems lacked HIT support for comprehensive resident assessment. Several studies have presented factors that affect electronic health record (EHR) adoption in LTC institutions [6] and provided descriptions of the early users experiences with EHRs in LTC institutions [7]. A recent study [8] examined the adoption and utilisation of EHRs in LTC institutions and identified the barriers preventing the implementation of EHRs. Assessment of the needs of healthcare [9] isthemost important task for LTC institutions. The care staff should know the overall functional status of residents before providing required care services. Any assessment of LTC residents

124 2 International Journal of Distributed Sensor Networks must be multidimensional and comprehensive enough to measure all health-related aspects of residents [10]. To assess care needs, Japan uses the care needs certification scale [11] to assess body function, actions in daily life, living function, cognitive function, behavioural and psychological symptoms of dementia, social adaptation, medical use over the past fourteen days and independence in daily life, which are classified into seven levels according to care needs. Germany uses ADLs and IADLs [12] as assessment tools. These tools areclassifiedintothreelevelsofcareneedsintermsof care time needed. Korea employs quantised ADLs to assess care subjects, which is classified into three ranks. The above assessment tools could be used for the general populace but are not widely used for LTC residents because they cannot meet the needs of residents with cognitive functional disorders. For this population, the International Classification of Function, Disability and Health (ICF) was published by World Health Organization to standardise descriptions of health and disability and can be used to assess residents cognitive function using the ICF and its qualifiers [13]. In relation to the fee schedule of LTC institutions, the Minimum Data Set 2 (MDS 2) [14] isthemandated assessment tool in American nursing homes. MDS 2 lists 18 risk assessment items and assessment criteria in total. The Resource Utilization Groups III (RUG-III) [15] casemix system provides resident-specific means of allocating resources based on the variables of care costs for residents with different care needs. RUG-III uses a large number of rehabilitation-related and medical-care-related explanatory variables, and it is followed by the Japanese LTCI classification system [16]. A TAI-based system [17] was designed mainly as a resident management tool and includes variables related to the resident s functional status. Researchers [18]proposed RUG-III and a Japanese classification system for LTCIs. Their system contained rich medical explanatory variables and explained total medical cost better than the classification in a case-mix system. However, even medical components had a positive effect on explanatory power; the use of medicalneeds-related variables will increase the complexity of a classification system, consequently restricting their utility to fee schedules. Okochi and Takahashi suggested that a simpler coding system, such as a TAI-based system, will be helpful to improve resident management and to represent the practical care needs of the residents. Without the aid of a software system, the types of care classification and care needs assessment are difficult to achieve because the assessment content mainly comprises complex health conditions and care needs. The assessment criteria also differ from each other for different assessors. Most previous researchers [19 22] would refer to the content of MDS when developing comprehensive care need assessment tools. Most software developers would also apply MDS as the assessment scale for the assessment package. These software packages offer a variety of features beyond data entry, including functions to help facilities support care tasks and improve efficiency in meeting multiple reporting requirements. However, the entire assessment process of these systems is complicated and time consuming, which is not suitable for Taiwan s institutions. By referring to a TAI-based system assessment content, we designed the RCTC system based on a conceptual model that follows the assessment/classification/placement sequence for LTC institutions. This sequence can be designed to resemble the sequence of tasks carried out by practitioners in the institution, and it can accommodate advanced HIT features that integrate the RCTC system with this sequence to enhance the institution administrator s ability to assess and provide resident care needs, track the quality of care service, make resident placement decisions, and monitor multiple performance indicators for the institution. InTaiwan,thereisintensivedebateaboutLTCinsurance [23], and however, there is no unified standard for resident assessment, so the institutions lack useful resident needs assessment tools to deliver personalised care. Taiwan is still inthepreliminarystageofemployinghittohelpprovide proper care services for the residents of LTC institutions. Currently, data processing tasks are mostly still performed manually or semimanually. Some commercial LTC systems are designed as general electronic forms packages. Therefore, we developed the RCTC system to help care appraisers to assessresidents overallfunctionalstatus,requiredcareneeds, care methods, and estimated care time. This pilot study of the RCTC system cooperated with an eighty-eight-bed LTC institution. This institution belongs to a health service company. The health service company provides personal LTC service and owns two regional hospitals and five LTC institutions. We gathered current resident care methods, assessment tools, care-related records, and resident assessment procedures for this study. The care staff also offered professional knowledge for this project. The objectives of this study are the following: (1) building up a resident assessment system for LTC institutions with a classification and grading concept, (2) developing a RCTC system to automatically generate personalised care plans and weekly healthcare schedules for individual residents in the institution, (3) improving effectiveness in management of medical institution, healthcare quality [24, 25] for residents, and working efficiency of staff. 2. Classifying Residents by Types of Care Traditional pigeonholed resident placement constitutes a depersonalised care service and is deemed not socially acceptable [26]. Classifying residents into categories by their care needs would be useful for resident placement decisions and would reflect humanistic considerations. Residents would receive appropriate care service, which would be helpful for the improvement of care quality. The classification concept was based on the premise that the care staff should provide residents with the right care services at the proper time using the most economical means, which can be achieved through the use of reasonable assessment procedures for the classification of residents according to their care needs. Thus, the placement of residents into appropriate districts with proper care plans would help residents to live at their maximum potential with the acceptance of care services.

125 International Journal of Distributed Sensor Networks 3 RCTC system HIT system Nurse charting system Nursing care plan templates Demographic and medical history and health status User-defined assessments HIT extend Resident profile Assessment tool Mental, movement, eating, toilet, medical treatment Care plan Types of care, care time Weekly care schedule RCTC extend LTC system Reimbursement Bed allocations Staff assignment Maintaining proper staffing ratios Figure 1: Conceptual framework of the RCTC system. The RCTC system was designed mainly by referring to the TAI [15] (Typology of the Aged with Illustrations) scale s classification and grading concept. We modified some of the items of the TAI assessment instrument to better fit the Taiwanese population. Figure 1 shows the proposed conceptual framework of the RCTC system. The RCTC system as a whole includes a basic resident database, assessment module, care plan module, weekly care schedule module, and management module. Because the care services are different from those of the traditional HIT system according to the five functional status assessment aspects mental, movement, eating, toilet (urination and defecation functions), and medical treatment the RCTC system can automatically compute the care needs and form personalised care plans and weekly care schedules for residents. Thus, the care staff can implement more flexible care services for individual residents compared to traditional care plans. As for the weekly care schedule of the residents and the number of residents in specific categories in the institution, the LTC system can be implemented in accordance with the estimated resources, such as reimbursement, bed placement, staff assignment, and maintaining proper staffing ratios, to enable effective management and operational efficiency. The resident assessment procedure first evaluates the care needs and then classifies residents into categories of types of care. After assessment, three types of care and seven care grades are possible. Each element starts with mental status, which is mainly used to assess cognitive function, including verbal functions and language skills, concept formation and reasoning, perception, orientation, and attention. The functional levels are graded in light of cognitive disorders. In terms of movement, the assessment mainly considers the ability to complete daily activities independently, including personal hygiene, dressing and undressing, medication usage, and shopping. The worst situation is that the resident is unable to turn over in bed by himself/herself. In terms of eating, the assessment mainly considers the ability to perform actions related to eating, including if the patient can eat by himself/herself, if fixed dishware is needed, and food type.theworstsituationisthatthepatienthastobefed by nasogastric tube or receive nutrition through intravenous injection. In terms of toilet functions, the assessment mainly considers the situation related to defecation and maturations, including if the patient can go to the toilet by himself/herself, if a diaper is needed, and the loss of control of defecation and maturations. The worst situation is that a Foley catheter is needed. In terms of medical treatment, the assessment mainly considers the need for medical treatment, including regular medical examination, emergency medical treatment, medical nutrition management, and special medical measures received in the previous two weeks. Each assessment option has a score. The RCTC system determines the grade in accordance with the scores to demonstrate the severity of each assessed criterion. Then, the type of care is determined according to the grade assigned to each element. The types of care fall into total care, partial care, and no care. The cumulative score of all subjects is calculated to judge the required care time. The higher the score is, the better the situation is and the shorter the care time needed. 3. Resident-Oriented Assessment LTC institutions offer formal care services that provide living accommodations for residents who require on-site delivery of around-the-clock supervised care services, including professional health treatment and personal care services [27]. The concept of types of care or levels of care is used to classify residents on the basis of similar characteristics and to facilitate the delivery of appropriate care services [9]. Thus, a resident classification system should be resident-oriented, and a comprehensive assessment of resident characteristics related to health conditions and service requirements is assessed to reach a type of care decision. After classification by type of care, residents could be placed in an institutional facility that provides the needed care services. The early LTC resident assessment tools mostly focus on the individual assessment at a single level. The major levels are assessment of cognitive function, assessment of activity function, and assessment of behavioural function [28]. Cognitive function refers to mental status, including memory, attention, and comprehension or language skills, which can identify dementia or signify the declining situation

126 4 International Journal of Distributed Sensor Networks of patients with dementia in terms of cognitive function and dependence [29]. For example, the Mini-Mental State Examination (MMSE) is widely used to assess the cognitive function of the aged, while the Short Portable Mental State Questionnaire (SPMSQ) is used to assess mental status. Activity function assessment refers to testing and quantifying the ability to conduct functional movement. This assessment can be used to monitor the overall improvement situation of individual cases, including physical, mental, emotional and social functions. Therefore, it is an important reference to predict care needs and make care plans. The Barthel Index (BI) can be used to monitor the development situation in the rehabilitation field. Independence in Activities of Daily Living (ADL) is used to assess the movement independence of chronic disease patients and the aged in their daily lives. Instrumental Activities of Daily Living (IADL) is an important index to assess the independence of the aged. Functional Independence Measure (FIM) is used to assess the rehabilitation of multiple-disabled patients. Behavioural function assessment is mainly used to assess the behaviour problems of disabled patients, especially patients with brain damage. In 1980, the WHO proposed the concept of International Classification of Impairment, Disability and Handicap (ICIDH), which influenced the concept of health functional assessment and its application. Many studies have explored theinfluenceofhealthfunctionclassificationonltc[30]. Different health functional assessment tools were thus developed to assess LTC care needs. The most typical of these tools are the following. (1) The Residential Assessment Instrument (RAI) is composed of the Minimum Data Set (MDS), Resident Assessment Protocols (RAPs), and Specification. MDS contains all core items that must be assessed. RAPs are used to form nursing care plans [31], provide diagnostic logic, and help evaluators to confirm the care required by residents, in accordance with the care planning of individual cases [20]. (2) TAI is composed of classification and grading assessment scales, an assessment scale of the required care for the aged, degradation and aging process charts, and a summary table of each unit. TAI classifies the types of care and then grades them. The 12 types of care can be specifically divided into 6 scopes and 11 categories. The type of care, degradation and aging stage of care grades, and the care time needed in every stage are demonstrated by a diagram [32]. Results from previous study [26] suggested that assessment should be based on not only medical diagnosis but also impairment and disability for long-term care. Classification systems developed for LTC should use more sophisticated methods than those used in other care areas. One similar work [5] employed functional status, behavioural status, and medical diagnosis as variables for logistic function modelling to predict the prior probability of levels of care. A study [33] developed a classification system using functional level, disease category, risk factors, and health indices. Therefore, resident assessment must be multidimensional for LTC, and the assessment must be comprehensive enough to measure all elements for each resident. To evaluate the practical care needs of residents, mental, movement, eating, toilet, and medical treatment were the five main categories of variables used in this study. Moreover, the type of care classification for a resident was determined after a comprehensive assessment of the resident. The results of the resident s type of care can be helpful for the care of residents because the care service aspect is emphasised and curing disease is usually not a viable goal. Following classification by types of care, a resident can be placed in an institutional facility that manages the provision of substantial care services. Accordingly, residents can then be classified in terms of their required grade of care. 4. The Innovative Resident Care Process Currently, most LTC institutions in Taiwan can only provide package services instead of individual services for residents. Before the completion of the RCTC system, the original care process of residents was as shown in Figure 2(a). The care appraisers manually and subjectively conducted the assessment by various scales as required, such as the Barthel Index, independence index of ADL, MMSE, SPMSQ, and Behavioral Rating Inventory of Executive Function. For normal situations, the care appraiser should fill in the corresponding care plan according to the health conditions of the resident. For example, if the resident encountered one care problem, a corresponding care plan should be filled in. If the resident encountered multiple care problems, multiple care plans should be filled in. However, because LTC residents often encounter multiple care problems at the same time, the final care plan will be rather complex and difficult to maintain without the aid of a software system. For this reason, the care staff will usually care for all of the residents with the same careplan.thatis,asidefromspecialmedicalsituations,every resident will be treated under the same care plan, including eating,bathing,toilet,andmovement.iftherearenospecial circumstances, residents are not reassessed after admission. Figure 2(b) shows the RCTC system innovative care process, which is composed of four stages: resident assessment, care plan, care implementation, and regular assessment. The care appraisers could comprehensively assess residents with the aid of the RCTC system, including mental, movement, eating, toilet, and medical treatment. After a comprehensive assessment, the RCTC system will automatically determine thetypeofcareofeachresidentandpredicttherequiredcare time. It is helpful for care administrators in the institution to place residents into different districts according to the type of care and organise suitable manpower according to the overall predictedcaretime.asidefromtypeofcareandcaretime,the RCTC system can automatically form personalised care plans and weekly care schedules for each resident. Care staff in the institution can not only determine the health conditions of residents but also explain the current situation and care methods to residents relatives. If any amendment is needed or care emphasis is added by the resident s relatives, it can be recorded in the weekly care schedule. The overall weekly care schedule is based on different types of care, by which the care staff can implement their care work. An additional advantage of the RCTC system is that the staff can rapidly reassess whether there is any change in the resident s health condition. Moreover, the RCTC system will proactively notify

127 International Journal of Distributed Sensor Networks 5 (1) Resident assessment (2) Care plan (3) Care implementation (4) Regular assessment Assessment purposes? Assessment tools? Stationary care plan Package services (a) Before the implementing of RCTC (1) Resident assessment (2) Care plan (3) Care implementation Mental Movement Eating Toilet Medical treatment Personalised care plan (4) Regular assessment Automatic notification Types of care grading Customized care weekly schedule (b) After the implementing of RCTC Figure 2: Comparison of resident assessment processes. the staff to conduct regular assessments every three months to make new care plans and weekly care schedules. Additionally, the innovative resident assessment process assesses residents with the aid of the system instead of the written assessment method used in previous procedures. After the completion of the assessment, the care plan is formed automatically, which not only improves the situation of filling in many care plan forms but also increases communication methods with the relatives of residents. Moreover, the care staff can better understand the care content so that theycanmakefewermistakes.therctcsystemcanprovide personalised care plans to improve upon the drawbacks of package services. 5. The Architecture of RCTC System The methodological approach of RCTC system development is illustrated in Table 1 andhasthreestages:inception,objectoriented analysis and design, and implementation. Each stage is defined by work content, method, and output. The software development process includes eliciting information on systems and needs, mapping care processes into the system, and soliciting ideas for the software from institutional care staff. We developed a pilot run system for an institution and incorporated the resident assessment processes, assessment questions, care services, and required care time. We employed the ontology methodology to enable interdisciplinary team members to elicit requirements semantically and derive implementation models that meet those requirements, and the UML methodology was used to describe the static and dynamic structure of the RCTC system to generate conceptual models, dynamic models, and user interfaces. For knowledge-oriented systems, the ontology can formally represent knowledgeasasetofconceptswithinadomain,aswellasthe relationships among those concepts. The ontology s vocabulary and taxonomy abilities provide a conceptual framework for sharing, analysing, and retrieving data in a specific domain. Therefore, ontology allows system developers to focus on the application domain structure rather than implementation details and allows the system developers to reuse and share application domain knowledge across different software platforms and programming languages. In this study, ontology helped care administrators, care staff, and system developers to utilise appropriate pieces of knowledge when facing complex care problems and situations. Figure3 illustrates a schematic ontology class hierarchy structure for the RCTC ontology. The RCTC ontology presents a formalised description of concepts of resident classification by type of care. The information presented in this ontology was extracted from major resident charts, care plan templates, care guidelines, and administrative information, and it was discussed with the care administrators of the institution. The ontology class hierarchy structure was constructed on the basis of the resident assessment process, and it represents the five major subcategories, under the category of RCTC ontology. The five subcategories are resident management, care management, nutrition management, social worker management, and administrative management. There are three major categories under the resident management subontology: resident demographics, relative information, and admission notes. This subontology includes all basic concepts and health information that are relevant to residents in the institution. The care management includes major assessment tools, care plans, and weekly care schedules for the classification of residents by type of care. The nutrition management, social work management, and administrative management subontologies include information related to management of daily activities of the institution. The proposed RCTC system is an assistance system for assessing institutional residents, as well as for generating personalised care plans and weekly care schedules. Our systemisbasedontheworldwideweb(www)technique because the WWW is a standardised, cross-platform environment. WWW applications can be effective in creating virtual working platforms, which provide easy ways to collaborate and communicate with coworkers. Therefore, it is appropriate

128 6 International Journal of Distributed Sensor Networks RCTC ontology Is a Is a Is a Is a Is a Resident management Care management Nutrition management Social worker management Part of Part of Part of Part of Part of Administrative management Part of Resident demographic Relatives information Admission note Assessment tools Care plan Part of Part of Part of Part of Part of Part of Part of Part of Part of Part of Part of Part of Part of Part of Part of Part of Part of Part of Part of Part of Part of Part of Part of Record at Record at Record at Record at Record at Record at Part of Part of Part of Part of Part of Part of Part of Part of Part of Name Gender Birthday Blood type ID Height Weight Main symptom Education degree Language Marital status Living conditions before admission Income source Communication methods Current living place Admission manner Institution district Floor Room no. Bed no. Admission date Discharge date Charge Name of relatives Relationship Telephone Address Hospital Sent date Return date TAI scale Japanese careneeds certification scale Barthel score MDS Body functions Life functions Cognitive function Mental mobility impairments Social life adaptation Part of Part of Part of Part of Part of Part of Part of Part of Part of Part of Toilet Eating Movement Mental Medical treatment Body functions Life functions Cognitive functions Mental mobility impairments Social life adaptation Care weekly schedule Part of Part of Part of Weekday Weekend Special arrangement Figure 3: The RCTC ontology.

129 International Journal of Distributed Sensor Networks 7 User interface Care appraiser Administrator Assessment module set Management module set Resident management Resident assessment System administration Scale tools management Care plan generation Care weekly schedule arrangement Care staff scheduling Resource allocation Institutional database Assessment tools table Types-of-care and grading table Care weekly schedule table Resident demographic and relatives table Care plan table Other institution administration related tables Figure 4: RCTC system architecture. Table 1: The RCTC system development stages. Content Method Output Stage 1: inception Data collection (1) Resident profile (2) Assessment tools (3) Existing charting forms (4) Resident assessment/care process Literature review Conducting interview Requirement analysis Research scopes Design objectives System requirements Stage 2: object-oriented analysis and design Conceptual modeling Dynamic modeling User interface design Ontology modeling UML methodology Conceptual models Dynamic models User interfaces Stage 3: implementation Database design Modular design System implementation Object-oriented programming (1) MS-SQL server (2) IIS server (3) Visual studio RCTC system to use the WWW to develop an assistance system for resident management. Inthisstudy,weusedMicrosoftSQLServer2008tobuild an institutional database based on the resident assessment tools and added tables at the care administrator s request. We used the C# programming language to code the whole system. The platform is available for the nursing staff, care staff, and institution administrator to obtain the required residents information. As shown in Figure 4, the RCTC system architecture includes an institutional database and two module sets. The module sets are described in detail below Assessment Module Set. This module set is primarily provided for user interfaces. It contains resident management, resident assessment, care plan generation, and weekly care schedule arrangement modules. The resident management module is used to manage residents in the institution. The resident assessment module determines the type of care of each resident and calculates care grade by classification and grading scales. The care plan generation module provides suggestions about care services and what care aspects should be emphasised. The weekly care schedule arrangement module helps care staff deliver better care service for the resident. Depending on the needs of the care staff, the interface for the assessment scales is customisable in that it displays adjustable items and uses a graphical interface to highlight the important changes after the assessment. It is therefore able to assist the care staff in understanding the changes in the health condition of the resident in the complex relationships among the five care aspects. Moreover, the resident s information,

130 8 International Journal of Distributed Sensor Networks including their demographic personal details, discharged disease abstract, and assessment results, is maintained by the interfaces of this module set Management Module Set. This module set is mainly used by institution administrators to set up the system permissions, create and modify assessment tools, and estimate the overall resources needed for the institution. This module set is composed of system administration, scale tools management, care staff scheduling, and resource allocation modules. The system administration module offers the overall system functions for institution management. The scale tools management module is designed to maintain the assessment scales, including scale items, item weights, and algorithms for the estimation of care time. Because there are still many uncertainties in Taiwan s regulations and many factors involved in the assessment of residents, the use of assessment tools may be changed to comply with Taiwanese laws, so we retained flexibility in this module. Currently, the RCTC system mainly refers to the TAI scale, Japanese care needs certification scale, and Barthel score. The care staff scheduling and resource allocation modules can generate weekly care plans for individual residents and calculate the demand for manpower and resources, which can be valuable references for the institution administrator to manage and monitor resource consumption. As shown in Figure 4, the resident assessment module is the most critical module of the RCTC system. The resident assessment procedure first evaluates care needs with respect to five functional status assessment attributes (with scores ranging from one to one hundred) of the resident. Then, it classifies residents into types of care categories (with grades ranging from one to seven). That is, three types of care and seven care grades are defined after evaluating by overall care time. The sequence diagram in Figure 5 explains how groups of objects work together to achieve the resident assessment process. This sequence diagram contains one main object on the appraiser s end. From this starting component, the care appraiser will receive the assessment demands from the RCTC server. The care appraiser can select the proper assessment tool to evaluate the health conditions of the resident, and it can also check the resident s record to see the resident s entire history. That history is retrieved from perspective tables in the institutional database. When any assessment is completed, the system will automatically compute the care needs and generate a personalised care plan and weekly care schedule for each resident. This procedure will enable care staff to deliver the required care services to their residents. 6. Design of the Assessment Algorithm for Care Time RCS can assess the estimated case time of residents, which employs decision trees as its core. The design process is described in detail below Step 1: Conceptual Design of Decision Trees. The structure of the decision tree is similar to a binary tree, which has a root node; each node has a left and right child node. The decision Table 2: Decision tree table. Fields Fields name Data type Note Code Node code Varchar(20) Key Name Node name Varchar(20) Parent Root node code Varchar(20) LR Left or right Varchar(1) Node type Node type Varchar(1) Node code Type code Varchar(20) Lrule Leftrule Varchar(50) Rrule Rightrule Varchar(50) tree s node not only stores numeric values, but also stores text. Ifthedecisiontree snodeisanumericvalue,thenthevalue of the left node must be smaller than that of the right node. Wedesignedthetreestructureasdatabasefieldstorecordthe relationship between nodes Step 2: Decision Tree Design. We employed a depthfirst traversal algorithm to implement the recursive call of thedecisiontree.asshowninfigure 5, theback-endweb server retrieves a decision tree structure from the database and generates HTML tag <ul> and <li> to include the tree s structure. When a user starts to use a decision tree, the system will first create the current root node and its left and right child nodes according the node value. Then the decision subtree structure was included in the <ul> and <li>xml tag. The tree generation process will end when the decision tree build is complete. Finally, a j Query organization chart plugin was used to generate a visualized decision tree Step 3: Connecting to the Assessment Results. After the completion of the decision tree, we need to connect the assessment results with the decision tree. We first added four fields including node type, type code, left side rules, and right side rails into a decision tree table (Table 2). The tree traversal direction depended on the current subtree s assessment result. As shown in Figures 6 and 7, thenode s typeandtypecodewereusedtoidentifythescalesubjectand assessment dimensions and then traversed the left side rule or rightsiderulebytheruleofthenode.theassessmentresult is determined once the expected traversal was made Step 4: Decision Tree Judgment Logic. Using decision tree judgments logic to determine the case time of results output. The judging process will call their own, when the node type is results outputcaretime.ifcannotbefoundnodewhobehalf node type or node location, output Error. As shown in Algorithm 1, Cannot be found node occurs, set of judgments of the decision tree set error, need to be corrected. 7. System Implementation In this study, the RCTC system provides an institutional database that stores relevant data such as the resident s demographic data, disease history, assessment results, specific care needs, and type of care and the care appraiser s interactions with the caregivers. These stored data help the care staff

131 International Journal of Distributed Sensor Networks 9 Assessment Appraiser interface Choose resident Demographic Query resident Enter assessment Display interface resident information Fill in assessment scale Sent assessment results Display assessment scale Display score Assessemnt scale Assessment tool Assessment Assessment aspects questions Component Question Option Question s options Calculate score Care plan Assessment Save assessment date, result, care plan Save answers of questions and options Assessment details Assessment decision trees Grades Types-of-care Care weekly scheduler Display assessment result Determine types-of-care, grade, care time Assessment result Care time Care grades Types-of-care Care plan Question and option Calculate care time Judge care grades Care grade Types-of-care Figure 5: Sequence diagram for the resident assessment process. <ul> <li>root node</li> </ul> Insert Insert <ul> <li>left node</li> <li>right node</li> </ul> Insert <ul> <li>left node</li> <li>right node</li> </ul> <ul> <li>left node</li> <li>right node</li> </ul> Query the child nodes of the left point generated structure of <ul> <li>. Query the child nodes of the right point generated structure of <ul> <li>. Figure 6: Decision tree building process. Node type-question Need beside guardianship Eating Need someone part to assist Need to fully assist Life function 31.2< > Node type-dimensions 19.1 Node type-end Figure 7: Node type determination.

132 10 International Journal of Distributed Sensor Networks Function RecurisiveTree (string parentnode, string LeftOrRight) If (readnote) { Switch nodetype Case Question: readevaluate if (Judgment value at LeftRule) RecursiveTree (node, left) else if (Judgment value at RightRule) RecursiveTree (node, right) else response error break; Case Dimensions: readevaluate if (Judgment value LeftRule) RecursiveTree (node, left) else if (Judgment value RightRule) RecursiveTree (node, right) else response error break; Case Result response root : name (code) = score break; default: response error break; } else { response error } end Algorithm 1: Algorithm for decision tree. to assess the care needs based on their residents health conditions, care plan arrangement, and wellness activities, as wellasthestatusofcaredecisions,careactionsundertaken, and other relevant information that could aid in the provision of proper care. The RCTC system provides information on residents care type, care method, estimated care time, and personalisedcareplan,aswellastheweeklycareschedule for residents in each district. The care appraisers can add or modify the residents basic data, hospitalisation records, and other information. To retain the flexible extension of the system, the RCTC system follows the modular design principle, enabling the institution care appraisers to modify and maintain assessment tools and care plan templates, thereby enhancing its usability. We show several screen layouts, including the system administrative environment, scale tools layout, assessment screen, personalised care plan, and estimation of institutional overall resource needs. Figure 8 depicts the function of assessing resident s care needs. Care appraisers assess the resident by asking questions relating to the five functional status assessment aspects, and they choose the proper care plan corresponding to the selected option. The personalised care plan offers suggestions on each aspect of the assessment results. The assessment module automatically computes the grades of thecareneedsforeachaspect.carestaffcanfollowthecontentofthecareplantoimplementtheircarework.figure 9 shows the assessment result determined by the system, which shows the required care time for nine care-need components, grades resident care needs as level seven, and recommends thetypeofcareas heavydegreeofcareneeds. Thecare appraiser and care staff can quickly review the overall situation and care needs of the resident by using the information shown as a report and a radar chart. According to the care plan, the weekly care schedule presents the care methods and care notes of the resident in different timetables. Care staff of the institution should record the additional requirements provided by family members of the residents in the sheet during their communication. The desire to balance the individual resident s care needs with the constraint of manpower limitations led us to develop our resident classification mechanism. The assessment/classification sequence is intended to be used by care appraisers, who are involved in resident assessment and care service delivery. For institution management issues,

133 International Journal of Distributed Sensor Networks 11 Figure 8: Assessment of a resident s care needs. Figure 9: Resident assessment result. management data were collected on the selected residents not only through assessment/classification by the care appraisers, who may or may not be independent of the institution, but also by service caregivers within institutions. Because the primary resource used by LTC institutions is manpower, classification/placement is intended to determine how much care manpower should be put into each resident based on overall estimated care time. The assessment/classification provides information on residents who are under specific types of care with different care-level grades, and the classification/placement information can be used for reimbursement, bed assignments, and decisions about staff assignment, recruitment, and staffing ratios, to achieve effective management and operational efficiency. 8. Conclusions The RCTC system presented in this study provides care practitioners in LTC institutions with a comprehensive assessment system that includes classification and grading concepts. The RCTC conceptual model follows an assessment/classification/placement sequence that can be used for resident assessment, classification, and placement decisions for LTC residents. The major functions of the RCTC system are to assess LTC residents based on five functional status assessment aspects, classify residents into different types of care, and grade the care needs in terms of severity. Moreover, this RCTC system offers personalised care plans and weekly care schedules, which make the care methods more flexible andofhigherqualitythantraditional packageservices. During our research, we proposed that the estimation of aresident soverallcaretimecanbeemployedtoestimate the resources the institution will require, including care manpower and consumptive materials. Currently, the system is under a pilot run with our cooperative institution. We hope to offer an assessment framework sample of an LTC institution in Taiwan that can be widely used in the future. We also plan to cooperate with the hospital to implement the system outside of LTC institutions and use it to realise seamless service in hospitals. Conflict of Interests The authors declare that there is no conflict of interests regarding the publication of this paper. Acknowledgments This research was partially supported by National Science Council of Taiwan, under Research Projects NSC E and NSC H CC3 and by the MSIP (Ministry of Science, ICT and Future Planning), Korea, under the ITRC (Information Technology Research Center) Support Program NIPA-2013-H supervised by the NIPA (National IT Industry Promotion Agency). References [1] D. W. Bates, M. Cohen, L. L. Leape, J. M. Overhage, M. M. Shabot, and T. Sheridan, Reducing the frequency of errors in medicine using information technology, Journal of the American Medical Informatics Association,vol.8,no.4,pp , [2] K. S. Bay, P. Leatt, and S. M. Stinson, A patient-classification system for long-term care, Medical Care,vol.20,no.5,pp , [3] B. A. Hamilton, Evaluation design of the business case of health information technology in long-term care, 2006, aspe.hhs.gov/daltcp/reports/2006/bcfinal.htm. [4] E. Tomasi, L. A. Facchini, and M. F. Maia, Health information technology in primary health care in developing countries: a literature review, Bulletin of the World Health Organization,vol. 82,no.11,pp ,2004. [5] L.J.CavaiolaandJ.P.Young, Anintegratedsystemforpatient assessment and classification and nurse staff allocation for long term care facilities, Health Services Research, vol.15,no.3,pp , 1980.

134 12 International Journal of Distributed Sensor Networks [6] B.Cherry,M.Carter,D.Owen,andC.Lockhart, Factorsaffecting electronic health record adoption in long-term care facilities, Journal for Healthcare Quality, vol.30,no.2,pp.37 47, [7] B. J. Cherry, E. W. Ford, and L. T. Peterson, Experiences with electronic health records: Early adopters in long-term care facilities, Health Care Management Review, vol. 36, no. 3, pp , [8] T. Wang and S. Biedermann, Adoption and utilization of electronic health record systems by long-term care facilities in Texas, Perspectives in Health Information Management, vol. 9, article 1g, [9] M. Brahami, B. Atmani, and N. Matta, Dynamic knowledge mapping guided by data mining: application on healthcare, Journal of Information Processing Systems,vol.9,no.1,pp.1 30, [10] P. Densen and E. Jones, The patient classification for long term care developed by four research groups in the United States, Medical Care,vol.14,no.5,pp ,1976. [11] S. Matsuda and M. Yamamoto, Long-term care insurance and integrated care for the aged in Japan, International Journal of Integrated Care, vol. 1, article e28, [12] A. Melanie, S. Ralf, S. Alexander, S. Susanne, and W. Sarah, The German social long-term care insurance: structure and reform options, IZA Discussion Paper 2625, 2007, [13] J. Okochi, S. Utsunomiya, and T. Takahashi, Health measurement using the ICF: test-retest reliability study of ICF codes and qualifiers in geriatric care, Health and Quality of Life Outcomes, vol. 3, article 46, [14] C. Hawes, J. N. Morris, C. D. Phillips, B. E. Fries, K. Murphy, and V. Mor, Development of the nursing home resident assessment instrument in the USA, Age and Ageing, vol. 26, supplement 2, pp.19 25,1997. [15] L. Martin, B. E. Fries, J. P. Hirdes, and M. James, Using the RUG-III classification system for understanding the resource intensity of persons with intellectual disability residing in nursing homes, Journal of Intellectual Disabilities, vol. 15, no. 2,pp ,2011. [16] Long-term care insurance in Japan, english/topics/elderly/care/index.html. [17] J. Okochi, T. Takahashi, K. Takamuku, S. Matsuda, and Y. Takagi, Reliability of a geriatric assessment instrument with illustrations, Geriatrics & Gerontology International,vol.5,no. 1, pp , [18] J. Okochi and T. Takahashi, The effect of medical needs to the disability based patient classification system a critical review of disability based patient classification, BMC Health Services Research,vol.7,articleA7,2007. [19] L. Berger, Information technology feature: the IT payback, McKnight s Long-Term Care News,pp.34 37,2006. [20] F. Landi, E. Tua, G. Onder et al., Minimum data set for home care: a valid instrument to assess frail older people living in the community, Medical Care,vol.38,no.12,pp ,2000. [21] V. Mor, A comprehensive clinical assessment tool to inform policy and practice: applications of the minimum data set, Medical care,vol.42,no.4,pp.50 59,2004. [22] J. M. Schols, H. F. Crebolder, and C. van Weel, Nursing home and nursing home physician: the Dutch experience, Journal of the American Medical Directors Association, vol.5,no.3,pp , [23] P. Nadash and Y. C. Shih, Introducing social insurance for longterm care in Taiwan: key issues, International Journal of Social Welfare,vol.22,no.1,pp.69 79,2012. [24] N. Ibrahim, M. Mohammad, and V. Alagar, Publishing and discovering context-dependent services, Human-Centric Computing and Information Sciences, vol. 3, article1, [25] R. Sumathi and M. G. Srinivas, A survey of QoS based routing protocols for wireless sensor networks, Journal of Information Processing Systems,vol.8,no.4,pp ,2012. [26] A. Følstad, K. Hornbæk, and P. Ulleberg, Social design feedback: evaluations with users in online ad-hoc groups, Human- Centric Computing and Information Sciences, vol. 3, article 18, [27] Health Canada, long-term cacilities-based care, [28] S. Hedrick, J. Papsidero, and C. Waynard, Analysis of alternative definitions of function in studies of long term care. Project to Analyze Existing Long Term Care Data, edited by, I.W. Scanlon, Washington, DC, USA, The Urban Institute, [29] L. K. Langley, Cognitive assessment of older adult, in Assessing Older Persons: Measures, Meaning, and Practical Applications, R. L. Kane and R. A. Kane, Eds., pp , Oxford University, New York, NY, USA, [30] OECD, Long-term care for older people, The OECD Health Project,OECD,Paris,France,2005. [31] D. Liu and N. G. Castle, Health information technology in nursing homes, Journal of Applied Gerontology, vol. 28, no. 1, pp.38 58,2009. [32]T.Takahashi,J.Okochi,K.Takamuku,andS.Matsuda, The introduction of typology of the aged with illustrations, Casemix Quarterly,vol.3,no.1,pp.3 14,2001. [33] K.S.Bay,P.Leatt,andS.M.Stinson, Cross-validationofapatient classification procedure: an application of the U method, Medical Care,vol.21,no.1,pp.31 47,1983.

135 Hindawi Publishing Corporation International Journal of Distributed Sensor Networks Volume 2014, Article ID , 10 pages Research Article Home Appliance Management System for Monitoring Digitized Devices Using Cloud Computing Technology in Ubiquitous Sensor Network Environment Yun Cui, 1 Myoungjin Kim, 1 Yi Gu, 1 Jong-jin Jung, 2 and Hanku Lee 3 1 DepartmentofInternetandMultimediaEngineering,KonkukUniversity,Seoul143701,RepublicofKorea 2 Digital Media Research Center, Korea Electronic Technology Institute, Seoul , Republic of Korea 3 Center for Social Media Cloud Computing, Konkuk University, Seoul , Republic of Korea Correspondence should be addressed to Hanku Lee; hlee@konkuk.ac.kr Received 30 August 2013; Revised 28 December 2013; Accepted 4 January 2014; Published 18 February 2014 Academic Editor: Young-Sik Jeong Copyright 2014 Yun Cui et al. This is an open access article distributed under the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited. The number of service techniques available for digitized home appliances is rapidly increasing as a result of various advances in digital technology. Users can now easily control and monitor home appliances via sensor networks formed among home appliances in ubiquitous environments. However, home appliances generate such large amounts of metadata about their status every month that in order to provide home appliance monitoring services to users, an approach that is able to store, analyze, and process these large amounts of metadata is needed. We propose a system that uses UPnP to collect metadata from home appliances and cloud computing technology to store and process the metadata collected from ubiquitous sensor network environments. Our proposed system utilizes a home gateway and is designed and implemented using UPnP technology to search for and collect device features and service information. It also provides a function for transmitting the metadata from the home appliances to a cloud-based data server that uses Hadoop-based technology to store and process the metadata collected by a home appliance monitoring service. 1. Introduction With the unfolding of the information age, home appliances, including various types of digital devices, became capable of communicating with one another through the Internet. Many smart devices are now being equipped with new features. For example, users can now control their smart TVs via their mobile phones and watch the media contents stored on a PC on their smart TVs. A home network enables all connectable devices to be connected in a single network inside of the home so that they can communicate with each other. In the new millennium, home networks have evolved in a consistent way. Simultaneously, the amounts of log data and metadata used for monitoring the status of home appliances have increased significantly. With the ultimate goal of realizing a way to cope with these changes, many researchers have carried out studies focusing on how to process and analyze the huge amounts of data generated by sensors deployed throughout smart homes [1 5]. Studies indicate that the metadata from devices deployed in home networks are generated by the sensors integrated inside the home appliances. These metadata should be collected, analyzed, and processed in a ubiquitous sensor network environment. To facilitate this, a relay device that enables outside communications with the home network is needed. A homegateway is such a device [6 9]. Home appliances can be monitored and controlled via a network deployed inside the home. However, it is impossible for the home network to communicate with an external network outside ofthehomeasis.homegatewaysareusedtoovercome this limitation of home networks. Most home gateways use diverse technological elements to collect metadata and perform associated processes for devices deployed in a home network. OSGi is used to support these diverse technological elements by processing and analyzing the metadata they collect [10]. In addition, OSGi is convenient because it provides the functions necessary for the home network on

136 2 International Journal of Distributed Sensor Networks a single platform from its bundles and allows developers to implement the functions they require. As the large amounts of metadata generated by home appliances deployed in home networks increase significantly, home gateways that use OSGi are encountering problems such as insufficient computing power and shortage of data storage space. Researchers are now therefore looking to utilize cloud computing technology to handle the huge amounts of data generated by the appliances. Cloud computing is a state-of-the-art technology that has been receiving a lot of attention in the computing field. In cloud computing, distributed processing of massive amounts of data can be achieved using spare resources, which results in savings, both economically and temporally. Furthermore, cloud computing brings technological advances [2, 11 14]. Cloud computing hasbecomeaneffectivesolutionforprocessingmassivemetadata generated by home appliances connected via home gateways. Combining the various elements of technology mentioned above, this paper proposes a cloud-based system for monitoring and controlling home appliances. Our proposed cloud-based home digital devices monitoring and control system consists of two major parts, the home network and a data server. It also includes a home gateway and a simulator. The home gateway is responsible for collecting device features and the service capabilities of the individual devices deployed in the home network and sendingthedatacollectedtothedataserver.thesimulator provides information about the home appliances. The home gateway and the device simulator exchange data with each other using the Cling library. The Cling library, which supports the Android mobile OS, is a Java library of functions that implements the UPnP service. It also has a function that enables an external network to access a home network. This function makes the Cling library a good solution for achieving communication between the home gateway and the data server. In our proposed system, the home gateway does not establish the network using OSGi, unlike the approach proposed in [7], in which OSGi is used to set up the network. While a gateway that uses OSGi provides a number of useful functions to collect and manage data from devices residing in the home network, the objective of the home gateway proposed in this paper is to play the simple role of a bridge between the home network and adataserverinthecloud.asaresult,usingthecling library, we designed and implemented a function that enables communication between a UPnP-capable module and an external network in order to collect data from the electronic devices deployed in the home network and send them to the external server. In our proposed system, the data server stores and maintains the metadata generated by the home appliances. In addition, the data server provides a device monitoring service to the users. Metadata is stored in the data server using the Hadoop Distributed File System (HDFS) [15], which ensures scalability and security, even if the size of the data increases significantly. HDFS is a Hadoop-based file system that supports massive data processing effectively. It stores data with a scalable and distributed structure and maintains a copy of the data to insure against any occurrence ofdataloss.tomeetusers requirementsforprocessingand extracting the data stored in HDFS, we designed a progress module using MapReduce in the data server. The data server presents the data to the users via HTTP after extracting the required data using MapReduce [16]. In this process, if the user wants to make a change to a device s state, the data server receives the change request from the user and extracts specific commands for the designated device. Upon extracting the commands, the data server then sends them to the home gateway so that the commands can be executed. Our proposed system facilitates the monitoring and control of a number of home appliances by cooperation of the data server and the home gateway [17]. The remainder of this paper is organized as follows. Section 2 discusses related studies, home networks, smart homes, and cloud computing. Section 3 gives a detailed explanation about the system architecture designed by utilizing the results from previous studies. Section 4 highlights the data flow that occurs during the data communication process. Section 5 presents the results obtained from our system implementedonthebasisofthedesignedarchitecture.finally, Section6 concludes this paper. 2. Related Work Many researchers have proposed approaches for controlling digital devices and using their service functions outside of thehomenetwork[1, 6, 18]. In [7], Kang et al. implemented aupnpavarchitecturalmultimediasystemusingahome gateway supported by the OSGi platform to provide internal and external multimedia services. In their system, they used the UPnP bundle of OSGi to provide users with an external multimedia streaming service. The multimedia services are provided by a multimedia sharing system based on the UPnP AV Architecture [19]. Although the system was implemented using OSGi, due to congested multimedia content, the system could not support a large amount of multimedia data storing and management approaches. In [20], Cui et al. proposed a UPnP-based multimedia sharing system that utilizes cloud computing. This system provides multimedia content storing and management functions via cloud computing, which is used to transcode the multimedia content to provide a personalized multimedia streaming service. In [14],Kim et al.developed a multimedia content transcoding module using MapReduce and HDFS. Díaz-Sánchez et al. [12] also proposed media cloud, a middleware for set-top boxes that classifies, searches for, and delivers multimedia content inside home networks and across any cloud that interoperates with UPnP and DLNA. Media cloud provides an easy to manage, cost-effective solution for using cloud computing to share content among federated home networks. It also allows devices from different home networks tocommunicateasiftheyareonthesamelocalnetwork. A similar trend is taking place with sensors. With the development of small sensor nodes that can be connected to a network, research on sensor networks, which function by connecting several sensor nodes to each other, is underway [21]. The research is being conducted in areas such as home security, healthcare, health management, and environmental monitoring. In home networks that use DLNA and

137 International Journal of Distributed Sensor Networks 3 ECHONET, the information that the sensors acquire can be used extensively by various devices. In [22], Jeong et al. implemented a large-scale middleware for ubiquitous sensor networks (called Lamses) that provides a USN middleware mechanism to focus on network processing. Lamses supports various heterogeneous sensor networks and mobile RFID. Unfortunately, RFID and other sensed devices have security vulnerabilities and generate a lot of data communication traffic. Hence, many researchers, such as Gao and Xiao [23] and Chen et al. [24], are currently researching core network issues and security problems. A system that utilizes a ubiquitous data source server (UDSS) a sensor device for home networks for use by sensors in the home, has also been proposed [25]. In the proposed system, by connecting a sensor to an ordinary home appliance and using the values acquired by the sensor, control of the home appliance is made possible. However, to achieve cooperative behavior like this, configuring that behavior directly on a PC connected to the sensor device is necessary. Configuration of a behavior cannot be performed from outside of the home, which is very inconvenient [25]. In [11], Xu et al. proposed a cloud-based framework for enabling smart home monitoring services. The system architecture resolves several issues including certification, media capture, Data-Cloud storage, and NAT traversal. The system offers remote control for surveillance and automatic sensor-control monitoring. In addition, it utilizes elastic storage and intelligent processing with the help of cloud computing. We implemented a prototype of this system to verify the utility of this architecture. In [26], Wei et al. proposed a cloud architecture based on an existing cloud system to support smart homes. In their proposed architecture, clouds can provide more humanized services for digital home appliances in smart homes, and smart home nodes can form peer-to-peer networks and publish, lookup, and use services in the cloud. They designed a mechanism that connects the smart home nodes and a cloud server to form a peer-to-peer network and introduced it as a webserviceinthecloud,sothatsmarthomeuserscannot only be suppliers but also consumers of cloud services. In line with the related works outlined above, we propose a cloud-based home appliance monitoring and control system. Utilizing the strong points of previous research results, we design and implement a home gateway to collect metadata from home appliances and transmit them to a data server and a cloud-based data server to store and manage the metadata generated by home appliances. 3. Proposed System Architecture The objective of our proposed system is to store and maintain status information about home appliances (such as refrigerators, air conditioners, humidifiers, and boilers) sent by a home gateway using a cloud-based data server. The data server stores and maintains metadata generated by home appliances on HDFS and also provides monitoring services for users by transmitting metadata to their smartphones when they request data for a specific device. Metadata associated with home appliances are collected by the home gateway in XML format. After collecting the metadata, they are sent by the home gateway to the data server through a series of processes. The home gateway is capable of collecting information such as device features, service capabilities, and status information about devices deployed in the home network using UPnP. Figure 1 depicts the structure of our proposed system. The proposed system can be divided into two parts according to the component dealing with the data: the home network part, which monitors the status of devices and collects metadata and transmits them to the data server, and the cloud-based data server, which manages information received from the home network in a coordinated fashion. The home gateway is the most important part of the home network. It can communicate with one or more home appliances in the home network via a router in both wireless and wired fashion. In a home network connected via a router, the home gateway uses the UPnP function to discover and gather information on device features, service capabilities, and the status of the connected home appliances. The home gateway plays two important roles in the home network. First, it detects and collects information on device features and service capabilities advertised by the electronic devices. Second, it communicates with the cloud-based data server and sends metadata to it. In other words, the home gateway functions as a bridge between the home network and the data server across the boundary of the network. The data center resides in cloud storage and maintains the metadata sent by the home gateway. It is designed to analyze and store metadata about home appliances deployed in the home network. Technically, the data center is implemented using MapReduce and HDFS. The metadata stored in the data center are used to provide a monitoring and control service to the users. End users can remotely check the status of any desired devices using their own smartphones. Further, in conjunction with the data server, they can also control the services with which the devices are equipped. The individual modules comprising the system architecture are described in more detail below Home Gateway. Inthehomenetworkestablishedusing the UPnP function, the home gateway detects and gathers information about home appliances and their service features. As a result, UPnP is considered to be a very powerful technology. However, connecting a UPnP-based home network to an external network is a very challenging task. For this reason, we propose to make the home gateway responsible for passing the data collected from the devices residing in the home network to the external data server. This gateway function can be realized by complicated cooperation among a number of modules. The modules that have to be implemented in the home gateway are as follows: network bridge module (NBM), device metadata parsing module (DMPM), device subscription function (DSF), action transmission module (ATM), and device registering function (DRF). Figure 2 illustrates the internal structure of a home gateway. DRFisabasicfunctionprovidedbyUPnPthatdiscovers and collects device feature information advertised by the

138 4 International Journal of Distributed Sensor Networks Smart device Data server Home Home network Home gateway Router Refrigerator Air conditioner Humidifier Boiler PC TV Printer Figure 1: Architecture of our proposed system. Network bridge module Home gateway Data temporary memory Device subscription module Action transmission module Device discovery and registering module Device metadata parsing module HTTP/HTTPU UDP TCP Figure 2: Internal structure of a home gateway. devices in a home network. To accomplish this, DRF uses the simple service discovery protocol (SSDP), a network protocol for advertising and discovering the services and device features of individual electronic devices. SSDP is textbased and establishes a network among devices residing in a home network using HTTPU. Consequently, electronic devices use SSDP to advertise their features and service information to other devices in a home network. The home gateway uses SSDP in the DRF to collect metadata from the devices. In collecting this information from the devices, SSDP adopts amulticastmechanismsothatitalwaysdiscoversthedatavia port The discovered data are stored in temporary DRF storageforusebydsfandatminthefuture. The information retrieved by DRF is basic information about electronic devices, such as devicetype, friendlyname, model description, model name, model number, UDN, and model URL. On the basis of the information about the device, DSF can request more detailed information associated with a specific device. When the request is made, DSF obtains service metadata in XML format. DSF accesses the URL of a specific device via the subscription command provided by UPnP. It can fetch the service names, which can be supplied by the device in XML, and can also get information on the current status of the device. In order to achieve these functions, DRF is implemented by GENA, which provides a status alarm function based on HTTP. Thus, DSF retrieves information from the devices through GENA and stores a device s status information in a temporary space. From the information acquired, we can check the command codes to control the service. These command codes are used by ATM. ATMusesdatagivenbyDRSandDSFtopasscommand codes to electronic devices and control device services. ATM controls the service features of devices in two ways. First,

139 International Journal of Distributed Sensor Networks 5 Data server Monitoring service module Mapreduce Network bridge module Metadata extraction module Hadoop distributed file system Resource virtualization CPUs Storages Networks Figure 3: Internal structure of the cloud-based data server. ATM can access the devices using information provided by DRF, such as URL, UDN, and model name. Second, ATM can use the command codes sent by NBM to control the devices. ATM sends the commands through the simple object access protocol (SOAP), a protocol that facilitates the exchange of XML messages over HTTP, HTTPS, and SMTP. SOAP is the underlying protocol used to deliver basic messages in web services. There are several types of message patterns in SOAP. It is designed with a design pattern combining header and body, which is implemented in XML. Thus, SOAP is the most suitable protocol for delivering data and commands over a HTTP-based environment. SOAP is also used as the primary command-delivery mechanism in UPnP. The home gateway facilitates the collection of device data in a home network and control of the devices. In addition, the home gateway has a function for communicating with the cloud-based data server. NBM, in the home gateway, establishes a connection with the data server and transfers information about device features and service capabilities to the data server through the connection. Information such as device metadata and service metadata is sent to the NBM residinginthedataserverviathenbminthehomegateway. To make a connection between the home network and the external network, NBMs are required to be implemented in both the home gateway and the data server. In this case, the communication is carried out via HTTP and information is exchanged as XML-formatted data. When a user sends commands to a specific device deployed in the home network, thecommandenteredbytheuserispassedtothedevice throughnbm.inthisway,userscancontrolspecificdevices remotely.inthiscontext,nbmcanbeconsideredthemost important module since it acts as a networking bridge between the home network and the external network. In UPnP-based data communication, XML data is used as a vehicle for transferring data and HTTP is used as the protocol. The home gateway, devices, and data server communicate with each other in XML format. Therefore, metadata information associated with a specific device has to be extracted from XML-formatted data. To extract the necessary information from the XML formatted data, we implemented a DMPM inside the home network. DMPM specifically parses XML data sent via HTTP and extracts metadata information such as device type, model name, and URL. The information extracted by DMPM is either immediatelyusedbyothermodulesorstoredinatemporary space for future use. The home gateway acts as a bridge between the home network and the cloud-based external data server. In addition,itisinchargeoftheexchangeofdataandcollectionof information to and from the devices deployed in the home network. Metadata information associated with the devices is sent to the data server via home gateway. Using the metadata information, the data server in conjunction with various modules can offer a device monitoring service to users. In the next section, we give details about the cloud-based data server Cloud-Based Data Server. The data server delivers metadata information to the smart devices, stores and manages device information, and provides monitoring services to users. The metadata include the device information, command codes, and service information. Metadata are deposited in HDFS and used to provide monitoring services through MapReduce. Because metadata information is passed in XML format, the data server needs a function to extract XML data. To handle metadata generated by the electronic devices, the data server uses a combination of HDFS, MapReduce, metadata extraction module (MEM), NBM, and MSM. Details about each module are given in the following sections. Figure 3 depicts the internal structure of the cloud-based data server. In order to keep communicating with the home gateway residing in the home network, the data server is required to have a counterpart NBM to the NBM in the home gateway. The NBM allows the data server to send device control commandstothehomegatewaywhentheusermakesa request for a specific device. It also allows the home gateway to send metadata information generated by the electronic device to the data server. When the user sends a command to a specific device, the command has to be delivered to the home gateway. To do this, NBM is equipped with a UPnP bridge feature that enables it to provide host functions based on the UPnP functional demands. Using NBM, the data server can communicate in both directions and receive/send XML data via HTTP without any restrictions. XML data sent bythehomenetworkarepassedtoitscounterpartnbm in the data server. The NBM then sends the XML data to the MEM because the metadata have to be extracted in

140 6 International Journal of Distributed Sensor Networks Smart device Data server HTTP Home appliances Status Info. Control point Data transfer Home network SOAP/HTTP/SSDP/GENI UPnP device Data transfer HTTP HTTP HTTP WAN based Internet Home gateway Home appliance (Humidifier, fridge simulators) Figure 4: The data flow in our proposed system. Home appliances Status Info. transfer order to get separate information such as a device s feature, service information, and state information. The extracted information is then stored. The MEM extracts detailed information about each device from the XML data sent by the home gateway. In other words, the MEM parses the XML data and extracts individual field values from it. For more accurate data extraction, the MEM is designed and implemented using the UPnP function. OncethedatahavebeenextractedbytheMEM,theyare stored in a distributed storage space that is named according to the device with which the data is associated. HDFS is considered the most suitable for storing the data because the data are deposited in a distributed storage space. Hadoop-based MapReduce takes the extracted metadata and separates them according to the name of the devices, specifically, the model name. In this module, a mapping process is carried out to investigate and store the metadata delivered accompanied with the model name. When the data are stored in HDFS, three duplicates are made and stored in a distributed manner due to the technical nature of HDFS. In termsofscalabilityofdata,hdfsisalsocapableofeffectively storing and maintaining the data even if the data consistently increaseinsize.eventually,thedatastoredinhdfsareused to provide electronic device monitoring and control services to end users. The metadata stored in HDFS are transferred to an MSM when a user makes a request. The MSM establishes a oneto-one connection with the user s smart device. To extract status information about the device specified by the user, the MSM acquires the metadata associated with the device fromhdfs.itthenpassesthedatatotheuserintextformat via HTTP. On successful delivery of the information to the user, he/she can then check the condition of the device on his/her smart device. In addition to the monitoring service, themsmiscapableofcontrollingthedevicespecifiedbythe user. The user can check the available control functions on his/her smart device. When the user clicks a command, a request is sent to the MSM located in the data server, which then passes it to MapReduce. The data server then extracts information from the command using MapReduce and sends ittothehomegatewayvianbm.inshort,themsmactsas aninterfacetoprovideservicestotheuser. 4. Data Flow In this section, we discuss the data communication architecture among the components, focusing on the data server and the home gateway of the home network. Figure 4 illustratesthedataflowthatoccursamongcomponents.allthe components of the proposed system are capable of transmitting data in XML format. As shown in Figure 4, the home network comprises electronic devices, the home gateway, and a router. The router facilitates the connection between the internal network and the external network. All home appliances generate information about themselves, such as their features, service capabilities, and current status.devicefeaturesandservicecapabilitiesarepredefined by respective vendors, while status information is generated by a sensor integrated with the device. Sensors differ according to the device with which they are associated. However, a sensor is designed to meet some criteria to properly perform its basic function as specified. For example, a refrigerator

141 International Journal of Distributed Sensor Networks 7 is divided into a freezer compartment and a refrigerator compartment. Each compartment has a distinctive sensor to measure temperature with different criteria. The user can check the temperature for each compartment using the relevant sensor. Sometimes there may be a need to adjust the temperature in the compartment depending on the food we want to keep in the refrigerator. In addition, the refrigerator could provide a function that retrieves the expiration date of the food stored in it. These kinds of information are dealt with by the data server. The refrigerator only passes the data, such as the expiration date, to the data server. In the case of a humidifier, the sensor would need to measure the water level of the water tank and transfer that data to the data server. The sensor is used to measure the device s condition except for the basic features of the device. Information regarding the device s features is advertised within the home network via the router. The home gateway collects the information advertised via the router, checks it, and registers it in the DRF. The home network is configured based on UPnP, so the data are exchanged among the devices in XML format. For this reason, the device features and service information are extracted using the DMPM. Specifically, DMPM extracts the device s URL and model name from the advertised data. After these data are extracted, the information is stored in a temporary space. The information is then utilized by the DSF to collect detailed information about the device. On the basis of the information registered about the device, the DSF makes a request for detailed service information to the specific device. When the device receives this request from the DSF, it sends service function metadata back to the DSF. As with the other types of information, the service function metadata are extracted by the DMPM. The information is then used when theactioncommandistransmittedtothedeviceinthefuture. Once the DSF obtains service function information about a specific device, all of the data are sent to the data server through the NBM, a communicating module that uses the HTTP protocol. The data server obtains various pieces of information about the home appliances, such as device features, service capabilities, and status information, from the home gateway via NBM. Once the information has been received, the MEM examines it and divides it into separate metadata such as model name, model URL, devicetype, friendlyname, model description, service description, and action name. Those metadata are then deposited in HDFS using MapReduce, and are used as basic data for providing device-monitoring services to the smart device of the user. When the user sends a request to the data server, the data server sends model name and state information for the device using the MSM. In our proposed system, a user can monitor an appliance s state and control it on his/her smart device. In providing monitoring and control services for the devices, the user s smart device does not need to have data processing capabilities itself. The smart device only utilizes the services provided by the data server. This means that the function required by the smart device is very simple. Thus, if the Internet is available, the user can monitor and control the devices via the data server. If the user attempts to change an appliance s Table 1: Specifications for the components of the development environment. Content CPU Data server 2.53 GHz Dual-Core Home gateway 2.9 GHz Quad- Core Appliances simulators 1.4 GHz Quad- Core Smart devices 1GHz Dual-Core RAM 2 GB 2 GB 2 GB 2 GB Hard disk 300 GB 150 GB 16 GB 16 GB OS Ubuntu Ubuntu Android Android Libraries Hadoop, Android Cling Cling Cling API state using his/her smart device, it is necessary to send the relevant command to the data server. The data server then fetches the commands that enable the user to control the device. To do this, the data server finds out the action command for the device in the metadata stored in HDFS in conjunction with MapReduce. Once the action command is found,itissentbacktothehomegatewayvianbm,together with information about the device. On receiving the action commandandthedevice sinformation,thehomegateway sends an appropriate command to the specific device. When the device gets the command sent by the home gateway, it changesitsstateinaccordancewiththecommand.oncethe change is complete, the related information is sent back to thehomegateway.asdataflowsfreelyaroundthesystem,the system s components carry out their functions in an organic fashion by interfacing with one another. Figure 5 illustrates the data flow activities for each step. 5. Implementation of Our Proposed System 5.1. Development Environment. We implemented all the devices comprising our proposed system using a generalpurpose computer and smart devices. Specifications for the variouscomponentsusedinthedevelopmentenvironment for the implementation are listed in Table 1.The data server was a four-node cluster because it needed to be capable of storing metadata generated by a number of home appliances and creating large amounts of log files accordingly. In order to prevent loss of log data, the data were stored in HDFS on the data server. Hadoop-based MapReduce was applied to process the data. The Cling library was used to realize communication with the home network. Because the Cling library has excellent support for UPnP and is Java-based, it is suitable for MapReduce, which runs on a Java-based platform. In addition, Cling provides a UPnP library based on Android. Our system was implemented using simulators acting as home appliances instead of actual devices. The simulator was developed with the Android-based Cling library. Finally, a smart device was implemented with a focus on enabling it to be used anywhere an internet connection is available because the smart device should be able to access the data server in order to allow the user to monitor and control the home appliances.

142 8 International Journal of Distributed Sensor Networks Smart devices Data server Home gateway Appliances Metadata transmission Advertisement Device collection Subscription Metadata transmission Metadata extraction and storage Sensor Monitoring service request Metadata extraction and storage (Hadoop based processing) Metadata selection Status view Action call Metadata transmission Action command transmission Action command selection Action command transmission Action command transmission Change status Figure 5: Data flow activities among the steps in our proposed system. Figure 6: Implemented home appliance simulators Implementation Results. We implemented all components, except for the smart devices, using the Java-based Cling library. We also implemented the simulators for home appliances such as refrigerators, humidifiers, and air conditioners using a Samsung Galaxy Note 10.1 tablet. We then tested the system in conjunction with the simulators. The simulation results are shown in Figure 6. After establishing a connection between the home gateway and the data server, we sent information about the home appliances to be stored in HDFS on the data server. The stored data were then transferred to a user s smart device by the MSM of the data server. Using the data sent by the MSM, userswereabletobrowseinformationabouttheconditionof the home appliances and control them at will. Figure 7 shows screen captures for the data server, the home gateway, and the smart device. To implement the home appliance simulator, we used an API provided by the Cling library. Table 2 lists the classes and functions provided by the API. The main roles of the home gateway are as follows: (1) collection of information about the home appliances, such as device features, service capabilities, and conditions; (2) sending the information collected to the data server; and (3) allowing the user to control the home appliances by means of commands sent by the data server to the device. We implemented these functions using the Cling library. The APIs used to collect information about devices are listed in Table 3. The home gateway is in charge of sending the device s statusinformationtothedataserver.todothis,weimplemented a Cling-based bridge in the home gateway. The bridge enables the home gateway to communicate with the data center. As mentioned above, NBM operates as a bridge. The main APIs used for implementing the bridge are listed in Table 4. Table 5 liststheapisthatareusedforstoringmetadatain HDFS on the data server. The main APIs used by MapReduce to extract information are included in the APIs shown in Table 5.

143 International Journal of Distributed Sensor Networks 9 (a) (b) (c) Figure 7: Implementation results for the data server (a), the home gateway (b), and the smart device (c). Table 2: Cling library API simulation classes and functions. Table 4: Classes and functions of the network bridge module. Class names Device Service UDN DeviceDetails AnnotationLocal ServiceBinder Registry SwitchableRouter Functions Generates the device features of devices Generates the service information of devices Generates UUID Used to advertise device features Binding service information provided by devices Registers device features and service information Used to connect to the home network Class names BridgeStartedEvent BridgeStoppedEvent ConfigureBridgeController EndpointController Bridge Functions Used to start a WAN bridge Used to stop the WAN bridge Configures HTTP-based address Used to connect with an endpoint Used for getting communication arguments Table5:TheAPIsusedtoimplementthedataserver. Class names Device Service UDN Registry SwitchableRouter RegistryListener ActionCallback Subscription Callback ActionArgument Value GENASubscription 6. Conclusions Table 3: APIs used in the home gateway. Functions Generates the device features of devices Generates the service information of devices Generates UUID Registers device features and service information Used to connect to the home network Used to collect device information Used to transmit action commands to control devices Requires detailed service information about devices Gets detailed service functions Periodically requires status information of devices Users can monitor and control home appliances from outside a home network using our proposed system. Home appliances form ubiquitous sensor networks via status sensors that they use to advertise device features and service information. According to the home appliance information advertised, a home gateway collects metadata from the home appliances Class names Configuration Path InputSplit FileSplit GetPath FSDataInputStream Functions Sets system states Path of files or data directory Splits the data stream Splits files Returns the split files path Used for storing files in HDFS and sends them to a cloud-based data server. The data server stores metadata on HDFS, processes them using MapReduce, and also uses them to provide a monitoring service to users. Users can also control the home appliances via the NBM providedbythedataserver.asaresultoftheprovisionofthe home appliance monitoring and status-controlling functions, users can easily confirm the status of home appliances. Our proposed system easily deals with the large amounts of home appliance metadata generated by processing them using cloud computing technology and effectively utilizing computing resources. In our proposed system, HDFS is used to store and manage home appliance metadata. However, an effective approach to store and manage semistructured data such as XML is needed. Therefore, as future work, we plan to design a real-time data monitoring system to effectively manage semistructured and unstructured data. In order to reduce the computational burden of smart devices, we also aim to design a zero-client interface using HTML5 to support the monitoring and control of home appliances. Furthermore,

144 10 International Journal of Distributed Sensor Networks we plan to research about home cloud services to handle multimedia devices and home appliances integration system. Conflict of Interests The authors declare that there is no conflict of interests regarding the publication of this paper. Acknowledgment This paper was supported by Konkuk University in References [1] E. U. Warriach, E. Kaldeli, A. Lazovik, and M. Aiello, An interpaltform service-oriented middleware for the smart home, International Journal of Smart Home, vol.7,no.1,pp , [2]H.Gu,Y.Diao,W.Liu,andX.Zhang, Thedesignofsmart home platform based on cloud computing, in Proceedings of the International Conference on Electronic and Mechanical Engineering and Information Technology (EMEIT 11),pp , Harbin, China, August [3] X. Ye and J. Huang, A framework for cloud-based smart home, in Proceedings of the International Conference on Computer Science and Network Technology (ICCSNT 11), pp , Harbin, China, December [4] M.Z.Bjelica,B.Mrazovac,V.Vojnovic,andI.Papp, Gateway device for energy-saving cloud-enabled smart homes, in Proceedings of the IEEE 35th International Convention (MIPRO 12), pp ,Opatija,Croatia,2012. [5] M.M.Islam,J.H.Lee,andE.-N.Huh, Anefficientmodelfor smart home by the virtualization of wireless sensor network, International Journal of Distributed Sensor Networks, vol.2013, Article ID , 10 pages, [6] Y. M. Baek, S. C. Ahn, and Y.-M. Kwon, UPnP network bridge for supporting interoperability through non-ip channels, IEEE Transactions on Consumer Electronics, vol.56,no.4,pp , [7] D.-O. Kang, K. Kang, S.-G. Choi, and J. Lee, UPnP AV architectural multimedia system with a home gateway powered by the OSGi platform, in Proceedings of the International Conference on Consumer Electronics (ICCE 05), pp , January [8] P. Bellavista, P. Gallo, and C. Giannelli, Discovering and accessing peer-to-peer services in UPnP-based federated domotic islands, IEEE Transactions on Consumer Electronics, vol. 58, no. 3, pp , [9] N. Ishikawa, PUCC activities on overlay networking protocols and metadata for controlling and managing home networks and appliances, Proceedings of the IEEE, vol.101,no.99,pp.1 12, [10] H.-Y. Hsueh, C.-N. Chen, and K.-F. Huang, Generating metadata from web documents: a systematic approach, Human- Centric Computing and Information Sciences, vol.3,no.7,pp. 1 17, [11] L. Xu, X. Zheng, W. Guo, and G. Chen, A Cloud-based monitoring framework for smart home, in Proceedings of the IEEE 4th International Conference on Cloud Computing Technology and Science, pp , Taipei, Taiwan, [12] D. Díaz-Sánchez, F. Almenarez, A. Marín, D. Proserpio, and P. A. Cabarcos, Media cloud: an open cloud computing middleware for content management, IEEE Transactions on Consumer Electronics, vol. 57, no. 2, pp , [13] Y. Pan and J. Zhang, Parallel programming on cloud computing platforms challenges and solutions, Journal of Convergence,vol.3,no.4,pp.23 27,2012. [14] M. Kim, S. Han, Y. Cui, H. Lee, and C. Jeong, A Hadoop-based multimedia transcoding system for processing social media in the PaaS platform of SMCCSE, KSII Transactions on Internet and Information Systems,vol.6,no.11,pp ,2012. [15] K. Shvachko, H. Kuang, S. Radia, and R. Chansler, The Hadoop distributed file system, in Proceedings of the IEEE 26th SymposiumonMassStorageSystemsandTechnologies(MSST 10), pp. 1 10, Incline Village, Nev, USA, May [16] J. Dean and S. Ghemawat, MapReduce: simplified data processing on large clusters, Communications of the ACM, vol.51,no. 1, pp , [17] J. Liu and S. -H. Chung, An efficient load balancing scheme for multi-gateways in wireless mesh networks, Journal of Information Processing Systems,vol.9,no.3,pp ,2013. [18] Y. Cui, M. Kim, and H. Lee, Social media sharing system: supporting personalized social media service using UPnP technology in cloud computing environment, Information, vol.15, no.5,pp ,2012. [19] UPnP Forum, UPnP AV Architecture 1.1, October [20] Y. Cui, M. Kim, and H. Lee, Cloud-based home media system model: providing a novel media streaming service using UPnP technology in a home environment, International Journal of Software Engineering and Its Applications, vol.7,no.4,pp , [21]Y.Kato,T.Ito,H.Kamiya,andM.Ogura, Homeappliance control using heterogeneous sensor networks, in Proceedings of the Consumer Communications and Networking Conference,pp. 1 5,LasVegas,Nev,USA,2009. [22] Y.-S. Jeong, E.-H. Song, G.-B. Chae, M. Hong, and D.-S. Park, Large-scale middleware for ubiquitous sensor networks, IEEE Intelligent Systems,vol.25,no.2,pp.48 59,2010. [23] J. Gao and Y. Xiao, Design for accountability in multi-core networks, Journal of Convergence,vol.3,no.3, pp.9 16,2012. [24] C.-W. Chen, Y.-R. Tsai, and S.-J. Wang, Cost-saving key agreement via secret sharing in two-party communication systems, Journal of Convergence,vol.3,no.4,pp.29 36,2012. [25]T.Matsuura,K.Hisazumi,T.Kitasuka,T.Nakanishi,andA. Fukuda, UDSS: sensor device for context awareness in home network, in Proceedings of the 4th International Conference on Networked Sensing Systems (INSS 07), pp , Braunschweig, Germany, June [26] Z. Wei, S. Qin, D. Jia, and Y. Yang, Research and design of cloud architecture for smart home, in Proceedings of the IEEE International Conference on Software Engineering and Service Sciences (ICSESS 10), pp , Beijing, China, July 2010.

145 Hindawi Publishing Corporation International Journal of Distributed Sensor Networks Volume 2014, Article ID , 9 pages Research Article GTrust: Group Extension for Trust Models in Distributed Systems Robson de Oliveira Albuquerque, 1,2 Luis Javier García Villalba, 1 and Tai-Hoon Kim 3 1 Group of Analysis, Security and Systems (GASS), Department of Software Engineering and Artificial Intelligence (DISIA), School of Computer Science, Office 431, Universidad Complutense de Madrid (UCM), Calle Profesor José García Santesmasess/n,CiudadUniversitaria,28040Madrid,Spain 2 Electrical Engineering Department, University of Brasilia, Campus Universitário Darcy Ribeiro, Asa Norte, Brasilia, DF, Brazil 3 Department of Convergence Security, Sungshin Women s University, Dongseon-dong 3-ga, Seoul , Republic of Korea Correspondence should be addressed to Luis Javier García Villalba; javiergv@fdi.ucm.es Received 30 October 2013; Accepted 20 December 2013; Published 10 February 2014 Academic Editor: Naveen Chilamkurti Copyright 2014 Robson de Oliveira Albuquerque et al. This is an open access article distributed under the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited. This paper proposes and describes a trust model for distributed systems based on groups of peers. A group is defined as a collection of entities with particular affinities and capabilities. All entities may have a trust and a reputation value of each other in the system. In many cases it may be necessary to trust the whole system instead of one particular entity. In such cases group trust represents the trust of their particular members. To achieve this, this paper presents a group trust calculation model. We implemented the proposed model in a P2P simulation tool and presented main results for group trust calculation. 1. Introduction Nowadays, distributed systems have become very complex environments in that hundreds of nodes have to collaborate in order to provide large-scale services. Some examples can be P2P networks, multiagent systems, grid systems, cloud systems, and so on. In these scenarios, trust between entities becomes a crucial factor as a way of determining the reliability of the different nodes in the system and to detect and predict misbehaviors and security threats. In these environments, trustcanalsohelpintheveracityofthesystem. Considering that the Internet nowadays has so many different types of things connected to it, there is no known way of dealing with the amount of information or data that is necessary to verify if systems are fully reliable. This amount of connected elements and its functions are commonly named Internet of things [1]. Once that so many possibilities of information and systems are connected and a lot of them interoperable, the amount of security threats also increases significantly because of the complexity, distribution, channels, data, and networks, which make the traditional security approaches inefficient. This also leads the technology and researches towards new manageability challenges. Ithasbecomeclearthatnewtrustmanagementsystems are necessary in order to accomplish security related to distributed system [2] considering the amount and variety of systems connected. The management of trust relationships between different peers belonging to a distributed system can be done using different approaches. The trust relationships can be manually established by each node in the network around the rest of the nodes. This manual approach does notscalewhenthenumberofnodesbecomesbiggerand bigger, so other approaches based on a trust model are used to automatically calculate how much a node can trust other nodes. Usually, a given node follows a trust model to determine if a node is trustworthy or not depending on two different aspects: (i) trust values are locally calculated

146 2 International Journal of Distributed Sensor Networks and (ii) trust values are provided by the rest of the nodes in the network (reputation). Current trust models are generally used to calculate one-to-one relationships between nodes. This means that an entity A needs to determine the trust value for all the other entities available in the distributed system, which has to communicate with A. In large-scale scenarios, this fact is an important lack of scalability in distributed systems due to the fact that A needs to calculate and maintain as many trust relations as the number of nodes necessary to accomplish A s work. Other scenarios consider several entities as a single node for trust purposes. For example, a distributed file system could be seen as a single entity even though it is composed of several subentities and a security threat in any of them may compromise the complete file system. This kind of simplifications may cause a lack of the accuracy of the current models. When A needs to communicate with a group of nodes having the guarantee that the group itself is trustworthy, A wants to independently form an opinion about the whole group. If this is the case, it may be necessary to establish communication with every group member. This makes entity A start an exhaustive process of discovering every member of the group. In most practical situations this is not a good idea becauseitwillmakeentityatrytofindmaybethousandsof nodes in a network. This consumes time and resources for entity A and may not be the best approach to achieving its objectives. To address these problems, the main contribution of this paper offers a novel extension to conventional trust models that enables the support for calculating the trust values for groups of nodes. A group is defined as a collection of entities with particular affinities and capabilities. Then, the entities which need to interact with a given group are able to find a trust value for the whole group directly, thus avoiding the necessity of discovering the whole group members and providing more accurate information about the group than approaches which consider the group members as a single entity. The extension provided has been implemented and validated by means of a set of statistics. In order to describe the proposal, this paper is organized as follows. Section 2 reviews some aspects and definitions about trust and reputation. Section 3 discusses some related work of different trust models for distributed systems. Section 4 presents the proposed group trust model. Section 5 shows implementation results and Section 6 ends this paper with the conclusions and future works. 2. Definitions The concept of trust and its definitions has been studied by many authors and is part of many research projects. Trust is recognized as an important aspect for decision-making in distributed systems [3 6], but there is no general consensus in the literature on the definition of trust and what trust management comprehends. For example, Patel [7] considers that trust in computational relations may be focused on the optimized selection of a communication partner and on Distrust Node exit Betrayal 0 1 Environment situations Figure 1: Trust or distrust and influences. Blind trust Table 1: Summary of common set of trust characteristics. Characteristic Trust is context aware Trust can be measured Trust changes with time Trust is socially aware Trust may be directional Example Entity A may trust B to download files but does not trust B to perform routing Entity A has more trust in entity B than A s trust in entity C The amount A trusts B may increase or decrease as interaction happens Entity A may trust entity C because C was presented by entity B, and A already trusts B Entity A may trust B, but B may not trust A the decision of agreements between two or more members of the network. That is particularly true in distributed environments where there is no certainty about the destination and the communication process can be easily deceived. Gambetta [8]considerstrustasaparticularlevel ofsubjective probability and Marsh [9]says that trust involvesprobability and this permits a representation of trust in values between zero and one. This approach creates a concept where a trust value ranges from 0, which means complete distrust, to 1 which means complete trust. This value may suffer interference of nodes, betrayal, and so forth. Moreover, complete trust is not desirable in most scenarios because it eliminates any possibilities to suspect any particular node. Figure 1 illustrates this consideration. There are common aspects in almost all the references about trust models that make us determine a common set of features related to trust. These features can be summarized in Table 1. As opinion, it has many subjective evaluations depending on several factors like situation observed (context), own trust value inference, information received by others in a social relationship, and so forth. The reputation value evolves with time and directly depends on the behavior of the entity being observed. Reputation may also represent indirect trust. It includes asking for the opinion of other parties with whom the entity has previously interacted in the past about a third entity. Reputation can also be defined as the common opinion of others regarding an entity [7], which may be used in the absence of trust formed from personal opinions. Reputation values take time to be acquired, but it can be easily lost

147 International Journal of Distributed Sensor Networks 3 in social aspects. Calculation of reputation values is done using past information that has been obtained during time and based in the information that was received from trusted parties.thesevariablesenableanentitytoformanidea about an unknown entity. It could be considered as a social evaluation of an individual or group of individuals. 3. Related Work There is an important number of research works providing trust models for different distributed scenarios such as grid computing, P2P, multiagents, ad hoc networks, wireless sensor networks, and cloud computing. This section provides a review of trust and reputation models in order to motivate our new trust model in context with the other works. Regarding trust models for Zhao and Dong [10] presentatrust model multidomain grids scenarios and consider that trust canbeusedtoevaluatetherelationshipbetweengridresource providers and grid consumers. Jingshan et al. [3] present a trust model for grids based on the information provided by the last service used. According to the authors, their proposed model achieved less resources occupied, increased flexibility, and prevention of threat of malicious evaluation and cooperative cheating in the grid. GridTrust [11] security framework is able to provide trust management in vertical andhorizontalapproachinthegrid.gsfisbasedonlayers and policies that control and monitor users activities in such a way that security and trust are guaranteed by their model. Regarding trust models for P2P, the novelty and strong point of the model proposed in DWTrust [12] isthatall the factors that have influence on the trust value for a particular node are represented as dynamic weights that adapt themselves depending on the trust policy of each node. AntRep [13] is another trust model where reputation evidence is distributed over a P2P network based on the swarm intelligence paradigm [14]. In AntRep each peer has a reputation table (RT) which is very similar to the distancevector routing table [15] but differs in the following aspects: (i) each peer in the RT corresponds to one reputation content; (ii) the metric is the probability of choosing each neighbor as the next hop instead of the hop count to destinations. EigenTrust [16] has become one of the most cited trust models for P2P networks. It achieves a decrease in the number of downloads of inauthentic files in a P2P file-sharing network by assigning each peer a unique global trust value, based on the peer s history of uploads. Regarding trust model for mobile ad hoc networks (MANETs),Changetal.[17] proposeamarkovchain-based trust model to determine the trust value for each one-hop neighbors in multicast MANET and the results indicate that the convergence speed is independent of the trust classes and of the initial values of the proposed model. Liu et al. [4] presentareputationmodelabletousesubjective opinions with familiar values in order to prevent selfish behaviors in MANET. Nodes accumulate reputation informationandcreateafamiliarityvalueusedtocomputeimpact of reputation recommendation. RRS for P2P and MANETs [18] present an enhancement of CONFIDANT [19], which is a robust reputation system for P2P and mobile ad hoc networks where everyone maintains a reputation rating and a trust rating about everyone else they care about. PTM [20] is a decentralized trust model which expressed trust relationships with fuzzy logic. These relationships can be established as direct or indirect. In the former, A will trust B without intervention of third parties. In the latter, the indirect trust relationships are given by recommendations from TTPs. A TTP is a peer who has a trust value higher than a certain threshold. Such recommendations are distributed using a pervasive recommendation protocol (PRP) among close entities. Regarding trust models for multiagents systems, TRAVOS [7] consists of a trust and reputation model for virtual organizations based in agents, in which trust is measured using probability. The evaluation of the amount of trust is based on past interactions and reputation obtained by other nodes. Sporas [21]is another reputationmechanism in agent systems where the reputation is computed recursively andwherethemorerecentaratingis,themoreweightit has. MTrust [22] uses a Bayesian network to calculate the trust value among entities in the network. It is focused on a mobile agent system, where the cooperative interactions among these agents and their respective visited hosts are ensured. Regret [23] (one of the most representative trust and reputation models in multiagent systems) manages reputation from three different dimensions: the individual one, given from direct interactions with the agent; the social one, from previous experiences of group members with the agent and its acquaintances; and the ontological one, given by the combination of multiple aspects in order to build a reputation about complex concepts. AFRAS [24] proposes a reputation mechanism in multiagent systems whose main characteristic is the modeling of an agent reputation and the interaction rating as fuzzy sets. In case the reader is more interested in how many of the previous models are being calculated, Gómez Mármol and Martínez Pérez [5] provide a very complete and comprehensible survey of trust models for distributed systems. Largillier and Vassileva [25] argue that group formation is a difficult task. It has many different contexts and groups can be formed based on users criteria or using methods that matches what users desire. In most cases it does not take into account previous successful or unsuccessful collaborations to forge new ones. Considering this, their work proposes a model of collaborative trust to help select the criteria that is the best fitted group for a task. Al-Oufi et al. [26] explain that in order to protect users it is important to identify trustworthy people. Their work extends the Advogato trust metric [27]so trustworthy users can be identified. The authors [26] claim that their model has advantages over existing representative methods because it is able to discover reliable users and prevent unreliable users. Easa et al. [28] say trust is used in soft security and proposes a group-based trust method to propagate information among peers. They also consider two factors called intermediate group confidence and group confidence used between two groups. Related to trust and the Internet of things, Saieda et al. [2] propose a new trust management system and design

148 4 International Journal of Distributed Sensor Networks a context-aware and multiservice trust management system fitting the new requirements that the authors consider important in their model. Schulz and Tjstheim [29]pointoutthat, when there are interactions with objects and services in the Internet of things, usually the users need to trust that their data is safe and that things will fulfill their promise. Wang et al. [30] consider trust management as a way of providing a potential solution for the security issues of distributed networks and propose a new distributed trust management mechanism for the Internet of things. They propose a model using sensor, core, and application layers which provide a general framework for the study of trust management for the Internet of things. All this preview work shows that trust is used in many different distributed technologies. Very few works consider the group aspect of distributed system as an important characteristic to help provide security and reliability in distributed systems. It is also important to remember that trust and reputation have increased significantly in recent years and nowadays they are being considered as important factors to help extend the functionalities in distributed systems. 4. Group Trust Model This section describes the extension for supporting groups in trust models. This extension enables the definition of trust values over both groups and single entities. A group is defined as a collection of entities connected together with common goals or even common contexts. Thus the entities are able to perform specific works in a common context like service offering. Moreover, the entities are also able to perform trust and reputation calculation of other entities in the system that considers any interaction. This group extension can be deployed over a system in which entities use any trust and reputation model that attends to the system needs. This work just considers that there is a trust and a reputation algorithm that can perform trust and reputation calculations and is considered as an extension over such algorithm. The trust added value is a consequence of the individual trust values of the group members. The added value represents a point of information for external entities so they can use it to infer the whole group trust value. To perform a trust calculation over a group, it is a requisite of our model that there should be a leader in the formation of the group. It is not easy to determine a leader for holding trust information and thus a method by consensus is adopted to determine the leadership of the group. Entities in a group may be able to agree to a minimum level of trust (trust threshold) in order to make a common analysis and commonly choose a leader based on trust. The problem is that not every entity has thesametrustvalueaboutanyotherentity.thisisbecause trust is calculated by every entity using its own ability and making use of its own inferences. Entities may agree in an ordinary value of trust and also agreethatthisvalueisenoughtoassumethatonespecific entity can represent the whole group. This assumption transforms the chosen entity to the leader of the group. In real distributed system scenarios where every node can perform its own decision, such agreement is very difficult because entities must exchange trust and reputation information that should have been defined previously considering, for example, security aspects as availability and integrity. However, if it is assumed that an entity should not trust other entity if its trust value is not inside a specific range, this entity could avoid exchanging information with another in the system thus leading to a complete failure of acquiring enough trust information to create trust consensus. Besides, this should consider that some basic factors are commonly known and used by every entity in the system and that may not be true in most distributed systems. If it is considered just voting schemas for a leadership choosing process, this may not represent consensus because what an entity does is just to vote. In other words, it is to choose one among many options. Voting schemas, in general, do not consider consensus in a distributed manner, which contradicts the trust aspects. It is not the objective of this work to develop a trust consensus algorithm or a trust consensus model in order to choose a leader. To have a consensus process enables entities to express their opinions about the leadership election process. It is important to observe that any entity in the group may be the potential leader of the group. One prerequisite is that the candidate entity has trust and reputation values expressed by the members of the group. Extending this view, any entity may announce its preferred candidate also depending on the context. That makes the leadership choosing process more complicated than just voting. For example, an entity could be responsible for taking other entities opinion and announcing inthenetworkwhohasthemostelevatedtrustvalueamong n+1members, where n is half of the whole group. Well, in large groups that can be a problem because this process may flood the network with messages for just choosing one entity as leader. Another point of view is that entities may inferthatatrustvalueisacceptableornotthusgeneratinga trust threshold. If an entity does not overcome a specific trust value than another entity, it may assume by its own means that the entity in focus may not be the leader for it. However, that may not be true for another entity that believes that it has enough trust to be the leader. Both assumptions lead us to a quantitative trust measure and not a qualitative approach and in some distributed system that is not adequate. It is not the objective of this work to develop a trust leadership algorithm or a trust leadership model. So, once a leader already exists in the groups, entities in the group have agreed that this leader is the representation of the group for new members and for the outside world. It is normal that entities may not be able to be actuated in every context available in the system. For example, one entity may be able to upload files but may not be able to perform matrix calculation. So it is a requisite to the model that the leader in the group knows every context in order to be able to calculate the trust information of the group for all such contexts available within the group. Thus, let s define how to calculate the trust value of the groups. Firstly, the reputation

149 International Journal of Distributed Sensor Networks 5 that entity B has for entity A in a particular context C is represented in the following equation by notation δ C a,b : δ C a,b = iv C a,b, (1) where i V is the reputation value calculated for the interaction i using the available reputation model in the system. In this case i V C a,b is one record that represents the expectation that B will fulfill A s requests in context C for the interaction i. Trust and reputation values may be stored as many individual records of every contextualized interaction with the same or different entity. Thus, an entity may have a collection of different reputation values for other entities in the system. Then, an entity B may calculate the final reputation about an entity A for a particular context C using δ C a,b = j i=1 i VC a,b j with j>0, (2) where δ C a,b is the final reputation that entity B has for A in context C and j represents the amount of interactions that A and B have done in context C. Following,oneentitymay haveasmanycontextsasitisprogrammedto.thenthefinal reputation regarding all contexts of one entity about another is given by the following expression: K a,b = x i=1 δc i a,b x with x>0, (3) where K a,b is the final reputation value that entity B has for A for all contexts and x is the amount of all contexts that B knows about A. ThiswayentityB is able to store all the reputation information about A in a given time period as one value. This value is used to perform the calculation of the trust value for the group. It is important to remember that reputation evolves with time, so this value can go up or down during time as B interacts with A. In our model, we consider that what best represents the trust value of a group is the reputation that every entity within the group has about all entities of the group that it is partof.then,thetrustvaluecanbecalculatedasanaverage reputation of all members inside the group. Considering this, the leader of the group receives and organizes the reputation values of the rest of the entities and computes the trust value of the group. For example, let s consider an example group of 5 members G = {M1, M2, M3, M4, M5} and M5 is the agreed group leader. In this case, M5 asks M1 about the reputation information about M2, M3, M4, andm5. After that, M5 asks M2 about the reputation information of M1, M3, M4,andM5 and so on. Then, M5 uses this information to calculate the trust value of the group. It may sound that M5 may manipulate the reputation value that it receives, which is true. To avoid that particular case, we assume that the leader role is an important function in the group so the leader must not cheat; otherwise the whole system may fail if it is trust based. It is important to remember that there is no common trust communication protocol for exchanging information in distributed systems. We assume in our model that the protocol to exchange information with the leader can be proactive, reactive, and hybrid depending on the scenario in which the protocol is deployed. Note that in some scenarios the members of groups already know who is the leader and then can proactively send such information. A new member canalwaysaskfortheleaderofthegroup,soweassumethat the node is able to find and communicate with the leader. Once this process is defined, the leader computes the final reputation of each entity as the average of the reputation values provided by the rest of entities within the group: ω n g = j i=1 K a,b j with j>0, (4) where ω n g is the average reputation of entity n as seen by the rest of the entities of group g and j represents the quantity of members in group g. After performing the final average reputation of every entity in the group, the leader can generate thefinaltrustvalueofthegroupusingallreputationvalues computed before. This computation process is represented: λ g = x i=1 ω n i g, with x>0, (5) x where λ g represents the final trust value of the group g and x represents the quantity of members in the group. Once the leader has the trust value of the group, the leader can send this information to all members of the group. In thecase wheretherearemanygroups (N), every group leader can perform its own trust value calculation and inform its trust value to other group leaders. Every group leader has theresponsibilitytostoregrouptrustinformation,sendit when asked, and distribute this value to new members, new leaders, and group outside requests as well. As seen the group role is very important, so the leader must be chosen carefully. In order to create a common process to perform group calculation the algorithm represented in Figure 2 can be used. 5. Implementation and Analysis The proposed model has been implemented and some statistics results have been obtained in order to validate it.inessence,atestbedhasbeensetupbymeansofa P2P simulation tool [31] to create a basic group and to develop all calculation processes. The simulation tool uses asynchronous interactions between machines and different scenarios were simulated according to specific policies in the network. Some assumptions were defined in the test environment in order to organize the tests. For simplicity, the testbed is accomplished assuming that peers do not lie about trust and reputation values in the network. However this behavior can also be detected using the underline trust modelusedintheproposal.allparticipantsaredoingthe same number of interactions in the testbed. The objective is to find standards and verify certain behaviors about trust and reputation values in the system.

150 6 International Journal of Distributed Sensor Networks Start Define group leader Ask reputation information of group members Entity calculates its reputation of group known members Send reputation values of entities to leader End No Perform individual reputation calculation Does the leader have all information to calculate group trust? Yes Store reputation information of group members Calculate group trust Send group trust when requested Figure 2: Algorithm for group trust calculation. When interactions between entities correspond (or not) to the expected behavior, it can be determined if the peer is trustworthy (or not). Some behavior patterns have been delimited as desirable in the system. Firstly, there are no errors in the communication transmission; secondly, the time for transmitting a file is determined by the quality level of the transmission. These parameters have been chosen in order to simplify the P2P environment, thus permitting the focus of the analysis on trust and reputation values considered in the interactions of peers and performing the calculation of the group trust value. The testbed is executed in machines with JXTA Shell [31] installed and configured. The simulated environment was composed of 500 nodes. These were defined as 5 different groups with a hundred nodes in each group. Each node only performs interactions within its group. Also, each peer performs at least 20 interactions in the network. The network topology is simple and uses 2 common layer switches. The purpose of this topology is to represent a P2P network connected directly to a LAN. The peers are configured in the same network segment with no additional hops. Each peer uses a different TCP port. This characteristic is to permit the P2P network to establish connections on different ports. The simulation testbed considered that the transmission delay and the integrity of the file are the parameters used to decide whether a peer is malicious or not. This means that the peer can send a corrupted file, delay its sending to another peer, or perform both. The interval of time values for the transmission is defined after some file transfer tests Table 2: Resumed interactions for test parameters. Source peer Destination peer Time (s) Speed (Kb/s) Peer1 Peer Peer3 Peer Peer1 Peer Peer2 Peer Peer6 Peer Peer9 Peer Peer10 Peer Peer6 Peer Peer7 Peer Description Table 3: Probable situations considered. (1) File corrupted and on time (2) File corrupted and a little delayed (3) File corrupted and completely delayed (4) File not corrupted and on time (5) File not corrupted and a little delayed (6) File not corrupted and completely delayed A-File load B-integrity check C were executed. Several interactions have been fulfilled for a filewithfixedsize(100kbytes),andastandardtimecouldbe defined for a successful interaction. Table 2 has some summarized definitions of the test interactions to limit the values expected. Based on this information it has been determined that the expected transference time of a file is up to 1 s, a short delay would be between 1 s and 2 s, and a completely delayed is above 2 s. The other parameter defined is the integrity of the received file. A hash calculation is used to verify this condition. Once this parameter is defined, the file load times are parametrically determined by the variable a,the file integrity check times are determined by the variable b,and the variable c determines the reputation feedback for the trust model associated with the given interaction. Table 3 shows the parameters used in our testbed to define how to infer some reputation for a peer. The reputation value c is determined by the following equation, where the parameter P represents the weight (importance) that the network administrator allocates to the integrity of the file: C=((P a) + ((1 P) b)). (6) Related to peers behavior, peers only accomplished interactions with appropriate parameters to verify the convergence of trust and reputation values. The underlying trust model used in the testbed is TRAVOS [7], which allows peers to

151 International Journal of Distributed Sensor Networks Trust of group 1 Trust of group 2 Trust of group 3 Trust of group 4 Trust of group 5 Figure 3: Group trust of an ideal environment. environment. When 20% of the peers start behaving in a maliciousmanner,thetrustvalueofthegroupdecreasesand tends to stabilize in a value near 0.8. The analysis shows that the increase of the trust coefficients provided by the good peers overcomes the decrease of the coefficient of the malicious peers. In this case the group is still considered trustworthy (λ g > 0.7) despite having malicious members, as represented in Figure Scenario 4: Random Behavior of 40 Peers in Group 1. In this test 40% of group 1 behaves randomly. This test simulates a situation where the P2P network is compromised and there is no guarantee that the peers in this group are trustworthy or not. When 40% of the group members are malicious, the trustcoefficientofthegrouptendstostabilizeinavalue near 0.6 which represents that the peers in the group are not trustworthy,andthusthegroupisnotconsideredtrustworthy because of thethreshold. Figure6 shows this result. realizethatsomemembersofthenetworkchangedtheir behavior. The testbed has been set up in order to have peer1 as leader of each group. Once all these parameters were set up, we defined different scenarios for our tests. Direct trust of the peers and reputation values based on context of the groups are calculated in the simulations in order to calculate group trust. The tested scenarios, its results, and analysis are presented in the following subtopics. In all graphs the x-axis is the number of interactions and the y-axis is the correspondent trust value in each round Scenario 1: All Peers Behave Accordingly. In this test all the peers in all groups behave as expected. This means that they fulfill their requirements and perform their defined context correctly. Note that the entire peer acts following the same behavior, without changing any aspect of its functional context. In this case, the trust value of the group is considered extremely trustworthy and it tends to stabilize in a value near1,thusavoidingblindtrust.whenthereisnomalicious peer in the network, the trust value of the group reflects the individual behavior of the peers in the group. This is considered the ideal world. Figure3 shows this result Scenario 2: Random Behavior of Peers. In this test all peers in all groups behave randomly after round 4. This meansthatitisnotknownforcertainbytheothergroup members whether a particular peer behaved accordingly or not. This test was set up in order to verify the results when nodes behave in a proper manner sometimes and then change their behavior with no particular reason. This can be considered the worst environment imaginable because it cannot be possible to predict if a node will or will not behave accordingly. This scenario is represented in Figure Scenario 3: Random Behavior of 20 Peers in Group 1. In this test 20% of group 1 behaves randomly. This test simulates a coalition of peers in order to modify the group trust. Such behavior is considered as if the nodes suffer some kind of attack or there are peers acting as black holes in the P2P 5.5. Scenario 5: Random Behavior of 60 Peers in Group 1. In this test 60% of group 1 behaves randomly; then the trust coefficient tends to stabilize in a value near 0.5, also making the group untrustworthy because of the threshold, as seen in Figure Group 1 Analysis. This test was performed to analyze group 1, compiled in one graph, as seen in Figure 8. When peers change their behavior, the trust value of the group decreases, thus making a group with constant behavior change to be untrustworthy considering a defined threshold of λ g = 0.7. When nodes behave randomly, the value of the group trust tends to 0.5. The reader may realize that the individual behavior of each member in the group influences the trust value of thegroupasawhole.theresultscanalsobeconsidered satisfactory because all the peers are initiated at the same time in the network and interact with each other the same number of times. Theseresultsalsoshowthatthetrustvalueofgroup1in the first scenario is originally high (moment in which all the peers have good behavior). After that, it starts to decrease in the moment the peers in the group change their behavior or acts forming a coalition. As a result, the group trust model canbeusedasaparametertointeractornotwithaspecific group. 6. Conclusion This work has reviewed different trust and reputation models in distributed systems. We developed a model as an extension to support the calculation of trust values of groups of entities. The proposed model has been validated in a P2P simulation tool.ourresultsshowthatitispossibletogenerateandto calculate group trust behavior in distributed systems. We consider that it is important that a trust leadership based algorithm or a trust consensus algorithm should be better studied in order to create leaders in groups in a distributed manner. It is also important to define a trust protocol

152 8 International Journal of Distributed Sensor Networks Trust of group 1 Trust of group 2 Trust of group 3 Trust of group 4 Trust of group 5 Trust of group 1 Trust of group 2 Trust of group 3 Trust of group 4 Trust of group 5 Figure 4: Group trust of the worst environment. Figure 7: Group trust of G1 when 60 nodes change their behavior Trust of group 1 Trust of group 2 Trust of group 3 Trust of group 4 Trust of group G1 trust-normal behavior G1 trust-random behavior G1 trust-20 bad nodes Figure 8: Group 1 synthesis. G1 trust-40 bad nodes G1 trust-60 bad nodes Figure 5: Group trust of G1 when 20 nodes change their behavior Trust of group 1 Trust of group 2 Trust of group 3 Trust of group 4 Trust of group 5 Figure 6: Group trust of G1 when 40 nodes change their behavior. as a platform to support trust based communications. We consider that as research areas that can be deeply studied. Using the concept of group trust, the proposed model in this paper can be used in bigger and more complex distributed systems architectures. As future work we will implement our group trust model in software agents, grid platforms, or cloud environments in order to evaluate its behavior in bigger systems. Conflict of Interests The authors declare that there is no conflict of interests regarding the publication of this paper. Acknowledgments Part of the computations of this work was performed in EOLO,theHPCofClimateChangeoftheInternational Campus of Excellence of Moncloa, funded by MECD and MICINN. The first author acknowledges the Laboratory for

153 International Journal of Distributed Sensor Networks 9 Decision Technologies at the University of Brasilia (LATI- TUDE/UnB) for its support to this work. References [1] M. J. Covington and R. Carskadden, Threat implications of the internet of things, in Proceedings of the 5th International Conference on Cyber Conflict (CyCon 13), pp. 1 12, Tallinn, Estonia, June [2] Y. B. Saieda, A. Olivereaua, D. Zeghlache, and M. Laurent, Trust management system design for the internet of things: a context-aware and multi-service approach, Computers & Security B,vol.39,pp ,2013. [3] S. Jingshan, Y. Jiabin, and Y. Fengshou, Grid trust model based on last service and hierarchy, Journal of Nanjing University of Aeronautics and Astronautics,vol.43,no.2,pp ,2011. [4] Y.Liu,K.Li,Y.Jin,Y.Zhang,andW.Qu, Anovelreputation computationmodelbasedonsubjectivelogicformobileadhoc networks, Future Generation Computer Systems, vol. 27, no. 5, pp , [5] F. Gómez Mármol and G. Martínez Pérez, Towards prestandardization of trust and reputation models for distributed and heterogeneous systems, Computer Standards and Interfaces,vol.32,no.4,pp ,2010. [6] T. Beth, M. Borcherding, and B. Klein, Valuation of trust in open networks, in Proceedings of the 3rd European Symposium on Research in Computer Security, pp. 3 18, Springer, London, UK, [7] J. Patel, A trust and reputation model for agent-based virtual organisations [Ph.D. thesis], Electronics and Computer Science, Faculty of Physical Sciences and Engineering, University of Southampton, Southampton, UK, [8] D. Gambetta, Can We Trust Trust? chapter 13, Department of Sociology, University of Oxford, [9] S. P. Marsh, Formalising trust as a computational concept [Ph.D. thesis], Department of Computing Science and Mathematics, University of Stirling, Stirling, UK, [10] T. Zhao and S. Dong, Trust-GSM: a trust aware security model for multi-domain grid, in Proceedings of the 5th Annual ChinaGrid Conference (ChinaGrid 10), pp.43 47,Guangzhou, China, July [11] S. Naqvi and P. Mori, Security and trust management for virtual organisations: GridTrust approach, in IFIP Advances in Information and Communication Technology, pp , [12] C. Huang, H. Hu, and Z. Wang, A dynamic trust model based on feedback control mechanism for P2P applications, in Autonomic and Trusted Computing,L.T.Yang,H.Jin,J.Ma,and T.Ungerer,Eds.,vol.4158ofLecture Notes in Computer Science, pp , Springer, Berlin, Germany. [13] W. Wang, G. Zeng, and L. Yuan, Ant-based reputation evidence distribution in P2P networks, in Proceedings of the 5th International Conference on Grid and Cooperative Computing (GCC 06), pp , Hunan, China, October [14] J. Kennedy and R. C. Eberhart, Swarm Intelligence, Morgan Kaufmann, Boston, Mass, USA, 1st edition, [15] W. Stallings, Data and Computer Communications, Prentice Hall,NewYork,NY,USA,7thedition,2004. [16] S. D. Kamvar, M. T. Schlosser, and H. Garcia-Molina, The eigentrust algorithm for reputation management in P2P networks, in Proceedings of the 12th International Conference on World Wide Web, pp , ACM, New York, NY, USA, May [17] B.-J. Chang, S.-L. Kuo, Y.-H. Liang, and D.-Y. Wang, Markov chain-based trust model for analyzing trust value in distributed multicasting mobile Ad Hoc networks, in Proceedings of the IEEE Asia-Pacific Services Computing Conference, pp , December [18] S. Buchegger and J. Y. le Boudec, A robust reputation system forp2pandmobilead-hocnetworks, inproceedings of the 2nd Workshop on the Economics of P2P Systems, pp , June [19] S. Buchegger and J. Y. le Boudec, Performance analysis of the CONFIDANT protocol, in Proceedings of the 3rd ACM International Symposium on Mobile Ad Hoc Networking and Computing (MOBIHOC 02), pp , ACM, New York, NY, USA, June [20] F. Almenrez, A. Marn, C. Campo, and C. Garca, PTM: a pervasive trust management model for dynamic open environments, in Proceedings of the 1st Workshop on Pervasive Security and Trust,2004. [21] G. Zacharia, Trust management through reputation mechanisms, Applied Artificial Intelligence,vol.14,no.9,pp , [22] S. Songsiri, MTrust: a reputation-based trust model for a mobile agent system, Proceedings of the 3rd International Conference on Autonomic and Trusted Computing, Springer, Berlin, Germany, vol. 4158, pp , [23] J. Sabater and C. Sierra, Regret: reputation in gregarious societies, in Proceedings of the 5h International Conference on Autonomous Agents (AGENTS 01), pp ,ACM,New York, NY, USA, June [24] J. Carbo, J. M. Molina, and J. Davila, Trust management through fuzzy reputation, International Journal of Cooperative Information Systems,vol.12,no.1,pp ,2003. [25] T. Largillier and J. Vassileva, Using collective trust for group formation, in Proceedings of the 18th International Conference on Collaboration and Technology, pp , Springer, Berlin, Germany, [26] S. Al-Oufi, H. N. Kim, and A. El Saddik, A group trust metric for identifying people of trust in online social networks, Expert Systems with Applications,vol.39,no.18,pp ,2012. [27] R. Levien, Advogato s trust metric, 2000, advogato.org/trust-metric.html. [28] F.R.Easa,A.G.Bafghi,andH.Shakeri, Agroup-basedtrust propagation method, in Proceedings of the 2nd International Conference on Computer and Knowledge Engineering, pp , October [29] T. Schulz and I. Tjstheim, Increasing trust perceptions in the internet of things, in Aspects of Information Security, Privacy, and Trust, L. Marinos and I. Askoxylakis, Eds., vol of Lecture Notes in Computer Science, pp , Springer, Berlin, Germany, [30] J. P. Wang, S. Bin, Y. Yu, and X. X. Niu, Distributed trust management mechanism for the internet of things, Applied Mechanics and Materials,vol.347,no.350,pp ,2013. [31] JXTA get connected, 2013,

154 Hindawi Publishing Corporation International Journal of Distributed Sensor Networks Volume 2014, Article ID , 10 pages Research Article Webit&NEU: An Embedded Device for the Internet of Things Jialiang Wang, Hai Zhao, Jiuqiang Xu, and Yuanguo Bi College of Information Science & Engineering, Northeastern University, Shenyang , China Correspondence should be addressed to Yuanguo Bi; Received 16 June 2013; Revised 13 December 2013; Accepted 19 December 2013; Published 14 January 2014 Academic Editor: Naveen Chilamkurti Copyright 2014 Jialiang Wang et al. This is an open access article distributed under the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited. The Internet of Things (IoT) is regarded as the future generation Internet, which ranges from radio frequency identification (RFID) to the ubiquitous computing systems such as wireless sensor networks and mobile ad hoc networks. With the rapid development of IoT, designing an effective low-cost embedded terminal device for the IoT become very necessary. A new embedded device, Webit&NEU, and its reduced embedded real-time operating system used for IoT are implemented by our China Liaoning Province Embedded Technique Key Laboratory in this paper. Besides, related modules in terms of RFID technique, wireless communication, and network protocol are also provided in this paper. Compared with several current solutions of connecting devices and Internet, it has the advantages of good real-time performance, light weight, and low cost. Besides, this paper also puts forward a localization algorithm for the Webit&NEU, and experimental test results in terms of real-time system ability, network communication performance, and localization algorithm show that Webit&NEU can work well and meet the actual requirements of the IoT. 1. Introduction The IoT is a new-emerging technique. After the age of Internet and wireless sensor network, people are gradually entering the era of IoT. The IoT refers to uniquely identifiable objects (things) and their virtual representations in an Internet-like structure [1]. In fact, the IoT includes technologies such as RFID, sensors, and smart devices. The concept of the IoT has become more and more popular now. RFID is often regarded as a prerequisite for the IoT [2, 3]. Supposing all objects of daily life were equipped with radio tags, thus they could be identified, tacked, and managed by computers, and then they will certainly bring great convenience to people s lives. Embedded Internet device Webit2.0 can make the standard industry device have the ability of accessing the Internet, and it has successfully been embedded to fieldbus, industrial devices, and so forth, so people can feel free to access those devices through Internet. Webit2.0 has been successfully developed by our China Liaoning Province Embedded Technique Key Laboratory and has been large-scaly manufactured by China Shenyang Neu-era Information Technology Stock Co., Ltd [4]. It is worth noting that Webit2.0 has obtained the China s product patent (nos. ZL and ZL ). The design and implementation of embedded device Webit&NEU are based on Webit2.0. Besides the basic functions of Webit2.0, new functions such as RFID data processing and wireless date communication are also added to thewebit&neu,anditsmicrocontrollerisalsoupdatedto Atmega128LsothatitcanwellbeusedinIoT. This paper mainly makes three novel research contributions. (1) It implements a light-weight and low-cost embedded device Webit&NEU used for the IoT and its reduced operating systems kernel Webit&NEU OS written in AVR assembly language. (2) It implements data collection module, wireless communication module, and network protocol module efficiently. (3) It proposes an improved localization algorithm for the practical application of Webit&NEU. This paper is organized as follows. The second section presents the current embedded system communication

155 2 International Journal of Distributed Sensor Networks RFID reader Client RFID reader Internet Server RS485 concentrator Client RFID reader Figure 1: Connection method of wireless communication. Handheld RFID reader Radio ware Server Internet Figure 2: Connection method of distribution access. Client Client solutions and related analysis. The third section describes the design and implementation of Webit&NEU hardware system, its light-weight operating system kernel, and related performance analysis. Besides, the wireless communication module between different devices, RFID data collection module, and thetcp/ipprotocolmoduleareprovidedinthissection.the fourthsectionproposedanimprovedlocalizationalgorithm for Webit&NEU, and performance comparison in terms of localization error and localization time is described in this section. Concluding remarks are given at the end of the paper. 2. Current Embedded System Communication Solutions The current common methods used to collect, communicate, and manage data are shown as follows [5, 6]. (1) RS485 is adopted in consideration of its good capability of long distance communication. The implementation includes connecting all RFID readers by concentrator and linking them by server. In this method, the server handles the data and also supports Internet interface so as to make users access them remotely as Figure 1 shows. (2) Handheld RFID is used reader to read and write RFID card, and then system sends data to nearby server by the method of wireless transmission and implements resources sharing by connecting to Internet. This method is adopted by some applications having high portable requirements as Figure 2 shows. (3) RFID and PC are usually connected by RS232 in this method. As the limit of communication capability for RS232, it is suitable for short distance communication. PC filters and handles these data collected by RFID reader and then sends these data to the server as Figure 3 shows. FormostoftheRFIDproducts,theconnectionmethod between different devices is usually implemented by USB or serial port [7], which is not only inconvenient to expand, but also is unsuitable for remote data transmission, so this is the bottleneck for these devices to be applied to the IoT. So we engaged to develop embedded device features: high efficiency, light weight, and low cost used in the IoT. RFID reader RFID reader RS232 RS232 PC PC Server Internet Figure 3: Connection method of bus access. Client Client 3. Design and Implementation of Webit&NEU 3.1. Design of Webit&NEU Hardware. Webit&NEU uses the basic structure of Webit2.0. It can manage the data collected from the RFID reader and CC2530 and store the necessary data to memory. Besides it can assemble HTTP packet according to different applications and then send them to Internet by the RTL8019AS. The product appearance of Webit2.0 is shown as Figure 4 shows (compared with one yuan coin). Its corresponding interface parts are described as below: (1) 14-bit full-duplex User I/O (TTL electrical level), (2) power supply interface (DC 9V), (3) network interface: IEEE802.3/RJ-45, (4) network data signal lights, (5) in-system programming interface. And its technique feature and performance are provided, respectively, as below. Technique feature: (i) no gateway and PC needed, (ii) independent design of hardware and software architecture, (iii) independent intelligent, (iv) independent function of network server, (v) independent addressability, (vi) embedded design, (vii) low cost, (viii) small in size.

156 International Journal of Distributed Sensor Networks Tag Tag. Tag Tag Webit and NEU RFID reader Webit2.0 CC2530. Webit and NEU RJ-45 Internet User User. 1 Tag. Tag RFID reader RJ-45 Webit2.0 CC2530 User Figure 4: Product appearance of Webit2.0. Figure 5: Composition architecture of Webit&NEU. Performance of Webit2.0: (i) Atmel AVR RISC processor, (ii) customized definition Web page, (iii) customized definition CGI programming, (iv) duplexing I/O interface with 14 bit TTL electrical level, (v) UART interface of TTL electrical level supports up to bps, (vi) 10 M Ethernet interface (RJ-45), (vii) in-system Programmability (ISP), (viii) protocol supported: ARP, ICMP, IP, TCP, and HTTP, (ix) Network interface: IEEE 802.3/RJ-45. Webit&NEU can not only implement wireless communication with each other by CC2530, but also send the data collected from RFID reader to the Internet through RJ-45 interface as Figure 5 shows. Thus while devices are embedded with the Webit&NEU, then they can be used to smartly identify, locate, track, monitor, and manage these devices having RFID tag. The Webit&NEU well inherits good feature such as real-time ability from Webit2.0, and the most important is that the design goal for Webit&NEU is the implementation of light weight and low cost. RFID reader collects data from devices having tags so as to identify devices [8]. The RFID data collection module, CC2530 wireless communication module, and Webit&NEU microprocessor module form the whole Webit&NEU hardware, and architecture illustration between Webit&NEU microprocessor module and RFID function module is shown as Figure 6 shows. The Webit&NEU microprocessor controls the RFID reader to achieve the function of data collection and management, and it mainly includes the below 4 procedures. (1) Initialization. Webit&NEU microprocessor implements the initialization of RFID reader by sending specific command (01H) to it. (2) Sending Commands. Webit&NEU microprocessor sends a request command (02H) to RFID reader so as to indicate that it will send commands, and RFID reader will return a response command (10H) to the microprocessor to show it has already prepared for receiving commands. After executing the above procedures, the microprocessor of Webit&NEU begins to send commands. (3) Data Verification. Thisisusedtomakesurethedata received are correct. It is implemented by the method of XOR operation. If this procedure is correct, then the system will execute data handling program. If not, the system will set the Flag in response frame as 1, so as to imply RFID reader that the data is wrong and order RFID reader to send it again. (4) Data Filtering. While microprocessor of Webit&NEU receives data from RFID reader, it will judge whether the data is anticipated according to message ID, number of data bytes, and data verification. If not, then the system will filter the wrong and incomplete data The Reduced Webit&NEU Operating System Kernel The Design and Implementation of Webit&NEU OS. A real-time operating system must have the ability to respond to external events timely. In order to coordinate and execute tasks effectively, Webit&NEU OS system provides the following services: interrupt handling, intertask communication, task synchronization, memory management, timing services, task priority assignment, and so on. All of these functions implemented by the unique design of system call can easily execute real-time applications. System call is a program supported to user by system, while users need to implement certain system functions; what they need to do is to call the accordingly system call in their application program, so it is convenient for users to use the system kernel. All the system calls are written in AVR assembly language. As one of the important advantages for AVR assembly

157 4 International Journal of Distributed Sensor Networks Power supply Crystal oscillator Power supply Crystal oscillator Address wire Webit2.0 Serial port MCU Control circuit Data wire Control wire RF writing/ reading chip Highfrequency filtering circuit Antenna matching circuit Antenna coil Network interface Expansion interface Figure 6: Hardware architecture between Webit2.0 and RFID. language is that it can easily and flexibly operate a certain storage location in memory, even its certain bit of the storage location, so system can handle task, message, interruption, timing, and so on effectively. So the Webit&NEU OS system can well be used to effectively manage data collection, wireless communication in the IoT [9, 10], and it has successfully been developed by our China Liaoning Province Embedded Technique Key Laboratory [11]. The system calls constitute the basic operating system and their brief descriptions are provided as follows. (1) Task management services: (i) CREATE TASK: creating and scheduling a task dynamically, (ii) DELETE TASK: deleting certain specific task, (iii) GET FUNCTION ID: obtaining task function ID (value ranges from 00 to 0FFH), (iv) SUSPEND: suspending a task being executed. (2) Intertask communication services: (i) ALLOCATE: allocating buffer space where a task creates a message to be sent, (ii) SEND MESSAGE: sending certain message to specific task, (iii) WAIT MESSAGE: allowing a task to wait for a message to come, (iv) DEALLOCATE: returning certain specific buffer to the system buffer. (3) Internal memory management services: (i) GET MEM: getting the address of certain memory having specific length is currently available in the system, (ii) RELESE MEM: returning certain memory with specific length to the system memory pool. (4) Interrupt-handling services: (i) DISABLE INTERRUPT: disabling some interruptions, (ii) ENABLE INTERRUPT: reenabling disabled interruptions, (iii) WAIT 3: synchronizing tasks with interruptions. (5) Timing services: (i) SET INTERVAL: setting a time interval after the interval event occurred, (ii) WAIT: waiting for interval event or timeout to occur. The design and implementation of data variable units and tables in the memory block bring great convenience for system to handle task information and its responding priority, interruption, message, timing, and so on. Besides, while designing the system calls, the data storage units and tables are all stored in the internal SRAM, so the speed in terms of task switching and information handling is improved to some extent. By using the above system calls, system can well synchronizeandschedulemultitasksintheiot.althoughonlyone taskcanbeexecutedontheprocessoratanygiventime,the multiplexingofalltasksmakesitappearasifallofthetasks are running simultaneously The Performance Test of Real-Time Ability about Webit&NEU OS. In order to evaluate the real-time performanceofwebit&neuos,wemainlytestedtwoperformance parameters of task switching time and max interruption inhabit time [12, 13] and then compared them with Webit5.0 OS (The fifth-generation operating system designed for Webit2.0) and TinyOS We chose the TinyOS as the comparison system because it also has relatively small kernel code size compared with the other operating systems. TinyOS is an open source, BSD-licensed operating system designed for low-power wireless devices, such as those used in sensor networks, ubiquitous computing, personal area networks, smart buildings, and smart meters. It is written in the nesc programming language as a set of cooperating tasks and processes. TinyOS started as a collaboration between the University of California, Berkeley, in cooperation with Intel Research and Crossbow Technology. A worldwide community from academia and industry uses, develops, and supports the operating system as well as its associated tools. TinyOS is officially released on August 20, TinyOS includes support for updated msp430-gcc (4.6.3) and avr-gcc

158 International Journal of Distributed Sensor Networks 5 Task switching time (μs) Webit and NEU OS Maximum Minimum Average Webit5.0 OS TinyOS Figure 7: Experimental test results of task switching time (μs). Max interruption inhibit time (μs) Wetib and NEU OS Maximum Minimum Average Webit5.0 OS TinyOS Figure 8: Experimental test results of max interruption inhibit time (μs). (4.1.2). A complete 6lowpan/RPL IPv6 stack. Support for the ucmini platform and ATmega128RFA1 chip [14]. We used the tools of logic analyzer (TLA603), arbitrary waveform generator (AWG2021), and digital storage oscilloscope (TDS1012). In the test programs, we added some control codes by transforming the value of high and low levels of pulses through I/O interface. For each experiment below, we tested 20 times totally, kept down the maximum and minimum, and then calculated their average as Figures 7 and 8 show. For the Webit&NEU OS, the reduction of task switching time and max interruption inhabit time is 8.41% and 13.44% separately compared with Webit5.0 OS, which means Webit&NEU can not only process task switching more quickly, but also fast respond to external interruptions. Compared with TinyOS 2.1.2, we can know that the task switching time of Webit&NEU OS is shorter, mainly because of the special design structure of TASK READY TAB (16 memory units ranging from 0283H to 0292H are used to store related information of every ready task). The low 4 bit stores the task priority, while the high 4 bit of each unit stores the task ITD. When a task is inserted into TASK READY TAB, it will be put in suitable place according to its priority so as to make sure all tasks in the task ready table are sorted by task priority. But for the max interruption inhibit time, TinyOS is better, because the two categories of interruption handling program (COMMON SERVER and TIMER0 SERVER) take longer time for the Webit&NEU OS. But compared with Webit5.0 OS, the Webit&NEU OS is better in terms of both task switching time and max interruption inhibit time RFID Data Collection Module. The RFID terminal function module of Webit&NEU can be summarized as below. The data collected from the RFID reader are sent to specific server by Internet so as to make the devices networked. In order to implement these functions, the architecture of terminal function module is mainly composed of network driver, data collection, and device driver module. The three modules and their composition are shown in Figure 9. Network driver module is the key to ensure RFID terminal devices are networked. In order to implement the connection between devices and Internet, the TCP/IP protocol is prerequisite. This system also implements the function of Web server so as to make clients access these devices by browser. RFID data collection module is the input data part for the whole system. This module implements communication between different devices by RFID technology. In order to guarantee the correction and security of data collection and transmission, firstly there should be a procession of establishing link before data frames transmission, and system uses handshake principle to implement synchronization between sender and receiver. Secondly, the data frames used tocommunicateshouldincludenotonlytheactualdata,but also the necessary control and check bit, which means data shouldbeverifiedsoastoguaranteethedatacorrection, so sometimes the data retransmission operation is required. Finally,thesystemfiltersthecontrolandcheckbitandthen obtains the effective data. DevicesdrivermoduleisthebasicpartoftheRFID terminal devices. This layer masks the complexity of microcontroller underlying hardware and supports simple interface to applications. This module includes Webit&NEU system initiation, serial port driver, and network driver Communication between Different Webit&NEU Devices. The basic communication mechanism for the Webit&NEU is sending and receiving the format message through CC2530 during different Webit&NEU devices. The implemented format of data packet includes link types, message length, bit identification, source/destination task, order/response identification and data as Figure 10 shows, which aims to implement wireless communication efficiently [15]. The first 4 bytes of the data packet is preamble which is used to implement the packet synchronization, and the

159 6 International Journal of Distributed Sensor Networks Embedded RFID terminal Network driver module Data collection module Device driver module Webserver Thin TCP/IP protocol Data verification Data write/read Data filtering Serial port driver System initiation Network card driver Figure 9: Architecture about RFID terminal function module of Webit&NEU. Bit: RSSI VAL CRC OK 6 0 LQI VAL Byte: n 2 Frame Frame control sequence Address Payload FCS MAC header (MHR) MAC MAC footer (MFR) Byte: (0to 20 + n) payload Preamble SFD Length MAC protocol data unit (MPDU) Figure 10: The communication format of data packet. following byte is start-of-frame delimiter (SFD), whose value is 0A7H used to indicate the beginning of the data packet, the next byte is the length of the data packet, and its low 7bitsisusedtosavethevalueoflength,soitsmaximum valueis127,thefollowingbytesarethephyservicedataunit (PSDU), and it is composed of MAC header (Including frame control, sequence number, and addressing number), payload, and MAC footer. While sending message between different Webit&NEU devices, the reduced message format is put into the payload of the data packet. In a word, the microprocessor of Webit&NEU controls thecc2530tosendandreceivedatasoastoimplement the wireless communication between different Webit&NEU devices Implementation of TCP/IP Protocol. As the storage resource of the ATmega 128L is relatively limited, so this paper aims to implement a reduced TCP/IP protocol so as to meet the basic requirements in the practical application. The design of TCP/IP protocol stack is using hierarchical structure. About link layer, it refers to the network card driver so as to implement the functions of receiving and sending Internet data frame, as this procedure depends on hardware address. Considering that IP protocol in the network layer is IP address, a dynamic translation between these two addresses is needed. System judges the type of the data packet if it is ARP request (value is 0806H), and then the system will call the ARP response module; if it is IP packet (value is 0800H), system will then upload the packet to network layer and then call the IP handling module. The system executes correspondingly operations according to the protocol value of IP packet existent in IP Input function; if the value is 0 01,thesystemwillturntoICMP execution module; if the value is 0 06,thesystemwillturn to the UDP handling module; if the value is 0 11,thesystem will turn to the UDP processing module. For the TCP protocol in the transport layer, if the port of TCP is 0 80,thesystemwillsendthedatatoHTTP server, and the HTTP server will respond to it after executing related operations. The whole data process flow in the TCP/IP is shown in Figure 11. In order to evaluate the network performance, we tested the parameters 20 times in terms of network response time, network upload rate and download rate. The maximum, minimum, and average values are described respectively in Figures 12, 13 and 14 shows. Network response time, network upload rate, and download rate are very important for evaluating the network performance, which can well reflect the network transmission velocity and real-time ability. The experimental test results show that Webit&NEU OS performs better than Webit5.0 OS in terms of network response time and network upload and download rate. Compared with Webit&NEU OS and TinyOS systems, Webit&NEU OS has little quicker network response time than TinyOS 2.1.2, but the network upload rate and download rate are slower than TinyOS In general, the network

160 International Journal of Distributed Sensor Networks 7 Beginning Receiving data frame Y ARP? N Y IP? N ICMP? N UDP? N TCP? N Y HTTP? N Y Y Handling and sending response Y Handling and sending ICMP response Handling and sending response Handling and sending ARP response Network upload rate (kbps) Webit and NEU OS Webit5.0 OS TinyOS Maximum Minimum Average Figure 13: Experimental test results about network upload rate (kbps). Network response time (μs) End Figure 11: Process flow of data packet. Maximum Minimum Average Webit and NEU OS Webit5.0 OS TinyOS Network download rate (kbps) Webit and NEU OS Webit5.0 OS TinyOS Maximum Minimum Average Figure 14: Experimental test results about network download rate (kbps). Figure 12: Experimental test results about network response time (μs). performanceofwebit&neucanmeetthebasicrequirement of data transmission so as to make sure that Webit&NEU can wellbeappliedtotheiot[6, 7]. 4. Localization Algorithm Designed for Webit&NEU One of the most important functions for this system is to manage the devices efficiently, so obtaining the position of certain device is very necessary for Webit&NEU. The development of Webit&NEU is devoted to make it well used in the practical application; thus, the research of their localization is very important, because it directly affected the communication ability and efficiency. Based on the traditional localization algorithm TL (used in the Webit2.0), we propose an improved localization algorithm ETL for Webit&NEU used in the IoT. In below discussion about localization algorithm, each Webit&NEU is regarded as a node. The localization of nodes in the two-dimensional space can directly be calculated by their distance between nodes. If only knowing the coordinates of the two reference nodes O 1 and O 2,itmaycauseuncertaintywhilecalculatingthe unknown reference node as Figure 15(a) shows; this is because the two circles form two intersections of P 1 and P 2, so it can not determine the accurate localization about the unknown node. Thus, at least three reference points are needed to determine the localization of the unknown node in two-dimensional space as Figure 15(b) shows. Calculating the distance between nodes above is under the assumption that the distance measurement is carried out with absolute precision; however, the measurement error is inevitable in the practical application environment. Supposing the error range is (0, ±s), while the practical distance of two nodes is r, then the measured distance range is (r, r ± s).

161 8 International Journal of Distributed Sensor Networks O 1 P 2 O 1 P O 3 P P 2 1 O 2 O 1 P 1 O 2 O 2 P 3 O 3 (a) (b) Figure 15: Analysis of node localization. (c) Due to the existence of error, the localization error will not formapoint,butasmallareaastheshadedareashowsin Figure 15(c). The localization error area caused by the three reference points of P 1, P 2,andP 3 is the region formed by the six arcs of F 1, F 2, F 3, F 4, F 5,andF 6 as Figure 16 shows. While the error value of s is small enough, then the error value can be approximated as the region formed by the hexagon of ABCDEF. Considerthefollowing: F 6 L 5 Q A 6 F 5 F Q 5 ε β 1,2 β 2,3 L 4 E P 2 L 3 area (Q 6 AQ 1 P) = 2area (Q 6 AP) = 2 ( 1 2 ε εtan β 1,2 2 ) P 1 β 3,1 Q 4 Q 1 P Q L 3 6 B Q 2 D area (ABCDEF) =2ε 2 (tan β 1,2 2 + tan β 2,3 2 + tan β 3,1 2 ) (tan β) =2tan β sec 2 β 0 (0 β π 2 ) F 4 F 1 L 1 C P 3 area (ABCDEF) =6ε (tan β 1,2 2 + tan β 2,3 2 + tan β 3,1 2 ) 6ε 2 tan β 1,2 +β 2,3 +β 3,1 6 =6ε 2 tan π 6 =2 3ε 2. So while the condition of β 1,2 =β 2,3 =β 3,1 =π/3is met, the localization error can reach the minimum value of 2 3ε 2. This means only when the place of the three reference nodes forms an equilateral triangle, the localization error can reach the smallest value [16]. While arranging reference nodes in the monitoring region, the above localization method canminimize the localization error. However, the three reference nodes of P 1, P 2,andP 3 should not be placed very close, because if the calculation distance of unknown reference nodes is too short, then the localization error will become very large. Because while the three reference points are geographically too close, itwillseemlikeapoint,soitcannotbeusedtolocatethe unknown node in this case. We can get the position of unknown node by using the improved localization algorithm as below. (1) F 3 Figure 16: Analysis of node localization error. Algorithm 1 (proposed ETL algorithm). Step 1. Calculate distance S ij (i =j) between any two nodes and store them to database. Step 2. Every node sends broadcast packet periodically, and each broadcast packet should include a message as the format of {ID, T Send, (a, b)}, ID is the identification of unknown node, T Send isthetimeofsendingmessage,and(a, b) is its coordinate. Step 3. After receiving the broadcast packet, system calculates the distance L i according to the time of sending message (T Send )andreceivingmessage(t Rec ) and then sends these dates in the database. Step 4. While database receives N messages totally, for each set of C 3 N, use the formula of cos a i,j =(L 2 i +L2 j +S2 ij )/2L il j tocalculatetheanglebetweenunknownnodeandtwoother reference nodes. Step 5. Use three nodes whose angle exists in the range of π/6 λ < α 1,2,α 2,3,α 3,1 π/6+λto calculate the position of F 2 L2

162 International Journal of Distributed Sensor Networks Localization error (cm) Localization time (ms) Number of Webit and NEU devices Number of Webit and NEU devices TL algorithm ETL algorithm TL algorithm ETL algorithm Figure 17: Comparison of localization error. Figure 18: Comparison of localization time. unknown node. For each eligible set in C 3 N, system calculates corresponding position of unknown node {H 1,H 2,...}. Step 6. Forallthepositionofunknownnode{H 1,H 2,...}, system calculates average value H avg. For TL algorithm used in Webit2.0, it gets the position by rough calculation firstly and then obtains the final value by repeated similar procedure. Compared with TL algorithm, the ETL algorithm we proposed avoids some repeatable process by the above specific filter. The subsequent experiments will evaluate the performance of localization error and localization time for the ETL algorithm. (a) The Test and Analysis of Localization Error. In order to locate the position of unknown devices, the two localization algorithmsarebothimplemented.wecomparedthetesting results by executing the Webit&NEU OS. The experimental region adopted is the indoor environment whose area is 4m 6 m. There are totally 17 Webit&NEU devices as the reference node, which are placed by the method of the above localization algorithm, and these devices form 20 equilateral triangles totally in this localization area. As the Figure 17 shows, while the number of Webit&NEU devices is 3, the localization error caused by the two algorithms is close, but, with the increase of the Webit&NEU devices number, the localization error of ETL algorithm decreases more quickly than that of TL algorithm. While the number of Webit&NEU devices reaches to 6, the localization error decreases smoothly for ETL algorithm, but correspondingly while the number of Webit&NEU devices reaches to 12, the localization error begins to decrease smoothly for TL algorithm. The comparison of the two algorithms verifies that the the ETL algorithm causes less localization error than TL algorithm; thus ETL algorithm is better to be used in the IoT because of its good localization accuracy. (b) The Test and Analysis of Real-Time Ability. In order to test the real-time ability ETL algorithm, we also implement ETL algorithm and TL algorithm separately based on Webit&NEU OS and evaluate the real-time localization ability while the number of Webit&NEU devices is increased continuously, and the experimental results measured are shown in Figure 18. As we can see from the above results, while the number of Webit&NEU devices is 3, 4, and 5, the localization time for these two localization algorithm is almost the same, but with the increase of the number of Webit&NEU devices, the localization time of ETL algorithm increases exponentially while ETL algorithm increases correspondingly smoothly. From the above analysis we can know that ETL algorithm is better than TL algorithm in terms of good-time ability and higher localization accuracy [17, 18]. Due to these features of good real-time ability and less localization error, thus ETL algorithm can be better executed for Webit&NEU in the practical application [19]. 5. Conclusions Webit&NEU is a new device designed to be used in the IoT in order to implement smart management of devices. This paper mainly introduces its hardware architecture, software system, wireless communication module, RFID module, and network protocol module. In order to evaluate its real application value, we improve traditional localization algorithm used for Webit2.0 and propose an ETL algorithm, and the experimental results show that ETL algorithm has good real-time ability and less localization error than the traditional one. Besides, experimental results in terms of task switching time, max interruption inhibit time, network response time, network

163 10 International Journal of Distributed Sensor Networks upload rate, and network download show that it works well and can also meet the basic application requirements for theiot.thewebit&neuoscanwellbeexecutedinsome experimental tests, but it still will be continuously optimized so as to better be applied to IoT. The design and implementation of embedded device Webit&NEU will provide a good solution for the IoT. The widely application of Webit&NEU will undoubtedly bring great convenience for people to implement smart management of devices. At the meantime, the design aim of light weight and low cost will certainly facilitate the widely usage of Webit&NEU. Conflict of Interests The authors declare that there is no conflict of interests regarding the publication of this paper. Acknowledgments This work is partly supported by the National Natural Science Foundation of China (no ), Fundamental Research Funds for the Central Universities of China under Grant no. N , Key Laboratory Project Funds of Shenyang Ligong University under Grant no kfs03, Educational Committee of Liaoning Province Science and Technology Research Projects under Grant no. L , and Cultivation Fund of the Key Scientific and Technical Innovation Project, Ministry of Education of China (no ). The authors thank Jan Vitek, Ales Plsek, and Lei Zhao for their help during studying at Purdue University from September 2010 to September The authors also thank the anonymous reviewers for their valuable comments. [9] R. Pellizzoni and M. Caccamo, Impact of peripheral-processor interference on WCET analysis of real-time embedded systems, IEEE Transactions on Computers,vol.59,no.3,pp , [10] J. Wu, J.-C. Liu, and W. Zhao, A general framework for parameterized schedulability bound analysis of real-time systems, IEEE Transactions on Computers, vol.59,no.6,pp , [11] Intel Corporation, IRMX 51 idcx 51 Distributed Control Executive User s Guide for Release 2.0, Order Number , Intel Corporation, Santa Clara, Calif, USA, [12] W. Dong, C. Chen, X. Liu, K. Zheng, R. Chu, and J. Bu, FIT: a flexible, lightweight, and real-time scheduling system for wireless sensor platforms, IEEE Transactions on Parallel and Distributed Systems,vol.21,no.1,pp ,2010. [13] J. Whitham and N. Audsley, Time-predictable out-of-order execution for hard real-time systems, IEEE Transactions on Computers,vol.59,no.9,pp ,2010. [14] [15]L.Sun,J.Li,Y.Chen,andH.Zhu,Wireless Sensor Network, Tsinghua University Press, Beijing, China, [16] Y. Zhou, The optimizing selection and error analysis of location reference nodes in smart space, Northeastern University, pp , [17] G.Cabri,L.Leonardi,M.Mamei,andF.Zambonelli, Locationdependent services for mobile users, IEEE Transactions on Systems, Man, and Cybernetics A, vol.33,no.6,pp , [18] J. Hightower and G. Borriello, Location systems for ubiquitous computing, IEEE Computers,vol.34, no.8,pp.57 66, [19] L. Yan, Y. Zhang, L. T. Yang, and H. Ning, The Internet of Things: From RFID to the Next-Generation Pervasive Networked Systems (Wireless Networks and Mobile Communications), Auerbach Publications, References [1] [2]G.D.AbowdandE.D.Mynatt, Chartingpast,presentand future research on ubiquitous computing, ACM Transaction on Computer-Human Interaction,vol.7,no.l,pp.29 58,2002. [3]G.Y.Xu,Y.C.Shi,andW.K.Xie, Pervasivecomputing, Computer Journal,vol.26,no.9,pp ,2003. [4] Z. Hai and C. Yan, Pervasive Computing,NortheasternUniversity Press, Shenyang, China, [5] A. H. Ho, Y. H. Ho, and K. A. Hua, Handling high mobility in next-generation wireless ad hoc networks, International Journal of Communication Systems, vol.23,no.9-10,pp , [6] S.-R. Yang and C.-W. Leong, A conformance test tool for next generation network applications and systems, International Journal of Communication Systems,vol.23,no.6-7,pp , [7] N. Bagherzadeh and M. Matsuura, Performance impact of task-to-task communication protocol in network-on-chip, Journal of Circuits, Systems and Computers, vol.18,no.2,pp , [8] K. Cho, S. Pack, T. T. Kwon, and Y. Choi, An extensible and ubiquitous RFID management framework over next-generation network, International Journal of Communication Systems, vol. 23, no. 9-10, pp , 2010.

164 Hindawi Publishing Corporation International Journal of Distributed Sensor Networks Volume 2014, Article ID , 11 pages Research Article iotsilo: The Agent Service Platform Supporting Dynamic Behavior Assembly for Resolving the Heterogeneity of IoT Euihyun Jung, 1 Ilkwon Cho, 2 and Sun Moo Kang 2 1 Deptartment of Computer Science, Anyang University, Jungang-ro, Buleun-myeon, Ganghwa-gun, Inchon , Republic of Korea 2 Division of Digital Infrastructure, National Information Society Agency, NIA Building, Cheonggyecheonno 14, Jung-gu, Seoul , Republic of Korea Correspondence should be addressed to Ilkwon Cho; ikcho@nia.or.kr Received 30 August 2013; Accepted 28 November 2013; Published 14 January 2014 Academic Editor: Luis Javier Garcia Villalba Copyright 2014 Euihyun Jung et al. This is an open access article distributed under the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited. Although a lot of researchers have painted a rosy picture of Internet of Things (IoT), there have been no widely accepted solution and related standards until now. To achieve the successful realization of IoT, the close collaboration of devices is the primary requisite. However, the heterogeneity of devices such as different hardware or network connectivity prohibits the realization of IoT. In order to overcome the heterogeneity issue, we suggested the agent service platform named iotsilo in which agents can communicate and cooperate on behalf of their devices. With this delegation approach, the iotsilo can support diverse devices without worrying about their differences. In designing an agent, several software design patterns are adopted to enable the agent to assemble behaviors for hiding the heterogeneity of devices. To investigate the effectiveness of the iotsilo, we developed eleven different types of the IoT devices to emulate real world things with Arduino, deployed the devices in both Korea and Japan, and then conducted three experiments. 1. Introduction A lot of research groups have predicted that Internet of Things (IoT) would be a core infrastructure to provide the future IT services enabling the smart society [1 3]. IoT is defined as a pervasive network infrastructure where various digital things including RFID embedded objects, sensors, or mobile phones are interconnected and communicate with each other. In the near future, several trillion devices would collaborate to serve people through the global machine-tomachine interaction provided by the IoT technology [4 6]. However, the research on IoT is still in its early stage, so thereisnoagreedandcommonsolutiontorealizeityet [7 9]. There are several issues to block the IoT realization; the most essential one is the heterogeneity caused by the various features of devices such as network types, identifier schemes, and interworking protocols [3, 4]. IoT needs a close collaboration of devices, but it is hard for the devices with the different features to interact with each other. For example, wireless sensors and RFID embedded devices cannot identify peers due to their different identifier schemes. Even, the same kinds of wireless sensors could not make a session if they have different network connectivities. Unfortunately, the current IoT enabling technologies such as wireless sensor networks (WSN) [10] and machine-to-machine (M2M) [11] cannot cope with these problems properly yet. Although WSN is consideredasthemaininfrastructureofiot[6], WSN usually consists of homogeneous devices sharing the same network typesandprotocols.asaresult,wsnoftenmakesisolated islands that cannot interoperate with each other, so it cannot work as the IoT technology by itself anymore. M2M has existed in the different form since the advent of computer networking automation such as telemetry or SCADA. Due to the stable cellular M2M communication and the sophisticatedm2mdatamodules,m2misconsideredas another prominent technology for IoT. However, M2M is driven under the telco business model, so it is inherently

165 2 International Journal of Distributed Sensor Networks lockedinthewalledgardenoftelcos.thatis,m2malsomakes a closed island similar to WSN except its size. In addition to this, M2M companies forced to use their proprietary M2M module to join their M2M networks, and thereby the third party groups had difficulty in joining it. For now, both WSN and M2M do not propose any particular ways to resolve the heterogeneity, whereas they stick to the monolithic design of the same hardware, network types, and even protocols [4]. In order to resolve the heterogeneity issue of IoT, the researchers of cyber-physical systems (CPS) have paid attention to the software agent and they believe that the agent technology would be a solution to hide nicely the heterogeneity of devices [12, 13].Intheproposedscenario,devices delegate their authority to the corresponding agents and then the actual collaboration is done in the agent level [12]. Since an agent identifies and communicates with other agents in the agent space, the differences of physical devices can be easily hidden. This seemed a good idea to use the agent technology for overcoming the heterogeneity, but a detailed structure or a reference model was not represented yet. Most of all, to hide the heterogeneity, the single type of agent has to contain all different functions of heterogeneous devices in it. It can be an intuitive approach, but it is impractical because all features and functions of various devices cannot be anticipated and implemented in the development phase. Additionally, even if this approach is adopted, whenever a new device appears or the functions of a device are changed, the agent must be modified in the source code and be redeployed again into the system. Therefore, new approach with which an agent transforms itself depending on the features of devices is needed. In order to fulfill the requirement, we proposed an agent architecture in which an agent assembles the needed behaviors for the corresponding device and links them into itscoreonruntimelikethelegoblock.sincethisarchitecture is much different from the ordinary software design, we adopted portable service abstraction (PSA) and dependency Injection (DI) design patterns. We also developed the agent service platform named iotsilo to evaluate the proposed agent architecture. In the iotsilo, various devices with different sensing functions and network connectivities delegate their authority to the corresponding agents. We conducted several experiments on the iotsilo after making various devices with Arduino [14] and deploying the devices in both Korea and Japan. The experiments showed that the proposed agent can assemble various behaviors depending on the corresponding devices and it enables acting for the totally different devices without affecting the agent itself at all. They also demonstrated that various devices with different network connectivities can collaborate through the cooperation of their agents instead of direct communication between heterogeneous devices. This evaluation indicated that the proposed agent model can be a novel method to overcome the heterogeneity of IoT. The remainder of this paper is organized as follows. The related research about the heterogeneity of IoT is stated in Section 2. In Section 3, the proposed platform is minutely described. This is followed by Section4 which discusses several experiments to show the effectiveness of the proposed platform. Finally, Section5 gives the conclusion. 2. Related Research There are several fundamental causes why existing technologies have failed to resolve the heterogeneity issue of IoT. First, devices have different identification schemes such as IPv6, RFID, or ZigBee address, so they cannot identify and locate others. In the conventional data network, a participant has to identify and locate peers using identifiers before it makes a session. IPSO [15], Auto-ID [16], and OID [17] research groups have been competing for occupying the major identifier in IoT, but any group does not answer the interoperability with other identifiers. Second, the different network connectivity inhibits devices from making an end-to-end session between them. For example, a device on a ZigBee network cannot make a session to another device on the Internet with IP connection. There have been some studies to address this [18, 19], but they partially solved the problem using a specialized gateway. Lastly, there is still no agreed common interworking protocol between heterogeneous devices. AlthoughJini,UPnP,HAVi,andsoforthweresuggestedas the interworking protocols, they all failed to unite devices or to interoperate with other standards. Since devices are made from many different manufacturers and they equip their own interworking protocols, they cannot exchange information at all even if devices manage to make a physical session between them. While existing technologies failed to show their abilities for the IoT realization, some research groups suggested the alternative architecture with a totally different approach. This architecture is called the IoT data platform [20]. Several platformssuchascosm(formerlypachube)[21] or Open.sen.se [22] are announced. In the IoT data platform, the IoT devices just have a responsibility to report their data using a predefined protocol while they ignore their heterogeneity such as the network types. The reported data can be used to provide the instant IoT services to its users regardless of the kinds of the reporting devices. The IoT data platform achieves some degree of success because it can hide the heterogeneity by separating devices and their data. COSM reports one million data feed into COSM per day, and the developers of COSM can easily make new IoT services with the reported data. However, this approach cannot provide a peer-to-peer collaboration between devices, because its purpose is to manipulate the transmitted data instead of accessing the devices directly. Separating devices and their data is a good attempt to make mash-up IoT applications, but the full-level IoT services such as the device collaboration cannot be achieved with this architecture. Besides, the CPS researchers expected that the software agent played an important role to realize IoT [12, 13]. It seems a novel approach, but they have not provided a reference model or a practical architecture until now. The most controversial issue of this approach is how an agent can hold vast functions of heterogeneous devices within it. It is impractical that the single agent contains every function of all

166 International Journal of Distributed Sensor Networks 3 devices in the world. The intuitive solution for this issue can bethecreationofallkindsofagentsfordevices.however, this solution also has a limitation. Different kinds of agents can suffer with the interoperability between them as their corresponding devices do. Behaviors iotsilo Behaviors 3. Architecture WiFi ZigBee 3.1. Conceptual Design. The heterogeneity of devices is natural and unavoidable because devices have to be designed to fit themselves to their applications and surroundings. Therefore, the forcing uniform scheme on the heterogeneous devices has no choice to fail. We decided that hiding the heterogeneity would be a key design factor instead of enforcing particular network types, identifier schemes, or protocols. For this purpose,weadoptedasoftwareagent[23]toactforadevice in a relationship of agency. An agent can communicate with other agents while it hides its device s network type, protocol, and other heterogeneous natures. Even while devices are located on the different networks and have different identifiers, they can painlessly communicate and collaborate with each other in the agent space. This approach enables the physical interaction of devices to be transposed onto the cyber interaction of agents, which is greatly flexible in responding to the heterogeneity of devices. Comparing to the severe limitation of the physical environment, the cyber environment allows devices a high level of freedom to collaborate with each other regardless of their different natures. The conceptual design of the iotsilo adopting the software agent is shown in Figure 1. In the iotsilo, a device delegates its authority to a corresponding agent. When a device wants to communicate or collaborate with other devices, it orders its corresponding agent to do it. Then, the corresponding agent performs a proper behavior on behalf of the device, such as reporting the device s data to the system or sending a request to other devices. If needed, the agent can communicate with other agents using the message bus in the iotsilo. Using the iotsilo, devices can make a logical channel between them using only their limited physical channels connected to their agents even if they cannot directly communicate with each other Agent Design. In designing agents, the extensibility is a major consideration that we have thought intensely, because an agent should handle many various devices without corrupting the running system. Generally, there are two main methods of the extensibility in the object oriented design (OOD): Inheritance and Composition. Although Inheritance is intuitive, the method is not recommended to make different kinds of classes inherited from the same parent class [24]. Inheritance should be used in the is-a relationship between classes. That means all classes using Inheritance are included in the same category. In the research, agents seem to be the same kind of agent which acts for a real world thing. However, agents are just the avatar of real world things which are totally different in their properties and operations. Therefore, the adoption of Inheritance is not proper for this Figure 1: The conceptual design of the iotsilo. case. Also, we were reluctant to choose Inheritance because of the several weaknesses: inheritance conflict, excessive dependency of intermediate parent classes, and compile time binding [24]. Comparing to Inheritance, Composition provides more flexibility. By dividing an agent into agent s core and its behaviors which perform the functions of its device, the agent can easily accept any kind of request from devices. However, the coherence between the agent core and its behaviors is still unavoidable even though Composition is adopted. In the research, the portable service abstraction (PSA) and dependency injection (DI) patterns are used to remove the coherence Portable Service Abstraction (PSA). In order to support every function of the various devices, the separation of the functions from the agent core is required. However, this separation leads to a new problem that an agent core has to know all method signatures of the device-dependent functions. That is, the agent core has no choice but to contain the name of all classes and method signatures in its source code as shown in Figure 2(a). Inthisstructure,whenever a new device is added, the agent core should be modified in the source code level and be redeployed. Therefore, a simple separation cannot be a solution in considering the vast diversity of devices. This problem has been referred to as the dependency in the OOD and the portable service abstraction (PSA) policy has been proposed to reduce the dependency between objects. The PSA recommends architects to use an interface rather than a concrete object for achieving low coherence between objects. In our design shown in Figure 2(b),devices functions are abstracted as a Behavior interface, which is implemented by actual behavior classes. Then, the agent core can call various device-dependent functions through the Behavior interface. Unlike the conventional system, the addition of new devices can be easily supported without affectingtheagentcoreandtherunningsystem.theresulting structure is to be the form of the Strategy design pattern. With the pattern, the agent core can act as a proxy of various devices by equipping various device-dependent classes which implement the Behavior interface Dependency Injection (DI). The PSA policy provides an agent to equip the behavior classes performing devicedependent functions without modifying the agent core, but

167 4 International Journal of Distributed Sensor Networks Temp Core -temp -camera -newdev gettemp() getcam()? +gettemp() Camera +getcam() <<interface>> Core Behavior process() -behavior +process() External config. New device Temp Camera New device +newop() +process() +process() +process() (a) (b) Figure 2: (a) The ordinary dependency. (b) The decreasing dependency by using the Behavior interface. <bean id= vibagent class= net.agentsilo.iotsilo.agent.thingagent scope= prototype > <property name= info ref= thinginfo /> <property name= behaviormap > <map> <entry key= report value= vib report /> <entry key= get data value= vib get /> <entry key= set cmd value= vib set /> </map> </property> </bean> <!-- Behavior Definition --> <bean id= vib report class= net.agentsilo.iotsilo.behavior.generalreport /> <bean id= vib get class= net.agentsilo.iotsilo.behavior.defaulthttpgetdata /> <bean id= vib set class= net.agentsilo.iotsilo.behavior.defaulthttpsetcmd /> Algorithm 1: The snippet of configuration file describing the vibagent and its behavior classes. the agent core still has to hold the behavior classes names in the source code because the instantiation of the behavior classes is unavoidable. We resolved this problem using the dependency injection (DI) [25]. The DI is a design pattern that allows removing hard-coded dependencies in the source code and injecting them from the external configuration. The DI ensures that the agent core does not know where devicedependent functions come from and it allows the selection among multiple implementations of the given interface at runtime. Using the PSA and the DI, the agent core can choose the proper behavior class from the external configuration without knowing the actual name of the behavior class as shown in Figure 2(b). In the snippet of the configuration in Algorithm 1, the vibagent acts for a vibration sensor and it has three directives: a report, a get data, and a set cmd. Generally, a directive indicates an agent s behavior open to the external entities such as other agents. The external entities can request a behavior to the target agent with the directive corresponding to the behavior instead of the actual class name of the behavior. With this mechanism, the same directive can be mapped to different behavior classes depending on the type of devices. The directives are linked to the actual implemented classes in the configuration file shown in Algorithm 1 and these classes are dynamically assembled into the agent core at runtime. For example, the report behavior is linked to the GeneralReport class that reports the vibration sensor s data to the iotsilo management console. In the proposed system, all the behavior classes should implement the Behavior interface which has a single method of process (object... values) and they are assembled into agents using the dynamic binding. In general, the Strategy pattern is enough to change behaviors depending on the types of devices. However, an agent has to support many functions of devices, so the agent maintains a behavior map which holds the pair of the directive key and the Behavior implemented class object value. In Figure 3, the vibagent s behavior map based on the configuration file in Algorithm 1 is described. When an external directive report is requested to the vibagent, the agent chooses the corresponding GeneralReport behavior class and then

168 International Journal of Distributed Sensor Networks 5 report Directive get data set cmd report Behavior vib get vib set vib report DefaultHttpGetData DefaultHttpSetData GeneralReport Behavior Execute Figure 3: The structure and process of the agent s behavior map. executes it. Using the method, agents can hold any kinds of devices behaviors unlimitedly. Common part { version : float alue, id : sender agent s id, 3.5. Protocols. In the iotsilo, two kinds of protocols are defined. First protocol is between a device and a corresponding agent and second is between agents. A protocol between a device and an agent should support exchanging various data with devices because devices can have their own unique properties and operations. However, if all requirements are defined in a single protocol interface, the protocol interface wouldbetoocomplex.itmeansthattheprotocolshouldbe not only extensible but also concise. In order to meet both conflicting requirements, we designed the protocol interface consisting of a common and acustompartasshowninfigure4. Inthecommonpart, version is the protocol version and id is the identifier allocated to a device and its agent. The common part is not unchanged and all agents handle it identically, but the processing of the custom part tagged as data can be varied depending on the devices nature. When a packet is received from its device, an agent processes the common part and it hands over the custom part to its corresponding behaviors using the given directive defined in the custom part. The arg inthecustompartcanbevariedbecausethehandlingmethod of the Behavior interface supports variable arguments. That is, agents leave the manipulation of the custom part to the indicated behavior, so the contents of the custom part can have a high degree of freedom. The protocol interface in Figure 4 is used in the communication flow from a device to an agent. In order to collaborate with other devices, the protocol should support the bidirectional communication, so it should have an interface called from an agent. In the opposite flow, the interface format is the same except for the number of directives. We decided that the protocol interface toward the device can have only a single directive in the custom part considering the processing power of an embedded device. The protocol interface between agents is similar to the one between a device and an agent as shown in Figure 5.The different part is that the protocol interface can only hold a single behavior. A device can issue multiple behaviors at the same time, but an agent only requests a single behavior at one time. The protocol interface between agents can also contain any kind of information in the custom part because the corresponding agent leaves actual processing of the custom part to the matching behavior. The id is a sender identifier Custom part } data : [ { directive : target directi e, arg : { args needed for directi e } } ] Figure 4: The protocol interface between a device and an agent. Common part Custom part { } id : sender agent s id, data : { } } directive : target directi e, arg : { args needed for directi e Figure 5: The protocol interface between agents. and a receiver identifier is omitted because each agent has its ownlisteningchanneltothemessagebusintheiotsilo.when a sender agent publishes a message to the target channel of the receiver agent, the message is asynchronously delivered through the message bus Message Bus between Agents. Between agents, multiple concurrent communication sessions could occur in the iot- Silo. Since all sessions are independent of each other, the ordinary request-response metaphor, for example, synchronous communication between agents, may cause the performance degradation and the deadlock of the system. Small amounts of mesh topology connections among agents are enough to halt the whole system if only a single agent delays a response. In order to avoid this catastrophe, we adopted a message oriented middleware (MOM) [26] to construct a message bus.

169 6 International Journal of Distributed Sensor Networks Publish { id : foo@silo.net, data : { } } Message bus Listen foo Physical channel Logical channel Figure 6: The message bus in the iotsilo. MOM relies on asynchronous message passing in which a message queue provides temporary storage when a destination peer is busy or is not connected. Therefore, a sender and a receiver do not need to connect to the network at the same time. It can solve problems with the intermittent connectivity. Even if a receiver fails for some reasons, senders can continue to send their messages unaffected. The messages will simply accumulate in the message queue for later processing when the receiver resumes. In the iotsilo, an agent is designed to have a listening channel named its identifier on the message bus. When an agent wants to send a message to a peer agent, the agent makes a message and it publishes the message with thepeeragent sidentifier.then,themessagebusintheiotsilo delivers the message to the destination agent asynchronously. This process is depicted in Figure 6. IoTisexpectedtobecomposedofatleastseveralmillion devices and various messages are exchanged in it. Therefore, the extensibility and the scalability are major considerations to choose MOM. In order to fulfill these design factors, we selected ActiveMQ [27] for the following reasons. First, ActiveMQ supports both string messages and blob message. In the current design, the protocol only uses string messages, but if the binary data is needed, the protocol can be easily extended by using the blob message. Second, ActiveMQ has a capability to build a cluster easily. The capability is important when the message bus of the iotsilo is required to be scaled up. The plan for the scalability of the iotsilo will be discussed in Chapter Experiments 4.1. Physical Things and Experiment Environment. In the research, the IoT devices were implemented using Arduino [14] in order to emulate real world things. Arduino is an open-source single-board microcontroller designed to make various electronic devices easily in multidisciplinary projects. Figure 7: The developed IoT devices. We made eleven different types of the IoT devices: a temperature sensor, a humidity sensor, an optical detector, a vibration sensor, a radioactive sensor, a camera sensor, a rainfall meter, a wind vane, a wind gauge, a speaker, and an illuminometer. The devices were connected on three kinds of networks: Ethernet, WiFi, and ZigBee. These totally different functions and network types enable the IoT devices to emulate real world heterogeneous things. The developed IoT devices are shown in Figure 7. In order to show the effectiveness of the iotsilo, the IoT devices were deployed at three locations: Anyang University and Pusan University in Korea and Waseda University in Japan. The points in Korea were connected to Korea Advanced Research Network (KOREN) and the point in JapanwasconnectedtoJapanGigabitNetworkExtreme (JGN-X). Each point was configured to have eleven devices

170 International Journal of Distributed Sensor Networks 7 Illuminometer Vibration sensor Optic detector KOREN/JGN Humidity sensor Temperature sensor ZigBee WSN Wireless AP Speaker Sink Ethernet WiFi Rainfall meter Wind vane Wind gauge Camera sensor Radioactive sensor Figure 8: The deployment structure of the IoT devices for experiments. as shown in Figure 8. In these experiments, we wanted to test three check points: the dynamic assembly of the various device-dependent functions into the agent s core, the collaboration between devices at a distant location with different network types, and the modification of behaviors without affecting other parts of the system Dynamic Assembly of Device-Dependent Functions. In modeling the agent, we focused on the agent s dynamic support of the device-dependent functions without modifying the source code of the agent itself. To evaluate this design purpose, we conducted an experiment in which the same kinds of agents act as a proxy of two different kinds of sensors: a camera sensor connected through the Ethernet and a temperature sensor through the ZigBee. The camera sensor reports binary image data to the corresponding agent which stores the data into a mass storage and renders the stored images to the management console. On the other hand, the agent of the temperature sensor stores the reported temperature data into a database and it renders thedataasabarchart.asshowninfigure9, each agent does not need to modify the agent core at all but just to load proper device-dependent behaviors in the external configuration as shown in Algorithm 2. At runtime, each agent loads device-dependent behaviors and assembles them into the agent itself as indicated in the configuration. This experiment showed that any kinds of devices can be easily supported bytheproposedagentwithoutthecodemodificationifthe corresponding device-dependent functions are indicated in the external configuration. After the experiment, we changed the report function from the CelsiusReport class to the FahrenheitReport class in the configuration of the agent for the temperature sensor. This subexperiment assumed the situation where a device changed its behavior after finishing the deployment of the IoT services. In this sub experiment, the reported temperature data is well interpreted as the Fahrenheit value without affecting other parts of the system and even the corresponding agent itself Collaboration of Heterogeneous Devices at Distant Locations. In this experiment, when an optical detector at Waseda University sensed the human body, it sent a request to a speaker at Anyang University to make a warning sound. Physically, the optical detector and the speaker were not directly connected and did not know each other s network status at all, but the optical detector communicated with the speaker using agents collaboration. The detector issued the report directive to its agent; then the agent chose the humandetection behavior class which sent a request message containing beep directive to the speaker agent. After receiving the request, the speaker agent chose the behavior corresponding to the beep directive. Then, the behavior issued a set cmd directive which contained a sound generation command to the speaker. The experiment gave the expected result although these devices were connected on different networks. The whole procedure is shown in Figure Addition of New Functions without Affecting Other Parts of the System. We pointed out that new devices would be

171 8 International Journal of Distributed Sensor Networks External config. iotsilo Directive report Behavior picture report Directive Behavior report cel report PictureReport FahrenheitReport CelsiusReport Figure 9: The agents assembly of device-dependent behaviors based on the external configuration. <bean id= temperatureagent class= net.agentsilo.iotsilo.agent.thingagent scope= prototype > <property name= info ref= thinginfo /> <property name= behaviormap > <map> <entry key= report value= cel report /> <entry key= get data value= temp get /> <entry key= set cmd value= temp set /> </map> </property> </bean> <bean id= pictureagent class= net.agentsilo.iotsilo.agent.thingagent scope= prototype > <property name= info ref= thinginfo /> <property name= behaviormap > <map> <entry key= report value= picture report /> <entry key= get data value= picture get /> <entry key= set cmd value= picture set /> </map> </property> </bean> <bean id= cel report class= net.agentsilo.iotsilo.behavior.celsiusreport /> <bean id= fah report class= net.agentsilo.iotsilo.behavior.fahrenheitreport /> <bean id= picture report class= net.agentsilo.iotsilo.behavior.picturereport /> Algorithm 2: The configuration of temperatureagent and pictureagent. often added and the already deployed devices would replace functions due to the dynamicity of IoT. In the experiment, we assumed that a radioactive sensor suddenly needed a warning sound for the high radioactivity level. However, the radioactive sensor did not have a sound function at the deployment time, so it needed a help from the speaker nearby. To add the function after the device deployment, we just only changed the radioactive sensor s report behavior from the GeneralReport behavior to the RadioActiveReport. The RadioActiveReport behavior sent a request to the speaker agent when it checked the high radioactivity level. In order to check this function, the radioactive beads were used as shown in Figure 11. The result showed that the addition of new function was successful without affecting other parts of the system. It means that the mash-up-style IoT services are always possible with the iotsilo. Additionally, though the radioactive sensor was connected with WiFi and the speaker was connected with

172 International Journal of Distributed Sensor Networks 9 iotsilo Directive Behavior ZigBee Set cmd set cmd beep speakerset beepbehavior Message Bus beep WiFi report Directive "report" Behavior humandetection Figure 10: The collaboration of heterogeneous devices located at distant location. Table 1: Comparison with other technologies. Factor for IoT iotsilo WSN M2M System integration with heterogeneous devices Good Moderate Poor Dynamic function extension Best Poor Poor Support of collaboration between heterogeneous Good Poor Poor devices Technology stability Moderate Good Good Scalability Poor Moderate Good Speaker Radioactive sensor Radioactive beads ZigBee,twodeviceseasilycommunicatedwitheachotherin spite of the difference of their network connectivity. 5. Evaluation and Future Research 5.1. Evaluation. In order to evaluate the effectiveness of the iotsilo, we compared the iotsilo with other IoT enabling technologies. As shown in Table 1, the iotsilo is much better than other technologies in supporting heterogeneous devices. The iotsilo can support various heterogeneous devices, but WNS and M2M have the limit of device types. The dynamic function extension is easily supported by changing the configuration of agents in the iotsilo, but other technologies should modify components in the source level. In WSN and M2M, it is difficult for heterogeneous devices to communicate with each other because the technologies force the devices to havethesamenetworkconnectivity.comparingtoit,the iotsilo successfully achieved the collaboration of devices on the different network connectivity by using the indirect communication containing conceptual directives. The iotsilo is superior to the conventional technology in resolving the heterogeneity issue of the IoT, but the iotsilo reveals some weaknesses of technology stability and the scalability due to its preliminary phase. Figure 11: The radioactive beads are put on the radioactive sensor and the sensor requests an alarm to speaker Federation of the iotsilos. From the evaluation, the scalability of the iotsilo should be improved. Since IoT is expected to be composed of at least several million devices, the scalability is essential property that all IoT platforms should support. The same numbers of agents corresponding to all devices have to be allocated in a single iotsilo, so the size of IoT depends on the capacity of the iotsilo. Generally, the methods of enhancing the scalability fall into two broad categories: scale-out and scale-up [28]. The scale-up is a simple way but it is not practical at all for the IoT environment because the number of devices can easily pass over the added resources. For this reason, the scale-out seems the only way to resolve the scalability issue of IoT. In the further study, we have a plan to design that multiple iotsilos can federate into a virtual single iotsilo in the manner of scale-out. Each iotsilo takes charge of each organization and delivers requests to other iotsilos when its devices want to collaborate with the devices belonging to the other iotsilo. Using this federation, the iotsilo can be extended to large scale easily Directory Service. In the current design, an agent has to know the target directive before calling it. That means

173 10 International Journal of Distributed Sensor Networks the directive should be contained in the behavior code. Since behaviors can be easily replaced in the configuration file, this structure does not severely affect the dependency between agents. However, if an agent can search other agents behaviors dynamically and understand semantically, the intelligent collaboration of agents may be possible. This kind of smart and automated collaboration of machines is another vision of the IoT. In the further research, we will study the semantic directory service with which every agent registers its behaviors and semantically searches others behaviors. For a long time, the related studies have been conducted in the field of the Semantic Web Services [29] and some meaningful methods such as SSWAP [30] have been produced. Therefore, it may not be difficult to make a directory service for IoT with the Semantic Web Services technology. Additionally, we also have a plan to create a new ontology which can describe agents behaviors and IoT services. 6. Conclusion Although many visions have been presented and expectation about IoT is growing rapidly, there has not been any solution for IoT yet. There may be many issues to prevent the realization of IoT, but the heterogeneity is the most essential one. In order to resolve the issue, we suggested the agent service platform, iotsilo. It enables agents to communicate and collaborate to make IoT services on behalf of their corresponding devices. The structures of agent and protocol were designed to hide the heterogeneity of devices and to extend easily the devices functions. With Arduino, various IoT devices were made to emulate real world things and they were deployed in both Korea and Japan. In order to show the effectiveness of the iotsilo, several experiments were conducted. The results showed that the iotsilo enables devicestocommunicateandcollaboratewitheachother regardless of their heterogeneous natures such as network types. The results also showed that the iotsilo enables agents to assemble or replace their device-dependent behaviors at runtime without corrupting the system. The iotsilo has shown a new way to resolve the heterogeneity issue for IoT, but there is some room for improvement because it is still in the experimental stage. For the further studies, we have a plan to design a directory service that enables agents to register and find behaviors semantically. In addition, we will study the scalability of the iotsilo to support a huge number of devices. Conflict of Interests The authors declare that there is no conflict of interests regarding the publication of the paper. References [1]T.R.Regina,T.Tome,andC.E.Rothenberg, Scenarioof evolution for a future internet architecture, in New Network Architectures,pp.57 77,Springer,2010. [2] D. Bandyopadhyay and J. Sen, Internet of things: applications and challenges in technology and standardization, Wireless Personal Communications,vol.58,no.1,pp.49 69,2011. [3] J. K. Ng, Ubiquitous healthcare: healthcare systems and application enabled by mobile and wireless technologies, Journal of Convergence,vol.3,no.2,pp.15 20,2012. [4] H. Sundmaeker, P. Guillemin, P. Friess, and S. Woelffle, Vision and challenges for realizing the internet of things, Cluster of European Research Projects on the Internet of Things, European Commission, [5] J. C. Augusto, V. Callaghan, D. Cook, A. Kameas, and I. Satoh, Intelligent environments: a manifesto, Human-Centric Computing and Information Sciences,vol.3,article12,2013. [6] L.Atzori,A.Iera,andG.Morabito, Theinternetofthings:a survey, Computer Networks,vol.54, no.15,pp ,2010. [7] F. Mattern and C. Floerkemeier, From the internet of computers to the internet of things, in From Active Data Management to Event-Based Systems and More, vol.6462oflecture Notes in Computer Science,pp ,2010. [8] L. Tan and N. Wang, Future internet: the internet of things, in Proceedings of the 3rd International Conference on Advanced Computer Theory and Engineering (ICACTE 10), pp.v5376 V5380, Chengdu, China, August [9] E. Cho and S. Helal, Expressive exceptions for safe pervasive spaces, Journal of Information Processing Systems, vol.8,no.2, [10] J.Yick,B.Mukherjee,andD.Ghosal, Wirelesssensornetwork survey, Computer Networks,vol.52,no.12,pp ,2008. [11] K. Chang, A. Soong, M. Tseng, and Z. Xiang, Global wireless machine-to-machine standardization, IEEE Internet Computing,vol.15,no.2,pp.64 69,2011. [12] M. Chen, J. Wan, and F. Li, Machine-to-machine communications: architectures, standards and applications, KSII Transactions on Internet and Information Systems, vol.6,no.2,pp , [13] M. Iqbal and H. B. Lim, Poster abstract: a cyber-physical middleware framework for continuous monitoring of water distribution systems, in Proceedings of the 7th ACM Conference on Embedded Networked Sensor Systems (SenSys 09), pp , November [14] D. Wilcher, Learn Electronics with Arduino,Apress,2012. [15] The IPSO Alliance, Technical Information, 2013, [16] E. Fleisch, What is the internet of things? An economic perspective, Auto-ID Labs White Paper WP-BIZAPP-053, 2013, WP-BIZAPP-53.pdf. [17] O. Dubuisson, Introduction to Object Identifiers (OID) and registration authorities, France Telecom Orange, 2013, [18] Q. Zhu, R. Wang, Q. Chen, Y. Liu, and W. Qin, IOT gateway: bridging wireless sensor networks into internet of things, in Proceedings of the IEEE/IFIP 8th International Conference on Embedded and Ubiquitous Computing (EUC 10), pp , Hong Kong, China, December [19] S. Hong, D. Kim, M. Ha et al., SNAIL: an IP-based wireless sensor network approach to the Internet of things, IEEE Wireless Communications,vol.17,no.6,pp.34 42,2010. [20] Postscapes, Internet of Things Platforms-Postscapes, 2013,

174 International Journal of Distributed Sensor Networks 11 [21] COSM, Cosm internet of things platform connecting devices and apps for real-time control and data storage, 2013, [22] Open.Sen.se, 2013, [23] H. S. Nwana and D. T. Ndumu, A brief introduction to software agent technology, in Agent Technology: Foundations, Applications, and Markets, pp , Springer, Wooldridge, Mo, USA, [24] K. Knoernschild, Java Design: Objects, UML, and Process, Pearson Education, [25] M. Fowler, Inversion of control containers and the dependency injection pattern, 2013, injection.html. [26] C. Edward, Message-oriented middleware, in Middleware for Communications, Q.H.Mahmoud,Ed.,pp.1 28,Wiley, Chichester, UK, [27] B. Snyder, D. Bosanac, and R. Davies, ActiveMQ in Action, Manning, [28] M. Michael, J. E. Moreira, D. Shiloach, and R. W. Wisniewski, Scale-up x scale-out: a case study using nutch/lucene, in Proceedings of the 21st International Parallel and Distributed Processing Symposium (IPDPS 07),March2007. [29] R. Studer, S. Grimm, and A. Abecker, Semantic web services: concepts, technologies, and applications, Springer, [30] D. D. G. Gessler, G. S. Schiltz, G. D. May et al., SSWAP: a simple semantic web architecture and protocol for semantic web services, BMC Bioinformatics, vol. 10, article1471, p.309, 2009.

175 Hindawi Publishing Corporation International Journal of Distributed Sensor Networks Volume 2013, Article ID , 9 pages Research Article Distributed Risk Aversion Parameter Estimation for First-Price Auction in Sensor Networks Xin An, 1 Shuo Xu, 2 Jiancheng Chen, 1 and Yuan Zhang 1 1 School of Economics and Management, Beijing Forestry University, No. 35 Qinghua East Road, Haidian, Beijing , China 2 Information Technology Support Center, Institute of Scientific and Technical Information of China, No. 15 Fuxing Road, Haidian, Beijing , China Correspondence should be addressed to Jiancheng Chen; chenjc1963@163.com Received 30 August 2013; Accepted 27 November 2013 Academic Editor: Luis Javier Garcia Villalba Copyright 2013 Xin An et al. This is an open access article distributed under the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited. Following the Internet, the Internet of Things (IoT) becomes a prime vehicle for supporting auction. The use of market mechanisms to solve computer science problems is gaining significant traction. More and more clues show that the bidders tend to be risk-averse ones. However, traditional nonparametric approach is only applicable for the case of risk neutrality in a centralized server. This study proposes a generalized nonparametric structural estimation procedure for the first-price auctions in the distributed sensor networks. To evaluate the performance of the aggregated parameter estimators, extensive Monte Carlo simulation experiments are conducted for ten different values of risk aversion parameters including the risk neutrality case in multiple classic scenes. Moreover, in order to improve the usability of the aggregated parameter estimators, some guidance is also given for real-world applications. 1. Introduction Auctions are suggested as a basic pricing mechanism for setting prices for access to shared resources, including bandwidth sharing in wireless sensor networks (WSNs) [1]. Following the Internet, the Internet of Things (IoT) [2, 3] becomes a prime vehicle for supporting auction [4]. Moreover,theuseofmarketmechanismstosolvecomputerscience problems such as resource sharing [1, 5, 6], load awareness [7], task allocation [8], and network routing [9 11], is gaining significant traction. In addition to the real-time concerns associated with auctions in distributed sensor networks (DSNs), privacy concerns are also very important [12 15]. Therefore, many protocols are proposed in the literature. For example, the protocol for a sealed-bid auction proposed by Franklin and Reiter [12] uses a set of distributed auctioneers and features an innovative primitive called verifiable signature-sharing. Recently, Lee et al. [14] put forward an efficient multiround anonymous auction protocol. When there are more than one party bidding the same highest price for auctioned objects, the protocol picks out the bidders offering the same highest priceinthefirstroundtothenextround. Risk aversion is used to explain the advertisers behavior under uncertainty. The auction model and the optimal mechanism design for risk-averse bidders have been studied by [16 18]. Within the private value paradigm, risk-averse bidders tend to shade less their private values relative to the risk neutral case, which often results in some overbidding [19]. More and more clues show that the bidders indeed tend to be risk-averse ones [20 22]. In recent years, in order to gain some insight from auction data, structural approaches, pioneered by Paarsch [23], have attracted extensive attention. Some of them rely upon a parametric specification of the bidders private values distribution, for example, the piecewise pseudomaximum likelihood estimation (PPMLE) approach [24, 25]. Laffont et al. [26] have proposed a simulated nonlinear least square estimation method, which allows general parametric specifications. However, all of the methods rely on some assumption on parametric structures. Without any parametric assumptions, Guerre et al. [27] have presented a computationally convenient nonparametric estimation procedure. But it is only applicable for the case of risk neutrality. Therefore, it cannot explain the agents behavior very well. To overcome this problem, this paper

176 2 International Journal of Distributed Sensor Networks O t to his/her bid b n,t for the auctioned object O t (t N T ),where b m,t is the mth player s bid for the auctioned object O t. The first-order condition for maximizing expected utility can be written as b 1,t b n,t b Nt,t t N T (N t 1) s (b n,t ) f (V n,t ;N t )(V n,t b n,t ) γf(v n,t ;N t ) =0. (1) Figure 1: The Schematic diagram for auction model. generalizes nonparametric estimation procedure, so that it canbeapplicableforriskaversioninthefirst-priceauction. Our previous work restricts us to a special case only with two different numbers of bidders [28] or to a centralized server [29], but in this study we lose these restriction to distributed sensor networks. The rest of the paper is organized as follows. After the risk-averse model for first-price auction with independent private value is briefly introduced in Section 2, a generalized distributed nonparametric estimation procedure is proposed for sensor networks in Section 3. In Section 4, extensive Monte Carlo simulation experiments are conducted, and Section5 concludes this work. 2. Preliminary: Risk-Averse Model A single and indivisible object O t (t N T = {1,2,...,T}) is sold through an auction to N t M ={m 1,m 2,...,m M } bidders with their respective bid b n,t (t N T,n N Nt = {1,2,...,N t }), wherem is a finite set with M 2elements. Thus, one can consider simultaneously T similar single item auctions. Figure 1 shows the schematic diagram for classic auction model. In this paper, the private value paradigm is considered, where every bidder has a private value V n,t (t N T,n N Nt ) for the auctioned object O t (t N T ). The private values V n,t s are drawn independently from a cumulative probability distribution (PDF) F(V;N t ),whichisassumedtobeknown to all bidders and is defined on compact support [V, V] with a probability density function (PDF) f(v;n t ). Intuitively, every bidder is potentially risk-averse with a von Neuman Morgensten utility function U( ) satisfying U ( ) > 0, U ( ) 0, andu(0) = 0. The bidders are symmetric in the sense that they share the same private value CDF F(V;m)and the same utility function U( ). Specifically, the distribution F(V;m) for the private values is a uniform distribution on [0, 1] with a constant relative risk (CRRA) utility function, U(x) = x γ with γ [0,1]. In this specification, 1 γis the coefficient of relative risk aversion, with γ=1corresponding to risk neutrality. Let b=σ(v) denote the equilibrium bid function. Under weak regularity conditions, the equilibrium bid function is strictly increasing and differentiable, so that its inverse s(b) = σ 1 (b) exists and inherits these properties [27]. Bidder n N Nt maximizes his expected utility E[Π n,t ]=[Pr(b n,t b m,t, m =n)]u(v n,t b n,t )=F(s(b n,t )) Nt 1 (V n,t b n,t ) γ with respect Since one can only access the equilibrium bid b n,t (t N T,n N Nt ) and cannot access private value V n,t (t N T,n N Nt ) in the real-world applications, it should be better to see the private value V n,t s as a function of equilibrium bid b n,t s. By rearranging (1), one can easily obtain V n,t =b n,t +γ F(s(b n,t );N t ) f(s(b n,t );N t )s (b n,t )(N t 1), t N T, n N Nt. Additionally, F(V n,t ;N t ) = V n,t, f(v n,t ;N t ) = 1,and (db n,t /dv n,t )=(1/s (b n,t )).Hencefrom(1), one can obtain (2) db n,t = N t 1 (1 b n,t ), t N dv n,t γ v T, n N Nt. (3) n,t To solve the differential equation (3), the following equilibrium bid function can be obtained: b n,t σ(v n,t )= N t 1 N t 1+γ V n,t, t N T, n N Nt. (4) Let G(b; N t ) and g(b; N t ) be the CDF and PDF of the bids with N t 1competing bidders, respectively. From (4), itisnotdifficulttoseethatbidb = σ(v) has the same CDF (cumulative probability distribution function) with V; namely, G(b; N) = F(s(b); N). In fact, Guerre et al. [27] have proofed that distribution function F( ; ) is unique and equivalent to G( ; ). Moreover, their PDFs also have some relevance; namely, g(b) = f(s(b))/σ (s(b)). Therefore, (5)can be simplified as V n,t =b n,t +γ G(b n,t ;N t ) g(b n,t ;N t )(N t 1), t N T, n N Nt. 3. Risk Aversion Parameter Estimation 3.1. Meta-Parameter Estimators. As suggestions by Campo et al. [22], let b (m k) α and b (m l) α denote the αth percentile of the CDF G( ; m k ) and G( ; m l ),respectively;theequation G(b (m k) α ;m k ) = α = G(b (m l) α ;m l ) holds for any different (5)

177 International Journal of Distributed Sensor Networks 3 number of bidders m k,m l M. Thus, equipped with percentile, (5)becomes { V α = { { G(b (m k) α ;m k ) +γ (m k 1)g(b (m k) α ;m k ) γα (m k 1)g(b (m k) α ;m k ), G(b (m l) α ;m l ) +γ (m l 1)g(b (m l) α ;m l ) =b (m γα l) α + b (m k) α b (m l) α =b (m k) α + (m l 1)g(b (m l) α ;m l ). Through simple arithmetic operations on (6), one can obtain the following estimator γ (m k,m l ) for risk aversion parameter γ: b (m k) α b (m l) α = γ (m k,m l ) α( 1 (m l 1)g(b (m l) α ;m l ) 1 (m k 1)g(b (m k) α ;m k ) ). On closer examination on (7), it is very surprising that G( ; ) disappears, which means that the estimators for γ have nothing to do with G( ;...). Whatismore,itisveryeasy to verify γ (m k,m l ) = γ (m l,m ) k (m k,m l M,m k =m l ).Inthis way, one can obtain M(M 1)/2 estimators for risk aversion parameter γ. Since these estimators cannot utilize the global information, but only local information, they are named metaparameter estimators in the work. In the next subsection, some weighted combinations of all meta-parameter estimators, called aggregated parameter estimators, will be introduced. These aggregated parameter estimators will be more helpful for the real-world applications Aggregated Parameter Estimators. It is very important to choose a proper estimator for risk aversion parameter in the real-world applications. But which one is better? Intuitively, it seems that some (weighted) combinations of all metaparameter estimators in Section3.1 may be appealing. In this study, two-level combinations are considered. Specially, for any m k M,theestimator γ mk, is first obtained from γ (m k,m ) l (m l M \m k ) using two kinds of weights: T ml /(T T mk ) and (m l T ml )/(L T m k T mk ),asseenintable 1. Here, L T = T t=1 N t, T mk and T ml are the times of auctions with m k and m l bidders, respectively. The main difference of Tables 1(a) and 1(b) is that the former only considers the times of auctions T m with m k bidders, but the latter considers both the times of auctions T m with m k bidders and the number of bidders m k. Thus, there are two inner aggregated estimators (6) (7) for each γ mk,. Similarly, one can combine γ mk, swithanother two types of weights: T mk /T and (m k T mk )/L T,asseenin Table 2.Thus,onecanobtainfouraggregatedestimatorsfor theriskaversionparameterγin total as follows γ a,a T mk T ml : γ (m k,m ) l, m k M T T T mk γ a,b T mk : m k M T m l M\m k m l M\m k γ b,a m k T mk : m k M L T γ b,b m k T mk : m k M L T m l T ml γ (m k,m ) l, L T m k T mk m l M\m k m l M\m k T ml T T mk γ (m k,m l ), m l T ml γ (m k,m ) l. L T m k T mk Now there are only four aggregated estimators for risk aversion parameters for users to choose from. However, in fact, there is no apparent reason to prefer one estimator to another, and many factors may influence one s choice: such as, auctions number and the number of bidders. In this situation, Monte Carlo simulations and methods [30, 31] can be utilized to examine the performance of each estimator. Please see Section 4 for more details. But before this, the next subsection will describe in detail the parameter estimation procedure in the distributed sensor networks Distributed Parameter Estimation Procedure. Our distributed parameter estimation procedure is inspired by MapReduce [32, 33], a programming model for processing large data sets with a distributed algorithm on a cluster. The schematic diagram for our procedure is shown in Figure 2.In fact, two-hierarchical Mapper-Reducer structure is adopted in our estimation procedure. More specifically, the whole estimation procedure is summarized as follows. Step 1. Each bidder with a sensor prepares his/her bid price for each of interested auctioned objects. And then he/she submits his secrete bid price and ID to Mapper server through distributed sensor networks, whose ID was assigned by an auctioneer. Step 2. The Mapper server will decide to transfer the corresponding bid prices and IDs to different auction server. Usually, auctioned objects will be divided into several groups and deployed in different auction servers. Step 3. At the end of auctions, each of auctioned objects will receive different number of bidders. In Figure 2,auction servers with the same color mean the same number of bidders. Now, auction servers will transfer all bids to PDF & CDF estimation server. Step 4. The PDF & CDF estimation server will estimate the G(b; m i ), g(b; m i ), F(V;m i ), f(v;mi ) for m M and V n,t for t N T, n N Nt. (8)

178 4 International Journal of Distributed Sensor Networks Table 1: Inner weights for aggregated parameter estimators. (a) m 1 m 2 m k m M γ m 1, γ m 2,. γ m k,. γ m M, 0 T m1 T m2 T T m1 T T m2 0.. d T m1 T T mk. T m1 T T mm T m2 T mk T T m1 T mk T T m2. d T T mk 0. d. d T m2 T T mm (b) T mm T T m1 T mm T T m2. T mm T T mk. T mk T T mm 0 γ m 1, 0 m 2 T m2 L T m 1 T m1 m k T mk L T m 1 T m1 m M T mm L T m 1 T m1 γ m 2,. γ m k,. γ m M, m 1 T m1 L T m 2 T m2 0.. d m 1 T m1 m 2 T m2 L T m k T mk. m 1 T m1 L T m M T mm m k T mk L T m 2 T m2. d L T m k T mk 0. d. d m 2 T m2 L T m M T mm m M T mm L T m 2 T m2. m M T mm L T m k T mk. m k T mk L T m M T mm 0 Table 2: Outer weights for aggregated parameter estimators. m 1 m 2 m k m M T m1 T m 1 T m1 L T T m2 T m 2 T m2 L T T mk T m k T mk L T Step 4.1. To estimate G(b; m i ), g(b; m i ) [29, 34], G (b; m) = 1 mt m 1 g (b; m) = mt m h g T m m t=1 i=1 T m m t=1 i=1 1 (b i,t b), m M, K g ( b b i,t ), m M, h g T mm T m M T mm where 1( ) is a indicator function, K g ( ) isakernelwitha compact support, and h g is a vanishing bandwidth. Step 4.2. To estimate F(V;m i ), f(v;m i ) [29, 34], F (V;m) = 1 mt m 1 f (V;m) = mt m h f T m m t=1 i=1 T m m t=1 i=1 1 (V i,t V), m M, K g ( V V i,t ), m M, h f L T (9) (10) where h f is a vanishing bandwidth, K f ( ) is a kernel function. Step 4.3. To estimate V n,t according to (5). It is worth noting that Steps 4.2 and 4.3 are optional. If one does not want to estimate them, he/she can just skip these two steps. That is why they are shown by dotted boxes in Figure 2. Step 5. The estimated values for G(b; m i ), g(b; m i ) will be allocated by another Mapper to several γ estimation servers. The corresponding servers will estimate the meta-parameter estimators by (7). Step 6. Another Reducer will calculate the four aggregated parameter estimators according by (8). 4. Experimental Design and Discussions In order to evaluate the performance of all estimators, extensive Monte Carlo experiments are conducted in this section Experimental Design. In all the simulation experiments, the private costs are drawn from a uniform distribution on [0, 1], and the bidders have a CRRA utility function, U(x) = x γ, γ [0,1]. The experiment takes ten different parameter values of relative risk aversion: γ i = 0.1 i (i = 1,..., 10) with γ 10 = 1.0 corresponding to risk neutrality. In order to simulate the real auction data as closely as possible, the experiment considers three kinds of sample sizes T: 400, 2000, and 8000, which corresponds to small, median,

179 International Journal of Distributed Sensor Networks 5 γ a,a γ a,b γ b,a γ b,b Reducer: γ estimation server γ m 1,m 2 γ m 1,m 3 γ m 1,m k γ m 1,m M γ m M 1,m γ m k,m l M γ estimation server γ estimation server γ estimation server Mapper: load balance Ĝ(b; m 1 ) Ĝ(b; m 2 ) ĝ(b; m 1 ) ĝ(b; m 2 ) Ĝ(b; m k ) ĝ(b; m k ) Ĝ(b; m M ) ĝ(b; m M ) F( ; m 1 ) f(b; m 1 ) F( ; m 2 ) f(b; m 2 ) F( ; m k ) F( ; m M ) f(b; m k ) f(b; m M ) {{ n,t } N t n=1 }T t=1 Reducer: PDF and CDF estimation server O 1 O 2 O 3 O 4 O 5 O T 1 O T N 1 m 1 N 2 m 2 N 3 m 1 N 4 m 2 N 5 m 3 N T 1 m 1 N T m 3 {b n,1 } m 1 n=1 {b n,2 } m 2 n=1 {b n,3 } m 1 n=1 {b n,4 } m 2 n=1 {b n,5 } m 3 n=1 {b n,t 1 } m 1 n=1 {b n,t } m 3 n=1 Auction server Auction server Auction server Mapper: load balance Bidders Figure 2: The schematic diagram for distributed risk aversion parameter estimation. and large sample, respectively. In each of these samples, the number of bidder m M couldtakeonfourdifferentvalues: 3, 6, 9, and 12. We investigate three different patterns for the design matrix and the probability distribution of the ms. Table 3 illustrates the detailed T m s and their corresponding Pr(m)s for the three different designs. In Design A, each m M was equally likely, while in Design B, large ms were more likely than small ones, and small ms weremorelikely thanlargeonesindesignc. Specifically, we first generate randomly private valuations for different designs from the uniform distribution. We then compute numerically bids using (4) with different values

180 6 International Journal of Distributed Sensor Networks Table 3: Design matrices of the T n s. Sample size m Design A T m Pr(m) Design B T m Pr(m) Design C T m Pr(m) Table 4: The estimation performance of the aggregated estimators for γ in Design A. Mean STD Lower quartile Medium Upper quartile γ 2 γ 5 γ 10 γ 2 γ 5 γ 10 γ 2 γ 5 γ 10 γ 2 γ 5 γ 10 γ 2 γ 5 γ 10 (a) T =400 γ a,a γ a,b γ b,a γ b,b (b) T = 2000 γ a,a γ a,b γ b,a γ b,b (c) T = 8000 γ a,a γ a,b γ b,a γ b,b for γ. Note that the random numbers for the experiments are generated using the multiplicative congruential method with modulus (2 31 1), multiplier 397,204,094, and initial seed 2,420,375. To minimize the impact of noise, the above procedure is repeated 1000 times. In our experiments, let K(u) = (35/32)(1 u 2 ) 3 1( u l), h g = 1.06 σ b (mt m ) 0.2, h f = 1.06 σ V (mt m ) 0.2,where σ b and σ V are the standard deviations (STD) of observed bids and private valuations, respectively. However, the kernel density estimator is biased at the boundaries of the support. Similar to [27],10%observedpseudo-privatevaluesnearthe boundaries are trimmed Results and Discussions. From (7), it is not difficult to see that the percentile α may influence the estimation performance for the risk aversion parameter γ. In order to minimize the influence α on the estimators, the percentile α is optimized to select α from {0.9, 0.91,..., 0.99},sothatSTD of estimators for γ is minimized. Tables 4, 5, and 6 report the estimation performance of the aggregated estimators for risk aversion parameter γ {γ 2,γ 5,γ 10 } with α for Designs A, B, and C, respectively. Mean, STD, lower quartile, medium, and upper quartile areusedinthepaper.onecanseethatallthemeansand medium of the aggregated estimators are near to the real values, and STDs are not large, especially for large sample size. This means that our distributed risk aversion parameter estimation procedure is feasible. In fact, one can observe similar phenomena for other cases. The corresponding results for other cases are available from the authors upon request. In order to improve usability of these aggregated estimators, Table 8 gives some useful suggestions based on γ =arg min 10 x,y {a,b} i=1 γx,y 0.1 i. (11) 0.1 i Let us take the third row in Table 8 as an example. If the sample size of one real-world application approaches to 2000 and the distribution pattern of m follows Design B, the aggregated estimator γ b,b is better than the other. To evaluate further the goodness of fit of the distributed risk aversion parameter estimation procedure, we also calculatethe2-normbetweentheestimatedprivatevaluations(5) and actual ones, defined formally as T l 2 = 1 mt t=1 N t n=1 ( V n,t V n,t ) 2. (12)

181 International Journal of Distributed Sensor Networks 7 Table 5: The estimation performance of the aggregated estimators for γ in Design B. Mean STD Lower quartile Medium Upper quartile γ 2 γ 5 γ 10 γ 2 γ 5 γ 10 γ 2 γ 5 γ 10 γ 2 γ 5 γ 10 γ 2 γ 5 γ 10 (a) T = 400 γ a,a γ a,b γ b,a γ b,b (b) T = 2000 γ a,a γ a,b γ b,a γ b,b (c) T = 8000 γ a,a γ a,b γ b,a γ b,b Table 6: The estimation performance of the aggregated estimators for γ in Design C. Mean STD Lower quartile Medium Upper quartile γ 2 γ 5 γ 10 γ 2 γ 5 γ 10 γ 2 γ 5 γ 10 γ 2 γ 5 γ 10 γ 2 γ 5 γ 10 (a) T = 400 γ a,a γ a,b γ b,a γ b,b (b) T = 2000 γ a,a γ a,b γ b,a γ b,b (c) T = 8000 γ a,a γ a,b γ b,a γ b,b The corresponding results are shown in Table 7,where the aggregated estimator for risk aversion parameter γ is set by Table 8.FromTable 7, one can easily see that private values canalsobeestimatedverywell,especiallyforlargesample. 5. Conclusions Auctions are suggested as a basic pricing mechanism for setting prices for access to shared resources, including bandwidth sharing in wireless sensor networks (WSNs). Following the Internet, the Internet of Things (IoT) becomes a prime vehicle for supporting auction. More and More clues show that the bidders tend to be risk-averse ones. However, traditional nonparametric approach is only applicable for the case of risk neutrality in a centralized server. In this paper, A generalized nonparametric structural estimation procedure is proposed for the first-price auctions in distributed sensor networks. In order to evaluate the performance, extensive Monte Carlo simulation experiments are conducted for ten different values of γ including the risk-neutral case in multiple classic scenes. Though there are no unique estimators for risk aversion parameter, four aggregated parameter estimators are obtained and some guidance is also given for real-world applications. Similarly, our previous work on PPMLE [25] is also extended to distributed sensor networks. Additionally, if our distributed parameter estimation procedure is equipped with localization technologies [35 38], it can be utilized in more mobile situations.

182 8 International Journal of Distributed Sensor Networks Table 7: The 2-norm between the estimated and actual private values. T Mean STD Lower quartile Medium Upper quartile γ 2 γ 5 γ 10 γ 2 γ 5 γ 10 γ 2 γ 5 γ 10 γ 2 γ 5 γ 10 γ 2 γ 5 γ 10 (a) Design A (b) Design B (c) Design C Table 8: The guidance on choice of the aggregated estimators. Sample size Design A Design B Design C Small γ a,b γ a,a γ a,a Medium γ a,a γ b,b γ a,b Large γ a,b γ a,b γ a,a Conflict of Interests The authors declare that there is no conflict of interests regarding the publication of this paper. Acknowledgments This work was funded partially by Fundamental Research Funds for the Central Universities: Research on Entrepreneurial Mechanism, Clusters and Organizing Mechanism of Peasants in Forest Zone under Grant no JGTD , Beijing Forestry University Young Scientist Fund: Research on Econometric Methods of Auction with their Applications in the Circulation of Collective Forest Right under Grant no BLX and Key Technologies R&D Program of Chinese 12th Five-Year Plan ( ): Key Technologies Researcher on Large Scale Semantic Computation for Foreign Scientific & Technical Knowledge Organization System, and Key Technologies Research on Data Mining from the Multiple Electric Vehicle Information Sources under Grant no 2011BAH10B04 and 2013BAG06B01, respectively. Our gratitude also goes to the anonymous reviewers for their valuable comments. References [1] S. Y. Shah, Useofauctionsinwirelesssensornetworks[M.S. thesis], Rensselaer Polytechnic Institute, [2] D. Uckelmann, M. Harrison, and F. Michahelles, Architecting the Internet of Things, Springer, New York, NY, USA, [3] C. Perera, A. Zaslavsky, P. Christen, and D. Georgakopoulos, Context aware computing for the internet of things: a survey, IEEE Communications Surveys & Tutorials PP,vol.99,pp.1 41, [4] Z. Chen, H. Huang, Y. E. Sun, and L. Huang, True-MCSA: a framework for truthful double multi-channel spectrum auctions, Cluster Computing,vol.12,no.8,pp ,2013. [5] J.Ostwald,V.Lesser,andS.Abdallah, Combinatorialauction for resource allocation in a distributed sensor network, in Proceedings of the 26th IEEE International Real-Time Systems Symposium, pp , IEEE Computer Society, Washington, DC, USA, [6] C.H.S.Carvalho,I.Woungang,A.Anpalagan,andS.K.Dhurandher, Energy-efficient ratio resource management scheme for heterogeneous wireless networks: a queueing theory perspective, Journal of Convergence, vol. 3, no.4, pp , [7] X. Li, N. Mitton, A. Nayak, and I. Stojmenovic, Achieving load awareness in position-based wireless ad hoc routing, Journal of Convergence,vol.3,no.3,pp.17 22,2012. [8] N. Edalat, Auction-based strategy for distributed task allocation in sensor networks [Ph.D. thesis], National University of Singapore, [9] X.Su,S.Chan,andG.Peng, Auctioninmulti-pathmulti-hop routing, IEEE Communications Letters, vol. 13, no. 2, pp , [10]M.Yoon,Y.K.Kim,andJ.W.Chang, Anenergy-efficient routing protocol using message success rate in wireless sensor networks, Journal of Convergence,vol.4,no.1,pp.15 22,2013. [11] H. R. Lee, K. Y. Chung, and K. S. Jhang, A study of wireless sensor network routing protocols for maintenance access hatch condition, Journal of Information Processing Systems, vol. 9, no. 2, pp , [12] M. K. Franklin and M. K. Reiter, The design and implementation of a secure auction service, IEEE Transactions on Software Engineering, vol. 22, no. 5, pp , [13] K.Peng, Asecurenetworkformobilewirelessservice, Journal of Information Processing Systems,vol.9,no.2,pp ,2013. [14] C. C. Lee, M. S. Hwang, and C. W. Lin, The design and implementation of a secure auction service, Journal of Discrete Mathematical Sciences and Cryptography,vol.10,no.4,pp , [15] K. Peng, An improved fast and secure hash algorithm, Journal of Information Processing Systems,vol.8,no.1,pp ,2012. [16] E. Maskin and J. Riley, Optimal auctions with risk averse buyers, Econometrica, vol. 52, no. 6, pp , 1984.

183 International Journal of Distributed Sensor Networks 9 [17] S. Matthews, Comparing auctions for risk averse buyers: a buyer s point of view, Econometrica,vol.55,no.3,pp , [18] A. Goel and K. Munagala, Hybrid keyword search auctions, in Proceedings of the 18th International Conference on World Wide Web, pp , ACM, New York, NY, USA, [19] J. C. Cox, V. L. Smith, and J. M. Walker, Theory and individual behavior of first-price auctions, Journal of Risk and Uncertainty, vol. 1, no. 1, pp , [20] J.K.Goeree,C.A.Holt,andT.R.Palfrey, Quantalresponse equilibrium and overbidding in private-value auctions, Journal of Economic Theory,vol.104,no.1,pp ,2002. [21] J. Lu and I. Perrigne, Estimating risk aversion from ascending and sealed-bid auctions: the case of timber auction data, Journal of Applied Econometrics,vol.23,no.7,pp ,2008. [22] S.Campo,E.Guerre,I.Perrigne,andQ.Vuong, Semiparametric estimation of first-price auctions with risk-averse bidders, Review of Economic Studies,vol.78,no.1,2011. [23] H. J. Paarsch, Deciding between the common and private value paradigms in empirical models of auctions, Journal of Econometrics,vol.51,no.1-2,pp ,1992. [24] S. G. Donald and H. J. Paarsch, Identification, estimation, and testing in parametric empirical models of auctions within the independent private values paradigm, Econometric Theory,vol. 12,no.3,pp ,1996. [25] X. An, S. Liu, and S. Xu, Piecewise pseudo-maximum likelihood estimation for risk aversion case in first-price sealed-bid auction, Computational Economics,vol.38, no.4,pp , [26] J.-J. Laffont, H. Ossard, and Q. Vuong, Econometrics of firstprice auctions, Econometrica,vol.63,no.4,pp ,1995. [27] E.Guerre,I.Perrigne,andQ.Vuong, Optimalnonparametric estimation of first-price auctions, Econometrica, vol.68,no.3, pp , [28] X. An, S. Liu, and S. Xu, Assess the goodness of fit for risk aversion parameter of first price auction via nonparametric method, Journal of Management Science and Statistical Decision,vol.6,no.4,pp.38 42,2009. [29] X. An, J. Chen, and Y. Zhang, Risk aversion parameter estimation for first-price auction with nonparametric method, in Proceedings of the 2nd International Conference on Ubiquitous Context-Awareness and Wireles Sensor Network, pp , Springer, Heidelberg, NY, USA, [30] G. A. Mikhailov, Parametric Estimates by the Monte Carlo Method,VSP,Utrecht,TheNetherlands,1999. [31] I. Manno, Introduction to the Monte-Carlo Method, Akadémiai Kiadó,Budapest,Hungary,1999. [32] J. Dean and S. Ghemawat, MapReduce: simplified data processing on large clusters, Communications of the ACM, vol.51,no. 1, pp , [33] A. Sinha and D. K. Lobiyal, Performance evaluation of data aggregation for cluster-based wireless sensor network, Human- Centric Computing and Information Sciences, vol.3,no.13,pp. 1 17, [34] R.O.Duda,P.E.Hart,andD.G.Stork,Pattern Classification, John Wiley & Sons, 2nd edition, [35] Z. Pei, Z. Deng, S. Xu, and X. Xu, Anchor-free localization method for mobile targets in coal mine wireless sensor networks, Sensors, vol. 9, no. 4, pp , [36] S. Xu, X. Qiao, L. Zhu, Y. Zhang, and L. Lin, Fast but not bad initial configuration for metric multidimensional scaling, Journal of Information and Computational Science,vol.9,no.2, pp , [37] X. Lu, P. Lio, P. Hui, and H. Jin, A location prediction algorithm for mobile communications using directional antennas, International Journal of Distributed Sensor Networks, vol.2013, Article ID , 10 pages, [38] C. H. Oh, Location estimation using space time signal processing in RFID wireless sensor networks, International Journal of Distributed Sensor Networks, vol.2013,articleid634531,8 pages, 2013.

184 Hindawi Publishing Corporation International Journal of Distributed Sensor Networks Volume 2013, Article ID , 7 pages Research Article An Efficient Adaptive Anticollision Algorithm Based on 4-Ary Pruning Query Tree Wei Zhang, 1,2 Yajun Guo, 2 Xueming Tang, 1 Guohua Cui, 1 Longkai Wu, 3 andyingmei 1 1 School of Computer Science and Technology, Huazhong University of Science and Technology, Wuhan , China 2 School of Computer, Central China Normal University, Wuhan , China 3 National Institute of Education, Nanyang Technological University, 1 Nanyang Walk, Singapore Correspondence should be addressed to Wei Zhang; zwccnu@163.com Received 7 September 2013; Revised 19 November 2013; Accepted 25 November 2013 Academic Editor: Luis Javier García Villalba Copyright 2013 Wei Zhang et al. This is an open access article distributed under the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited. In radio frequency identification system (RFID), the efficiency in which the reader identifies multiple tags is closely related to the methods to solve the collision of multiple tags. At present, a reasonable solution is the introduction of 4-ary query tree (or n-ary query tree) to reduce the collision time slots and additional query is used to decrease idle timeslots. The advantage of a 4-ary tree anti-collision algorithm is that it is able to reduce collision timeslots, but it also increases the idle timeslots. To reduce these excessive idle timeslots the 4-ary tree anticollision algorithm brings, an anti-collision algorithm based on adaptive 4-ary pruning query tree (A4PQT) is proposed in this paper. On the basis of the information of collision bits, some idle timeslots can be eliminated through pruning the 4-ary tree. Both theoretical analysis and simulation results support that A4PQT algorithm can significantly reduce recognition time and improve throughput of the RFID system. 1. Introduction In the Radio frequency identification (RFID) system, if multiple readers and multiple tags, or a single reader and multiple tags, transmit simultaneously in the readers working area, there are three types of collision as readers and tags operate on the same wireless channel. First, it is the frequency interference among readers which is due to the overlapping of two or more readers in the working area. The signals transmitted by these readers are causing mutual interference; thus the readers cannot read properly the tag data within the region. Second, when a tag appears in two or more readers working area, it will not know which readers to communicate with. Third, when many tags appear in the working area of one reader, these tags respond simultaneously to the queries of the reader, which will result in collision. Generally,thefirsttwotypesofcollisionarecalledreaderto-reader collision, and the third is known as reader-to-tag collision. When collision occurs, the reader cannot read the tag s data. How to reduce the number of collisions? How to reduce the messages that tag transmits repeatedly and reduce the communication overhead of the RFID system? Therefore, an efficient anticollision algorithm for identifying multitag is of great importance for the wireless RFID system. RFID system for solving multitag collision problem is currently divided into ALOHA-based anticollision algorithms andtree-basedanticollisionalgorithms[1, 2]. ALOHA-based anticollision algorithms such as Slotted ALOHA, framed slotted ALOHA, dynamic framed slotted ALOHA, and enhanced dynamic framed slotted ALOHA are on the basis of a probabilistic method, in which each tag selects randomly a time slot in response to the query of reader in the event of a collision. If there are a large number of tags, the collision will happen when some tags repeatedly back off after the collision, so there will be a long time for the tag to be identified. Tree-based anticollision algorithms, such as binary search algorithms [2, 3], query tree algorithms [4 10], and tree-splitting algorithms [11] are based on deterministic methods. In the tree-based anticollision algorithm, the reader iteratively queries a subset of tags which match a given property until all tags are identified.

185 2 International Journal of Distributed Sensor Networks In the binary tree anticollision algorithm, the reader sends a query to tags. Upon receiving the query from the reader, tags compare their own ID numbers to the query number. Tags respond only when their ID numbers are less than or equal to the query ID number. When there are multiple tags simultaneously to respond, it will cause a collision. Then the reader will change the query number and set the highest collision bit to 0. Other bits after highest collision are set to 1. The query will be repeated until it can identify each tag. Some studies have improved identification efficiency of the binary tree anticollision algorithm by reducing the number of times to repeat searching [2, 3]. Wu et al. [3] proposed a binary tree algorithm based on ALOHA. The algorithm uses a dynamic, adaptive, and split method to adjust the frame length and brings frame length closer to the number of tags, so that the efficiency of the algorithm is very closetotheoptimalvalues. Query tree (QT) algorithm uses the query prefix of reader to split tags into two groups. The reader transmits a query string called prefix and the tags compare the string with its ID to see whether any of its ID contains the prefix. Only those tags whose IDs have a prefix matching to the string respond to the query. If the match succeeds, the tag transmits its ID to the reader. Collision occurs when multiple tags have the same prefix. In this case, the reader forms two new query prefixes by appending the old prefix with a binary 0 or 1. The reader then repeats the query with a different new prefix, and these tags are divided into two different groups until the number of tags in a group is one. However, in the basic query tree algorithm, the reader gradually inquires bit by bit. If the reader sends a query and no tag s ID matches the query string, it will result in an idle time slot and increase the traffic and query time. Many researchers have improved the query tree algorithm. In the 4-ary query tree anticollision algorithm [6], the query prefix updates by increasing two binary bit each time and the query by the reader no longer constitutes a binary tree but splits into a 4-ary tree. The 4-ary query tree algorithm can reduce the collision time slots but increase the idle time slots. In some n-ary tree anticollision algorithms [1, 7], in order to reduce idle time slots once collision occurs, the reader will add an additional query to determine the status of the collision bit and avoid to query the nonexistent query branching. So it will avoid many idle time slots. This way can improve efficiency but adds an extra query time slot. Wu et al. [3]alsointroduce a bit collision detection mechanism based on basic query tree anticollision algorithm to reduce the idle time slots, and it is only suitable for mobile reader to identify tags. Tree-splitting anticollision [11] algorithm uses a random number generator to split tags into a plurality of subsets. Each tag has a random number generator and a counter and responds to the query of reader when the counter value is 0. The start value of the counter of each tag is 0. When the collision occurs, the reader broadcasts collision information. These tags just responding to the reader will generate a randomnumber0or1andaddtothecounter.thosetagsthat has remained silent will set their counter value plus 1. Treesplitting anticollision algorithms are similar to the slotted ALOHA algorithm. These algorithms are random and their recognition efficiency is low. The new enhanced anticollision algorithm [12]splitthetagsintom+1groups, from G 0 to G m, where m is the tag ID. A tag belongs to G k only if the number of bit 1 in its ID is equal to k. When a collision occurs, the reader divides these collided tags into two subgroups, namely, subgroup 0 and subgroup 1. Each group or subgroup uses two counters, namely, C 1 and C 0, which represent the number of 0 and 1 in the remaining tag IDs. When the C 0 or C 1 is equal to 1 in a group, the reader can identify the tags in the groups. In order to reduce idle time slots the 4-ary tree brings, this paper proposes an adaptive 4-ary pruning query tree (A4PQT) anticollision algorithm, which adaptively prune the idle time slot branches without any additional query, so it greatly improves the identification efficiency of RFID system. The rest of the paper is organized as follows. Section 2 introduces the basic idea to design A4PQT. Section 3 presents A4PQT. Section4 describes the performance of the proposed algorithm. Section5 analyzes the experimental results of A4PQT. Section 6 compares A4PQT with query tree algorithm (QT), collision tree algorithm (CT) [4], and the improved 4-ary query tree algorithm (I4QTA) [6]. Section 7 draws some conclusions. 2. Optimal Split Tree for Anticollision Algorithm In order to improve the recognition efficiency, many anticollision recognition algorithms have been proposed based on the binary tree, 4-ary tree, and n-ary tree. There are three types of time slots in the identification process of treebased anticollision algorithms: readable time slots (one tag response), collision time slots (multiple tags response), and idle time slots (no tag response). If a tree-based anticollision algorithm in identifying the same number of tags requires minimum time slots, then the algorithm is optimal. Suppose there are five tags, which are 0100, 1001, 0010, 1011 and 0110, the tree-split process of anticollision algorithm to identify multiple tags by using a binary tree and 4-ary tree is shown in Figure 1. AscanbeseenfromFigure 1, adataframeofthetreebased anticollision algorithm can be expressed by a tree. A data frame is constituted by a number of time slots and each time slot corresponds to a branch of the tree. For an anticollision algorithm, the more total time slots are to identify a certain number of tags, the lower identification efficiency the algorithm will bring. In Figure 1(a), thereare fivetagstobeidentified.thealgorithmthatusesabinary tree-splitting method requires five collision time slots, one idle time slot and eleven total time slots. While in Figure 1(b), by using a 4-ary tree method to identify the same five tags, the algorithm requires three collision time slots, five idle time slots and a total of thirteen time slots. As can be seen, the binary tree-splitting method to identify the same number of tags will result in a relatively large collision time slot and a relatively small idle time slot the anti-collision algorithm can reduce the collision time slots by using the 4-ary tree splitting method, but it also increases the idle time slots.

186 International Journal of Distributed Sensor Networks Collision timeslots Readable timeslots Collision timeslots Readable timeslots Idle (a) Binary query collision tree Idle (b) 4-ary query collision tree Figure 1: The process of binary tree and 4-ary tree anticollision algorithm for identifying tags. The total time slots of a n-ary tree anticollision algorithm and [5] t (m) =1+B L=0 B L [1 (1 B L ) m mb L (1 B L ) m 1 ], where B isthenumberofbranchesofthetree,m is the number of tags, and L is the current level. When B value is different, the total time slots of an anticollision algorithm are different. It has been proved that when the B value is 3, the total time slots are the minimum; that is to say the identification process is most efficient, but it is not possible to construct a 3-ary tree. The advantage an anticollision algorithm has by using n-ary tree is that it is able to reduce collision time slots, but it increases the idle time slots. The basic idea of this paper is to take advantage of some information to cut some branches of the 4-ary tree (idle time slots). This corresponds to the way the 3-ary tree splits, which can guarantee minimum total time slots and greatly improve the anticollision recognition efficiency of 4-ary tree algorithm Ary Pruning Query Tree Anticollision Algorithm A4PQT uses tag IDs to split tags into four groups. The reader transmits a query prefix and the tags in the work area of the reader respond to the query. Manchester code is used to detect collision bits. If a collision occurs, the reader updates query prefix. From the beginning of the highest collision bit, a collision node is split by a 4-ary tree and A4PQT will cut some idle time slot nodes according to the characteristics of collision bits. Assuming X denotes the collision bit, the pruning principles are as follows. (1) (1) If the mode is the highest collided bit +0, the two branches 01 and 11 can be cut. (2) If the mode is the highest collided bit +1, thetwo branches 00 and 10 can be cut. (3) If the mode is the highest collided bit +X, dono pruning. In the first case, according to the quadtree splitting method, the node 01 and node 11 are apparent idle time slot nodes. Similarly, in the second case, node 00 and node 10 are idle time slot nodes, and the two nodes can all be pruned. In the third case, since next the highest collided bit is also a collision bit, it is not possible to determine whether there is an idle time slot node and we cannot cut any branch. The A4PQT algorithm consists of rounds of queries and responses. In each round, the reader transmits query prefix. Tags receive the query prefix and check whether their own IDs contain the same prefix. If the tag s ID contains the query prefix, the tag sends its ID except the part which is the same as the received prefix. If there is a collision, the reader updates query prefix based on the principle of pruning. For the reader, astackisusedasaprefixpooltoholdprefixes.firstly,anull string ε is pushed onto the stack. The reader starts a query with a null string ε; if this causes a collision, the reader pushes new queries onto the stack and pops prefix to a new query until the stack is empty. The A4PQT algorithm is described as follows. (1) Initialize algorithm. An empty string ε is pushed onto the stack. (2) Determine whether the stack is empty. If it is empty, then go to step (7). (3) The reader pops a new query prefix from the stack, and sends to tags. Tags compare the query prefix with their IDs and check whether their IDs contain the prefix. If the matching is successful, tags respond to the reader query and send their IDs which do not

187 4 International Journal of Distributed Sensor Networks include the prefix. If the matching is not successful, then tags respond to nothing. (4) Reader receives response from tags. If only one tag responds, turn to step (6).Ifnotagresponds,turnto step (7). If multiple tags respond simultaneously, then a collision occurs. The reader updates query prefix based on principle of pruning and pushes new prefixes onto the stack. Suppose S is the set of query prefixes; the response of tag is b 1 b 2,...,b u 1 b u b u+1,...,r, If the reader detects the highest collision bit which is b u, then new query prefixes produce the following principles (4.1) If the b u+1 bit is 0, then cut two branches of 4-ary tree; only use Sb 1 b 2,...,b u 1 00 and Sb 1 b 2,...,b u 1 10 as two new prefixes and push thetwoprefixesontothestack. (4.2) If the b u+1 bit is 1, then cut two branches of 4-ary tree; only use Sb 1 b 2,...,b u 1 01 and Sb 1 b 2,...,b u 1 11 as two new prefixes and push them onto the stack. (4.3) If the b u+1 bit is a collided bit, then generate four new branches. The reader uses Sb 1 b 2,...,b u 1 00, Sb 1 b 2,...,b u 1 01, Sb 1 b 2,...,b u 1 10, and Sb 1 b 2,...,b u 1 11 as new prefixes and pushes them onto the stack. (5) Repeat step (2) step (4). (6) Identify tag. (7) Algorithm ends. Using the five tags (they are 0100, 1001, 0010, 1011, and 0110) as an example from Figure 1, the identifying process of A4PQT is shown in Table 1, and the collision tree pruning process corresponding to the identifying process of the example is shown in Figure 2. When the reader sends an empty string ε, all tags are responding. The reader receives theresponseoftagsanddetectsacollisionaccordingtothe Manchester code; the result is XXXX, where X represents a collided bit. The reader uses 00, 01, 10, and 11 as new prefixes according to the principle of pruning. When the query prefix is 00, identify the tag When the query prefix is 11, there is no tag response, resulting in an idle time slot. When the query prefix is 01, the reader judges that the collided bit is X0 and can cut two branches 01 and 11. When the query prefix is 10, the collided result is X1; this can cut the two branches 00 and 10. ItcanbeseenfromFigure 2 that, to identify five tags by using pruning, A4PQT algorithm generates four additional time slots, wherein the collision time slots are three and the idle time slots are one, while as shown in Figure 1, thetotal additional time slots of binary collision tree are six, and 4-ary collision tree are eight. So A4PQT algorithm can reduce the collision time slots and the idle time slots; it can improve the identification efficiency of RFID system. 4. Performance Analysis In RFID tag identification, the total time slots of an anticollision algorithm to identify all tags are important performance indicators; the less the total time slots, the better the performance of the algorithm. A4PQT algorithm has two kinds of situations in implementation process; one is split in accordance with the 4-ary tree but to cut two branches. The other is split by a completely 4-ary tree. Nodes on 4-ary tree include the root nodes, intermediate nodes, and leaf nodes. The degree of intermediate node is 4 and the degree of leaf node is 0. So the total nodes of a 4-ary tree are N=n 0 +n 4. (2) Here n 0 is the number of nodes where the degree is 0 and n 4 is the number of nodes where the degree 4. Since node degree is 4 has 4 children, leaf node has 0 children, and the root node is not a child of the other nodes. Therefore, the total nodes of a4-arytreecanalsobeexpressedas Then there is N=0 n 0 +4n 4 +1=4n (3) n 0 +n 4 =4n (4) The number of leaf nodes can be determined as follows: n 0 =3n 4 +1, (5) n 4 = (n 0 1). (6) 3 Let n be the number of tags to be identified. When pruning condition is satisfied, A4PQT algorithm produces no idle time slots; the number of leaf nodes is equal to the number of tags; namely, n 0 =n. The number of intermediate nodes is only half the number of intermediate nodes of a full 4-ary tree, namely, n 4 = n 0 1. (7) 6 So when meet the pruning condition, the total time slots of A4PQT algorithm are N=n+n 4 = n+(n 1) 6 = (7n 1). (8) 6 When does not meet the pruning condition, each collision time slot in A4PQT algorithm will generate 0 2 idle time slot. If there has 0 idle time slot, the number of leaf nodes is equal to the number of tags; that is, n 0 =n. According to formula (5), we have n=3n (9) Then the total time slots of A4PQT algorithm are N=n 0 +n 4 =n+n 4 = (4n 1). (10) 3

188 International Journal of Distributed Sensor Networks 5 Table 1: The identifying process of A4PQT. Round R T ε T R Collided Readable Collided Collided Idle Readable Readable Readable Readable Tag Tag Tag Tag Tag Collision timeslots Readable timeslots Idle Figure 2: The process of A4PQT for identifying tags. If there are 2 idle time slots, it means that one collision node will split two idle nodes; an idle node is also a leaf node, and then the leaf nodes include readable nodes (n tags) and idle nodes. Assumingthenumberofidlenodesism,byformula(5), we have n+m=3n (11) And m=2n 4 ;obtainn 4 =n 1; then the total time slots of A4PQT algorithm are N=n+m+n 4 =n+3n 4 =n+3(n 1) =2n 3. (12) Therefore, the total time slots of A4PQT algorithm are (7n 1) T=[,2n 3]. (13) 6 Throughput is another important performance indicator of an anticollision algorithm, which is the ratio between the number of tags to be identified and the total time slots required to identify them. Throughput reflects the recognition efficiency of the algorithm; the greater the throughput, the higher the recognition efficiency. Let S be the identification throughput of A4PQT algorithm; we have S= n N = n [(7n 1) /6, 2n 3]. (14) Communication complexity is also a performance index for an anticollision algorithm; it is said to identify all tags required to transmit total number of bits. The communication complexity includes the reader communication complexity and tag communication complexity. Let C(n) be a communication complexity of A4PQT algorithm to identify n tags, let C R (n) represent the communication complexity of reader, let C T (n) represent the communication complexity of tag; then there is C (n) =C R (n) +C T (n). (15) In A4PQT algorithm, the sum of the reader query prefix length and tag response length is equal to the tag length. Let L pre.i be the query prefix bits length of the reader, let L req.i be the bits length of tag response, and let L ID be the bits length of tag ID; then formula (15) can be expressed as C (n) = T i=1 (L pre.i +L req.i ), (16) where T is the total time slots of the A4PQT algorithm. Since L ID =L pre.i +L req.i, the communication complexity of the A4PQT algorithm is C (n) =[ 5. Simulation and Result (7n 1),2n 3] L 6 ID. (17) To test the performance of A4PQT, we compare our scheme with query tree algorithm (QT), collision tree algorithm (CT) [4], and the improved 4-ary query tree algorithm (I4QTA) [6]. CT eliminates the unnecessary idle time slots based on the basic query tree algorithm. I4QTA adopts a transposition way to reduce idle time slots. When two continuous collision bits occur, tags modify some bits to respond; the reader judges a tag prefix according to the transposed number, but this method also adds an extra cycle. Randomly generating 96 bits tag, the number of tags increases from 0 to 1000, and all results are the average of 100 experiments. In total time slots, Figure 3 shows the total time slots of QT, CT, I4QTA, and A4PQT for identification of all tags. It can be seen that the total time slots of CT are significantly less than that of QT; this is because the CT has no idle time slots. The total time slots of the I4QTA are slightly less than that of

189 6 International Journal of Distributed Sensor Networks Total time slots QT CT Number of tags 14QAT A4PQT Figure 3: Total time slots of the four algorithms to identify tags. Throughout Number of tags QT CT 14QAT A4PQT Figure 4: Throughput of the four algorithms to identify tags. CT. I4QTA split by 4-ary tree can reduce collision time slots and adjust its query prefix through the response of tags, which can reduce the idle time slots the 4-ary tree generates. A4PQT algorithmusesanadaptivewaytoprune4-arytree,soitcan more effectively eliminate some idle time slots, and the total time slots are less than those of the other three algorithms. Figure 4 shows the throughput of the four algorithms. With the increasing of the number of tags, the throughput ofqtisabout34%andthatofcttendstobe50%;the throughput of I4QTA is about 52%, while that of A4PQT tendstobe64%. In communication complexity, Figure 5 shows the total transmitted bits for identifying n tags using the QT, CT, I4QTA, and A4PQT algorithms. As can be seen, the total Communication complexity QT CT Number of tags 14QAT A4PQT Figure 5: The communication complexity of the four algorithms to identify tags. transmitted bits of A4PQT algorithm are always less than that of the other three algorithms. The experimental results show that A4PQT algorithm can significantly reduce the total time slots and the transmitted bits for identifying tags; thus A4PQT algorithm has very fast identification rate and very high throughput; this is because A4PQT algorithm can eliminate some idle time slots. 6. Discussion QT, CT, I4QAT, and A4PQT all belong to query tree algorithm. QT algorithm is a binary query tree algorithm. The reader sends a query prefix and tags will respond when their IDs have a prefix matching. The collision will occur when multiple tags respond simultaneously; then the reader will update the prefix by adding 0 and 1 in the original prefix. SincetheprefixinQTisbitwiseincrementallyupdating, this will cause no tag response when some queries are sent. This will bring idle time slot and increase the query time. CT algorithm is also a binary query tree algorithm. It is an improved version of the QT. The prefix in CT algorithm is updated by jumping; thus it can avoid the idle time slots. CT algorithm performance is also greatly improved and can avoid all idle time slots that the QT has. But it does not avoid the collision time slots. I4QAT is a 4-ary query tree algorithm which can reduce the collision time slots but meanwhile increases the idle time slots. I4QAT uses Manchester code to detect that the collided bit is one bit or two consecutive bits. If there are two consecutive collided bits, the reader adds an additional query to send to tags. When tags receive the query, they change the corresponding two bits. The conversion method is the 00 converted into 1, 01 converted into 01, 10 converted into 001, and 11 converted into I4QAT algorithm is to reduce idle

190 International Journal of Distributed Sensor Networks 7 time slots by using an additional query and it can eliminate theidletimeslots.sotheperformanceofi4qatalgorithm is greatly improved. However, I4QAT algorithms rely on additional queries to eliminate idle time slots; thus it also affects the performance of the I4QAT. It is also an additional burden to tags to change two bits. A4PQT algorithm is different from I4QAT algorithm. A4PQT algorithm does not execute additional queries; instead, it assumes there are fivetagswhichusepruningwaytoeliminatesomeofthe idle time slots. When a collision bit appears, all the invalid branches will be pruned to eliminate all the idle time slots. Even when there are two consecutive collision bits, A4PQT algorithm can still eliminate most of the idle time slots without having to execute additional queries. As shown in our previous performance analysis, the average time slots requested by A4PQT algorithm have been lower than those required by I4QAT algorithm. When the number of tags reaches 500, the average number of time slots requested by A4PQT algorithm will be about 700. Relatively, the average number of time slots requested by I4PQT algorithm will come up to 900. As to some other aspects in terms of performance, A4PQT algorithm also has an obvious advantage over I4QAT algorithm as well as other algorithms. 7. Conclusion In the multitag identification environment, the collision is the major issue affecting the performance of RFID system. Collision resolution strategy based on binary tree can reduce idle time slots but increase collision time slots, while the 4-ary tree resolution strategy is to reduce the collision time slots but increase the idle time slots. In this paper, we propose a novel 4-ary query tree algorithm called adaptive 4-ary pruning query tree (A4PQT) for identifying multiple tags based on the characteristics of collided bit, using pruning way to eliminate some idle time slots of 4-ary tree. The A4PQT algorithm makes use of the advantages of 4-ary tree scheme that has less collision time slots and overcomes shortcomings of its more idle time slots. Both theoretical analysis and simulation results show that A4PQT algorithm can significantly reduce the total time slots for identifying multitag and improve the throughput of identification tags. As evidenced in the performance analysis, the adoption of n-ary tree will increase the idle time slots while the collision time slots are decreased. We envision that the pruning strategy will be an important and effective way to improve the efficiency of identification in the Radio frequency identification (RFID) system. Our further research work will be focused on further study of proper pruning strategy which will eliminate the increased idle time slots and decrease the total time slots. system, in Proceedings of the 24th International Conference on Microelectronics,pp.1 4,2012. [3] H.Wu,Y.Zeng,J.Feng,andY.Gu, BinarytreeslottedALOHA for passive RFID tag anticollision, IEEE Transactions on Parallel and Distributed Systems,vol.24,no.1,pp.19 31,2013. [4] X. Jia, Q. Feng, and L. Yu, Stability analysis of an efficient anticollision protocol for RFID tag identification, IEEE Transactions on Communications,vol.60,no.8,pp ,2012. [5] D.R.HushandC.Wood, AnalysisoftreealgorithmsforRFID arbitration, in Proceedings of the IEEE International Symposium on Information Theory, pp , [6] Y. Kim, S. Kim, S. Lee, and K. Ahn, Improved 4-ary query tree algorithm for anti-collision in RFID system, in Proceedings of the International Conference on Advanced Information Networking and Applications (AINA 09),pp ,May2009. [7] P. Pupunwiwat and B. Stantic, Unified Q-ary tree for RFID tag anti-collision resolution, in Proceedings of the 20th Australasian Database Conference (ADC 09), Conferences in Research and Practice in Information Technology, pp , [8]H.GouandY.Yoo, AbitcollisiondetectionbasedHybrid Query Tree protocol for anti-collision in RFID system, in Proceedingsofthe11thIEEEInternationalConferenceonComputer and Information Technology (CIT 11), pp ,September [9]Y.JiangandR.N.Zhang, Anadaptivecombinationquery tree protocol for tag identification in RFID systems, IEEE Communications Letters,vol.16,no.8,pp ,2012. [10] J. Sung, D. Kim, T. Kim, and J. Choi, Heuristic query tree protocol: Use of known tags for RFID tag anti-collision, IEICE Transactions on Communications, vol.e95-b,no.2,pp , [11] M.-K. Yeh, J.-R. Jiang, and S.-T. Huang, Parallel response query tree splitting for RFID tag anti-collision, in Proceedings of the International Conference on Parallel Processing Workshops (ICPPW 11), pp. 6 15, September [12] Y.-H. Chen, S.-J. Horng, R.-S. Run et al., A novel anti-collision algorithm in RFID systems for identifying passive tags, IEEE Transactions on Industrial Informatics,vol.6,no.1,pp , References [1] J. Shin, B. Jeon, and D. Yang, Multiple RFID tags identification with m-ary query tree scheme, IEEE Communications Letters, vol. 17, no. 3, pp , [2]A.Rennane,H.Saadi,R.Touhami,andM.C.E.Yagoub, A comparative performance evaluation study of the basic binary treeandalohabasedanti-collisionprotocolsforpassiverfid

191 Hindawi Publishing Corporation International Journal of Distributed Sensor Networks Volume 2013, Article ID , 13 pages Research Article Improving Performance through REST Open API Grouping for Wireless Sensor Network Min Choi, 1 Young-Sik Jeong, 2 and Jong Hyuk Park 3 1 Department of Information and Communication Engineering, Chungbuk National University, 52 Naesudong-ro, Heungdeok-gu, Cheongju, Chungbuk , Republic of Korea 2 Department of Multimedia Engineering, Dongguk University, 30 Pildong-ro 1-gil, Jung-gu, Seoul , Republic of Korea 3 Department of Computer Engineering, Seoul National University of Science and Technology, 232 Gongneung-ro, Nowon-gu, Seoul , Republic of Korea Correspondence should be addressed to Jong Hyuk Park; parkjonghyuk1@hotmail.com Received 18 June 2013; Accepted 30 August 2013 Academic Editor: Naveen Chilamkurti Copyright 2013 Min Choi et al. This is an open access article distributed under the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited. With this growth of the Internet, it is expected that every device, including computers, will be connected to the Internet, as it is called IoT. For example, smartphones and even refrigerators require an address to connect to the Internet. In this research, we design Internet of things architecture, especially for wireless sensor networks. The architecture consists of wireless sensor networks with a microcontroller at the very bottom level. They are connected to smart devices at the next level. However, the computing capability of the smart devices is generally less powerful than that of the conventional devices. Thus, it is necessary to offload the computation-intensive part by careful partitioning of application functions. In this research, we focus on designing the concept of MapReduce like approach through the web service grouping of several web services into one. We propose two methods: REST API grouping and REST API caching. First, the web service composition results in reducing energy consumption and communication latency by composing two or more REST web services into one. Second, the web service caching technique provides fast access that is recently accessed or frequently accessed. We conducted the experiments with Jersey REST web service server. Experimental result shows that our approach outperforms conventional approaches. 1. Introduction With the fast development of the Internet technologies, web based architectures are becoming the major technologies for various fields of mobile computing. Nowadays, we are experiencing a major shift from traditional mobile applications to mobile cloud computing. The demand of Open API based development stems from the increasing use of smartphone applications [1, 2]. Community portal companies such as Google, Naver, and Yahoo are providing the Open API service for the access of their service. Before we go into more details, we briefly introduce the REST Open API based mobile application development approaches. Within a few years, we can expect a major shift from traditional mobile application technology to mobile cloud computing [3]. It improves application performance and efficiency by offloading complex and time-consuming tasks onto powerful computing platforms. By running only simple tasks on mobile devices, we can achieve a longer battery lifetime and a greater processing efficiency. Not only is this offloading with the use of parallelism faster but it can also be used to solve problems related to large datasets of nonlocal resources. With a set of computers connected on a network, there is a vast pool of CPUs and resources, and you have the ability to access files on a cloud. In this paper, we propose a novel approach that realizes the mobile cloud convergence in a transparent and platform-independent way. Users need not know how their jobs are actually executed in distributed environment, and users need not take into account whether theirmobileplatformsareiphoneorandroid.

192 2 International Journal of Distributed Sensor Networks Tocommunicatewithremoteprocedurecallbetween client and server, interface should be defined at first. To this end, WSDL and RPC were used for the specification. However, these previous approaches are relatively complicated and highly overloaded. Recently, REST architecture is first introduced by Fielding. REST web service is becoming popular and explosively used in the field of application development of web and smartphone. Therefore, today s many Internet companies already provide their services by both traditional SOAP based web service and RESTful web services [4, 5]. The main difference between REST web service and SOAP/WSDL web service is as follows. Due to the complicated characteristics of SOAP based web services, REST web service is introduced. REST web service removes the overhead from encoding/decoding of header and body during message transfer. The REST web service enables users anddeveloperstoeasilyusethewebservicesatremote or local sites. We need not add additional communication layer or protocols for REST web service, but we can easily achieve scalability and performance. This research evaluates the performance of mash-up architectures through RESTful Open API web services on smart mobile devices. It provides the analytical and experimental results for the performance evaluation of system models. We especially try to find an optimalnumberofparallelrestwebserverarchitectures under certain request arrival rates. And we show the performanceofproposedarchitecture,especiallythemeannumber of requests in the queue and the mean waiting time. The area of REST web service composition is underexplored and most research efforts are still at their initial state [3, 6 8]. In this paper, we propose a new conversion method from web service execution result to object. REST web service execution results are usually provided in XML format. Previous composition method is required to analyze web service execution result with XML parser [9]. Other previous composition approaches were exploited to synthesize program code from linear logic or first-order logic [1, 2]. These papers are well organized theoretically and logically, but they have limited capability in terms of automatic synthesis. However, in order to provide an easy way for web service composition, we convert REST web service to objects. The conversion changes web service to a directly readable format (objects) with OOP language. The objects are programming primitives generally available for all types of OOP languages, such as C++ and Java. Since OOP languages are very popular for developers, they can easily utilize OOPs to compose web services. REST web service is core technology for smartphone application development. This is because REST web service is the most appropriate way for accessing information through the Internet. Usually, a smartphone application needs information from several sources of (one or more) REST web services [1]. So, we need to utilize two or more REST web services composition to realize a target application [3, 6]. In this paper, we propose a server architecture for managing REST web services. This server is for managing web services so as to provide web server maintenance, especially on composition, deployment, and management of REST web services. It enables service developers to conveniently develop, deploy, upload, and run their composed web services with the use of general OOP languages. The rest of this paper is organized as follows. Section 3 describes the necessity of WSN using REST web service grouping. Section 2 shows conversion of REST web service to objects. Section 4 depicts management server architecture for REST web service system. Then, Section 5 shows performance evaluation. Finally, we conclude and summarize our work in Section 6. Wireless sensor network brings computing power to places and things that were previously not able to imagine to realize because they were cost prohibitive or physically impossible [10, 11]. This emerging wireless technology allows computingtogotoplacesneverbeforepossible,everywhere of our physical world. Via the Internet, a variety of computing devices, including wireless sensor network devices, are connected into a worldwide computing network and becoming the next generation communication devices. In this research, we make use of the Chipcon CC2420 RF Transceiver which is capable of 2.4 GHz communication by IEEE standard [12]. It reduces the load on the host controller and allows CC2420 to interface low-cost microcontrollers. Figure 1 shows a block diagram of RF sensor module and its connection to ATmega128 microcontroller. The left side of Figure 2 shows a CC2420 RF module. The right side of Figure 2 represents an example of real connection between ATmega128 microcontroller and the CC2420 RF module. During transmission, the FIFO and FIFOP pins are only related to the RXFIFO. The SFD pin is active during transmission of a data frame. The SFD pin goes high when the SFD field has been completely transmitted. In receive mode, the FIFOP pin can be used to interrupt the microcontroller when a threshold has been exceeded or a complete frame has been received. This pin should be connected to an ATmega128 s interrupt input port. The ATmega128 microcontroller communicates with smart devices to provide data which is applicable to arguments of REST Open APIs, for example, temperature, brightness, humidity, and any data from ADC. In 2004, the concept of MapReduce [13] was introduced as a novel programming model and implementation for a large set of computing devices. Map generates a set of intermediate key/value pairs and Reduce merges all intermediate values associated with thesameintermediatekey,sothatprogramswiththisare automatically parallelized and executed on a large cluster of computing devices. This research focuses on designing the concept of MapReduce through the web service composition of several web services into one. This REST web service composition results in reducing energy consumption and communication latency. This is because the conventional approach generates several consecutive connection requests to remote servers through REST Open API. In this case, two or more REST web services execution should be carried out on smartphones. However, our REST web service composition eliminates such several consecutive connection requests since several REST web services are composed into one. Moreover, the REST web service caching technique in this research provides

193 International Journal of Distributed Sensor Networks 3 Automatic gain control ADC Digital demodulator LNA ADC -Digital RSSI -Gain control -Image suppression -Channel filtering -Demodulation -Frame -Synchronization Serial voltage regulator TXRX control PA On-chip BIAS Power control X08C Σ 0 90 Freq synth TX power control DAC DAC Control logic Digital interface with FIFO buffers, CRC and encryption Digital modulator -Data spreading -Modulation Digital and analog test interface Serial microcontrollers interface 16 MHz Figure 1: Chipcon CC2420 RF module block diagram [12]. (a) (b) Figure 2: CC2420 RF module and ATmega128 board. optimization through data caching that is recently accessed or frequently accessed. REST web service is core technology for smartphone application development. This is because REST web service is the most appropriate way for information access through the Internet. Usually, a smartphone application needs information from several (one or more) REST web services. So, we need to utilize two or more REST web services for realizing target application. Algorithm 1 shows the Open API REST web service for the development of search applications on smartphones. Algorithm 2 shows the example of open API REST web service for keyword search of web documents. Assuming developing a web search application, two-phase search task is necessary; the first step is to check the validity of search keyword. This step is to prevent persons who are under 19 years of age to access adult data through the search engine. The second step is to search the keyword actually from web database. This step is to get the content from search engine after checking keyword validity of Algorithm 1. Algorithm 3 describes the list of error messages when there is failure in Open API request, for example, invalid input parameters, network failure, and authentication failure. Like above, the necessity for several REST web service composition is obvious. 2. Conversion of REST Web Service to Objects Web service composition requires a method to access data that are in XML format of web service execution result

194 4 International Journal of Distributed Sensor Networks Table 1: Real examples of the web service to object conversion. Class variables Num Latitude Longitude Name Street City County State Country Object instance Coex Mall Samsung-dong Kangnam-gu Seoul Republic of Korea Object instance Coex Square Samsung-dong Kangnam-gu Seoul Republic of Korea Web application (web service) Web application (web service) (a) (b) (c) XML Objects Composition XML (a) (b) (c) Objects Web application (web service) Figure 3: Conversion of web service execution result to objects. from OOP languages. Because OOP languages are familiar to developers, they can easily utilize the OOPs to compose web services. In order to realize REST web service composition, we propose a web service composition method by conversion of web service to objects. The reason why we convert the REST web service to objects is to make an easy way for manipulatingthewebserviceresultintocomposition.the objects mean that they are the programming primitives which are generally available for all types of OOP languages, such as C++ and Java. Figure 3 depicts the extraction process from web service to objects. Step (a) represents that web service execution results are returned as XML format. Step (b) describes the process that XML is converted to object. The process is to derive objects that are available for object-oriented languages. Actually, our web service manager converts the XML result to create Java program objects that represent the result data of web service execution. step (c) is to compose several objects to make another new web service that utilizes one or more web services. The reason why we convert web service to object is that the object-oriented language is the most convenient tool for developers to manipulate and understand easily. This is because previous REST web service composition proposals of H. Zhao [4] and X. Zhao [5] are well organized theoretically and logically, but these are difficult for developers and users to easily understand and manipulate with a familiar programming language. Algorithm 4 represents the result of REST web service execution. REST web service execution result is provided as XML data like above. After that, we proceed to the extraction process with the results of REST web service execution. In this example, we use web service from Yahoo. The REST web service is called Open API. It returns execution result by a type of XML data. When clients receive the XML, they first parse the XML and finally they get the data that they wanted. 3. The Necessity of WSN Using REST Web Service Grouping for MapReduce As shown in Algorithm 4, theabovexmldocumentcontains a single element ResultSet, which has subelements for head, locations, and item. The locations element contains a collection of item elements. The item element has several attributes: num, latitude, longitude, name, street, city, county, state, and country. In Algorithm 5, we see the class definition for conversed object mapping. For convenient use of objects in OOP languages, the result of data conversion should be provided by a real object instance which can be directly referenced on OOP program source code. To do this, we need askeletonclass,shownasasampleinalgorithm 5,inwhich simple types are mapped to each property variable and service developers can access the property values using get and set methods as shown in Algorithm 5. The above example in Algorithm 5of REST web service execution can be converted bythefollowingsetofobjectinstanceswhichhaveattributes of the following result in Table 1. For the example of Algorithm 4, thelocationselement contains a collection of <item> tags which has subelements. The tag <item> is the object identification element in our conversion system, and it has several attributes: num, latitude, longitude,name,street,city,county,state,andcountry.it is easy for a human to make a decision that <item> tag is repeated per object. However, it is not easy for a machine to decide which tag is corresponding to the object separator. This conversion process is repeated until reaching the end tag. That is why our system requires the separation of tag name forobjectidentificationatthebeginningoftheconversion process. 4. Management Server Architecture for REST Web Service In this section, we propose a server architecture for managing REST web services. This server is for managing web services instead of web so as to provide web server maintenance service, especially composition, deployment, and management. Figure 4 shows the architecture of our REST web service management system. The main role of the system is composition/deployment/management of REST web services. It enables service developers to conveniently develop, deploy,

195 International Journal of Distributed Sensor Networks 5 (c) A web service Object 1 Object 2 Service developer (e) Composition (a) Mobile web (smartphones) (a ) (b) Web server Composed web service (d) REST web service manager Database Web browser Figure 4: REST web service management system. (1) Request URL (2) Request parameter key string (mandatory): key string for authentication target string (mandatory): adult querystring(mandatory):searchkeywordasutf-8encoding -SampleURL (3) Response field adult integer: 0, 1 (0 non adult, 1 adult) Algorithm 1: Open API of REST web service for checking search keyword validity. upload, and run the composed web service by general OOP languages. Web browsers and mobile web browsers shown in (a) of Figure 4 are very popular on desktop and smartphones, respectively. They commonly utilize the HTTP to communicate with web server through port number 80. The web server in (b) of Figure 4 is an application daemon which receives request from web browser and provides the requested documents and data. Module (c) in Figure 4 represents web service or composed web service. It can be provided by platformindependent packaging technology, such as COM/COM+ and JavaBeans. This package can include directory structure that has a restriction on which directory should have a configuration file for our web service management system. Module (d) in Figure 4 is REST web service manager. It manages REST web services which can be either a native REST web service or a composed REST web service. It provides service to requests from clients. The service developers (e) in Figure 4 can upload their web service or composed web service onto our REST web service management system, so that web services can be launched and serviced on demand. This is quite useful for smartphone application developers. This is because the computing power of smartphones is generally less than that of other mobile computing devices, such as laptop computers and mobile tablets. Therefore, it is necessary to offload the computation-intensive part by the careful partitioning of application functions across the cloud computing platform. To this end, we make use of RESTful web service to realize distributed computing environment. During loading and running composed REST web service in (d) of Figure 4, dynamic binding is required for composed web service to use objects which are converted from web service. Likewise, the management system has to instantiate and dynamically bind the composed objects. This is because our system provides service concurrent users at the same time. At the time, it is not possible for all web services to be loaded onto memory. Some of them might be garbage collected during the runtime. Thus, we need to reload and bind the object when it is about to be referenced. So, our web service management system supports dynamic loading and binding for converted objects from web service. Likewise, we propose a REST web service management system that provides REST web service for clients such as