Research Article A Novel Data Classification and Scheduling Scheme in the Virtualization of Wireless Sensor Networks

Transcription

1 International Journal of Distributed Sensor Networks Volue 2013, Article ID , 14 pages Research Article A Novel Data Classification and Scheduling Schee in the Virtualization of Wireless Sensor Networks Md. Motaharul Isla and Eui-Na Huh Departent of Coputer Engineering, Kyung Hee University, Yongin-si, Gyeonggi-do , Republic of Korea Correspondence should be addressed to Eui-Na Huh; johnhuh@khu.ac.kr Received 25 February 2013; Revised 27 June 2013; Accepted 1 July 2013 Acadeic Editor: Al-Sakib Khan Pathan Copyright 2013 Md. M. Isla and E.-N. Huh. This is an open access article distributed under the Creative Coons Attribution License, which perits unrestricted use, distribution, and reproduction in any ediu, provided the original work is properly cited. Most of the nodes in a wireless sensor network (WSN) reain idle for the axiu period of their lifetie resulting in underutilization of their resources. There are any ongoing research studies to utilize the resources of sensor nodes in an efficient way. Virtualization of sensor network (VSN) is one of the novel approaches to utilize the physical infrastructure of a WSN. VSN can be siply defined as the virtual version of a WSN over the physical sensor infrastructure. By allowing sensor nodes to coexist on a shared physical substrate, VSN ay provide flexibility, cost effectiveness, and anageability. This paper proposes a QoS-aware data classification and scheduling fraework for VSN in the health care sector. We develop a tiny virtual achine called VSNware for health care applications, which facilitates QoS-aware forwarding of data packets, aintaining the reliability, delay guarantee, andspeed.thesiulationresultsalsoshowthattheproposedscheeoutperforstheconventionalwsnapproaches. 1. Introduction Recent advances in electronics have enabled the developent of ultifunctional sart sensor nodes that are sall in size and counicate in an untethered anner over short distances. A sensor network consists of a large nuber of tiny sensor nodes that are densely deployed over a specific target area [1 4]. There is a robust deployent of WSNs in the health care sector because of their sall size. Today s sart sensor node can efficiently onitor different vital signs such as the cardiac data, teperature, blood pressure, pulse rate, and saturation of peripheral oxygen (SPO 2 )ofapatient.in the health care scenario, applications deand different types of QoS requireents such as reliability, end-to-end delay, speed, and tieliness. There are any ongoing efforts to enhance the QoS issues of WSNs in the existing literature [5 7]. In this age of recession, providing QoS affordably in the WSN-based health care syste is a big challenge for the increasing worldwide elderly population, which is the largest deographic group in the developed countries. For this very reason, researchers are searching for cost-effective ways to support QoS in WSNs for the health care sector. Very recently, network virtualization has created a resonance aong the network research counity. The concept of sensor virtualization has also attracted a great deal of attention fro industry and acadeia [8 10]. Virtualization of sensor network (VSN) can be defined as the separation of functions for the traditional wireless sensor network (WSN) service provider into two parts: the sensor infrastructure provider (SInP), which anages the physical sensor infrastructure, and the VSN service provider (VSNSP), which develops VSN by aggregating the resources fro ultiple SInPs and offers services to the application-level users (ALUs). The WSN virtualization renaissance has originated ainly fro the realization that ost of the sensor nodes reain idle for ost of the tie in a WSN. Virtualization is one of the best ways to utilize the physical sensor infrastructure. VSN can provide a platfor upon which novel sensor network architectures can be built, experientally tested, and evaluated. In addition, virtualization in WSNs is expected to provide a clean separation of services and infrastructure and to facilitate new ways of doing business with sensor network resources aong ultiple service providers and applicationlevel users [8]. In this paper, we propose QoS-aware data classification and a scheduling fraework for the health care syste in VSN. The sensor node senses parallel data and forwards it to a

2 2 International Journal of Distributed Sensor Networks nearby node or gateway node. The gateway node classifies the data as urgent, suspicious, oderate, or noral. The classified data are passed through the decoding odule and are queued up in the VSN queue. Finally, the scheduling odule sends data to a specific path based on the priority, reliability, and delay requireents of the data packets. The ain contributions of this paper are as follows. (a) A business odel of the virtualization of sensor network is proposed. (b) A tiny virtual achine called VSNware for health care applications has been developed for QoS-aware data packet forwarding. (c) Packet classification and scheduling echaniss are suggested. (d) A detailed probabilistic analytical odel of the reliability and delay for different traffics is proposed. (e) Finally, the siulation results of the proposed schee are presented with respect to other approaches. The reainder of the paper is organized as follows. Section 2 reviews the background related to the conventional WSN, virtual sensor network, VSN, and related works. In Section 3, wediscussthedetailedarchitectureofthesensor nodes and the sensor gateway router. Section 4 describes the VSN network odel. Section 5 states the classification and scheduling of data packets. Sections 6 and 7 discuss a detailed atheatical odel of the delay and reliability for different data traffics. Section 8 presents the perforance evaluations and siulation results. Finally Section 9 concludes the paper. 2. Backgrounds VSN is a brand-new research approach in the field of WSN. Before proceeding further, we need to clarify few basic concepts and the difference between traditional WSN, virtual sensor network, and VSN. In this paper, VSN eans virtualization of a WSN as defined in the Introduction and in Section 2.3. The ter VSN in this paper is synonyously used for the process of virtualization of a sensor network and for the network that supports virtualization Traditional WSN Approach. Traditional wireless sensor network consists of a large nuber of sensor nodes that are densely deployed either inside the phenoenon of interest or very close to it [1]. A sensor node senses its surrounding environent, perfors necessary coputation and processing, and sends the sensory data through ultihop or directly tothecoordinatornode.thecoordinatornodeaybea fixed node or a obile node capable of connecting the sensor network to an existing counication infrastructure where a user can access the reported data. By integrating sensing, signal processing, and counication functions, a traditional sensor network provides a natural platfor for hierarchical inforation processing [2 4]. The traditional WSN is dedicated for the onitoring of a particular event. But in VSN environent, the sae infrastructure can be used by ultiple stack holders Virtual Sensor Network. The virtual sensor network consists of a collaborative wireless sensor network. It is fored by a subset of sensor nodes of a wireless sensor network, with the subset being dedicated to a certain task or an application at a given tie [11, 12]. In contrast, the subset of nodes belonging to the virtual sensor network collaborates to carry out a given application at a specific tie. A virtual sensor network can be fored by providing logical connectivity aong collaborative sensor nodes. Nodes can be groupedintodifferentvirtualsensornetworksbasedonthe phenoenon they track or the task they perfor. The virtual sensor network protocol should provide the functionality for network foration, usage, adaptation, and aintenance of a subset of sensors collaborating on a specific task [13]. 2.3.VSNandItsBusinessModel.Unlike the conventional WSN, the VSN environent has a collection of ultiple heterogeneous sensor nodes that coexist in the sae physical space. VSN is a type of network that creates a virtual topology on top of the physical topology of a traditional WSN. SInP in Figure 1 deploys different sensor nodes. In the traditional WSNinfrastructure,theproviderandtheserviceproviderare thesaeentity.vsndifferentiatesbetweentheinfrastructure provider and the service provider, thus providing a business odel of true virtualization. SInP deploys sensor network resources. It offers resources through prograable interfaces to different VSNSPs. Different interest groups can deploy sensor nodes and can ake individual infrastructures, which can be used by the VSNSP to run individual applications. The VSNSP gets resources fro ultiple SInPs to deploy VSNs by sharing the allocated virtualized network resources to offer end-to-end application user services. The VSNSP can obtain resources fro ultiple SInPs. ALUs in the VSN odel are siilar to those of the existing WSN, except that the existence of ultiple VSNSPs fro copeting SInPs provides a wide range of choices. Any end user can connect to ultiple VSNSPs fro different SInPs to use ultiple applications Related Works. Currently there are few approaches in the WSN [5, 11 16] thatfocusonthevirtualandoverlay sensor network rather than the purist view of the VSN approach introduced in this paper. Table 1 suarizes a few of the research projects that act as the background of the proposed research approach. It deonstrates the conteporary research direction in the field of virtualization of sensor networks in general. Recently, the Federated Secure Sensor Network Laboratory (FRESnel) has aied to build a large-scale sensor fraework. The goal of this project is to offer an environent that can support ultiple applications running on each sensor node [14 16]. It provides an execution environent that hides the syste details fro the running applications. The syste operates in a shared environent. The key characteristics of this approach are a virtualization layer that is running on each sensor node and provides abstracts access to sensor resources, which allows the anageent of these resources through policies expressed by the infrastructure

3 International Journal of Distributed Sensor Networks 3 Care giver Govt. office Doctors Patients ALU VSNSP-2 VSNSP-1 SInP SGR SGR SGR SGR SGR Figure 1: Business odel for VSN. Table 1: VSN research-related projects. Projects Research area URL FRESnel To build a large-scale federated sensor network fraework with ultiple applications sharing the sae resources VSNs Rando routing, virtual coordinates, and VSN support functions Sensor Planet Nokia-initiated cooperation, a global research fraework, on obile device-centric large-scale wireless sensor networks ViSE Virtualization of sensor/actuator syste, creating custoized virtual sensor network test beds DVM To build a syste that supports software reconfiguration in ebedded sensor networks at ultiple levels SensEye Multitier ultiodal sensor networks SenQ Coplex virtual sensors and user-created streas can be dynaically discovered and shared WebDust Multiple, heterogeneous, wireless sensor networks can be controlled as a single, unified, virtual sensor network

4 4 International Journal of Distributed Sensor Networks owner. A runtie environent on each node allows ultiple applications to run inside the sensor node. It also provides policy-based application deployent that enables ultiple applications to be deployed over the shared infrastructure. In MMSPEED [5], a novel packet delivery echanis for QoS provisioning was proposed. It provides QoS differentiation in ters of two qualities, such as tieliness and reliability. This approach is based on ultiple logical speed layers over a physical sensor network that is based on the conventional virtual sensor network. Based on the speed, it considers different virtual overlays. For virtual layering, it eploys virtual isolation aong the speed layers. This is accoplished by classifying the incoing packets according to their speed classes and then placing the into the appropriate priority queues. SenShare [15] is another platfor that attepts to address the technical challenge of supporting ultiple corunning applications in the sensor node. Here each application operates in an isolated environent consisting of an innode hardware abstraction layer and a dedicated overlay sensor network. Instead of using a virtual achine, SenShare uses a hardware abstraction layer. It is a set of routines in software that eulates soe platfor-specific details, thereby giving progras direct access to the hardware. The Mate [17] and Melete[18] systes are based on the virtual achine approach that provides reliable storage and enables the execution of concurrent applications on a single sensor node. The VSN approach proposed in this paper is based on the Mate and Melete systes. This odified version of virtual achine is called VSNware. VSNware provides an environent to support different applications for health care systes such as cardiac data, blood pressure, blood sugar, and teperature sensing. VSNware helps to provide the purist view of the virtualization concept. It does so by separating thesinpandvsnspasdiscussedintheprevioussections.by applying the purist view of virtualization in VSN, this schee can be efficiently used in the health care syste, which is the ain contribution of this paper. To the best of our knowledge, nopreviousresearcharticlehasexploredthevsnapproach to design a ubiquitous health care syste for the QoS-based vital data classifications and scheduling schee. 3. Architecture of VSN In this section we are going to introduce the detail architectural design of the proposed VSN approach and the description of its individual coponents elaborately Syste Architecture. Here we briefly describe the detailed syste architecture and the software architecture of the sensor node and sensor gateway router (SGR). In the following sections we will explain the architecture in detail. The syste architecture consists of three layers: the SInP, VSNSP, and ALU. The software architecture describes the virtualization of the individual sensor nodes and gateway router SInP. SInP consists of different sensor nodes. These sensor nodes sense different vital signs of the patient, such as the teperature, heart rate, blood pressure, and blood Applications Figure 2: VSN architecture. sugar. To sense the patient body in the VSN environent, we consider two types of sensor node: the fully functional device (FFD) and the reduced functional device (RFD) sensor nodes. SInP deploys sensor nodes in the hospital in a distributed anner. Each group of sensor nodes is divided into different logical areas, which are identified by circles to indicate the SGR doain. Each SGR doain ay consist of one or ore SGRs, which is an FFD sensor node. Each SGR supports sensor virtualization. In each doain there are any RFD sensor nodes that sense vital signs. The RFD is ore resource-constrained than the SGR/FFD VSNSP. The VSNSP consists of any virtual SGRs (s), which are the virtual representations of the processing, storage, and other resources of the SGRs. The links between the s are the dynaically allocated channels between the SGRs. Since the VSN schee is based on the IEEE radio specification, it has 16 channels. Each channel is considered to be an individual path that consists of ultiple links between SGRs. Each VSN provides a specific application service to the users. In Figure 2, we depict up to a axiu of 16 VSNs provided by a specific VSNSP as the underlying SInPs can support ALU. This layer consists of different application level users such as doctors, nurses, patients, or any other specialized users. Based on the application requireents, the ALU sends a request to the VSNSP. The VSNSP then aps the particular VSN according to the request of the specific application. Individual applications ay use ultiple VSNSP VSN 16 VSN 2 VSN 1 SInP

5 International Journal of Distributed Sensor Networks 5 Teperature Input/output Cardiac Sensor node Blood pressure Blood sugar VSNware Application anageent Ebedded linux CPU USB RF Storage Figure 3: Architecture of sensor node. fully functional sensor node that supports the concurrent processing of ultiple applications. It also consists of a physical layer, sensor network operating syste layer, VSNware layer, and application layer. The lower layer consists of the physical sensor resources, such as the central processing unit (CPU), RF odule, and storage odule. The sensor operating syste layer consists of a typical ultitasking sensor network operating syste. In this odel we use Ebedded Linux as it is used in the individual sensor nodes. ItprovidestheenvironenttoruntheVSNware.VSNware supports concurrent execution of the applications. VSNware consists of different odules such as forwarding/routing in VSN, input/output, and application anageent. Finally, the application layer provides the classified data, such as urgent, suspicious, oderate, and noral, over different VSNs based on the reliability and delay requireents. Urgent data Suspicious data VSN gateway Moderate data Noral data VSNware Virtual routing Input/output Application anageent Ebedded linux CPU channel Storage Figure 4: Architecture of SGR. resources. The user ay be a achine in the case of achineto-achine counication, which can involve individual coputers and any other sart device. 3.2.SoftwareArchitectureforSensorNodeandSGR. Figure 3 depicts the software architecture of a sensor node. It senses vital signs fro patients in a health care syste. A single sensor node perfors ultiple sensing tasks by using physical infrastructure virtualization as a service. The typical sensor node architecture consists of a physical layer, an operating syste (OS) layer, a virtualization layer, and ultiple sensing service layers. The lower layer consists of the physical sensor resources such as a central processing unit (CPU), USB odule, RF odule, and storage odule. Layer 2 consists of a typical ultitasking sensor network operating syste. We use Ebedded Linux in this case. Ebedded Linux provides the environent to host the virtualization layer. The virtualization layer supports concurrent service execution. The virtualization layer includes the input/output and application anageent odules. Finally, the application layer runs ultiple services over the virtualization layer, such as the sensing of teperature, cardiac data, blood pressure, and blood sugar. Figure 4 depicts the detailed architecture of the SGR. The SGR is one of the key coponents in the overall VSNarchitectureforthehealthcaresyste.AnSGRisa 4. Network Model of VSN In this section we propose the network odel of VSN based on graph theory. We consider a densely deployed largescale and heterogeneous wireless sensor network in which N sensor nodes and M sensor gateway routers are uniforly distributed. The nodes and SGRs ay deterine their geographical locations. In fact, networking in such a WSN is very dynaic and differs fro a traditional wired network. A node in a WSN is very tiny and consists of a sall processing and storage unit. Since VSN is based on the existing SInP, it inherits ost of the properties of a WSN. A link in VSN is the different channels used in a WSN. We use IEEE , which has 16 channels. We describe the network odel of a WSN by using the graph theory that follows the procedure discussed in [19, 20]. We also discuss the VSN node and VSN link ebedding. Virtual node ebedding in VSN is like the conventional network ebedding, but link ebedding is quite different fro the conventional approach. In this case, link ebedding is done by dynaically using different channels that consist of ultiple links SInP. We odel the S network as a weighted undirected graph and denote it by G SInP =(N SInP,L SInP ),wheren SInP is the set of physical sensor nodes and L SInP is the associated links. SInP sensor nodes are divided into two functionalities basedontheirprocessingcapabilityandstoragespace,thatis, coon widely deployed sensor nodes and a sensor gateway router. Each sensor gateway router in the SInP is associated with the CPU capacity weight value C(N SInP ) and its GPS location loc(n SInP ) on a globally understood coordinate syste. Each substrate link l SInP (i, j) L SInP between two substrate gateway router nodes i and j is associated with the bandwidthcapacityweightvalueb(l SInP ) denoting the total aount of bandwidth. We denote the set of all substrate paths by P s and the set of substrate paths fro the source node s to the destination node d by P s (s, d). Figure1 shows the substrate SInP network, where the sensing node is indicated by sall circles of different colors and the sensor gateway routers are indicated by a node with a wireless antenna.

6 6 International Journal of Distributed Sensor Networks 4.2. VSN Request by ALU. As we discuss the graph-based description of SInP, we also odel the VSN request as weighted undirected graphs and denote a VN request in ters of the service request as G vsn (N vsn,l vsn ).Weention therequireentonvirtualnodesandlinksofthesubstrate physical sensor network. Each VN request has an associated nonnegative value D V expressing how far a virtual node n vsn N vsn can be ebedded fro its preferred location loc(n vsn ). D V is expressed naturally as a link delay or round-trip tie fro the loc(n vsn ). 4.3.SInPNetworkResourcesMeasureent. To easure the differenttypesofresourceusageofthesinp,weusethe notion of utility. The substrate SInP node utility U SInP n (n SInP ) is defined as the total aount of processing power allocated to different virtual sensor nodes hosted on the substrate SInP node n SInP N SInP : U n SInP (n SInP ) = c (n vsn ). (1) The substrate SInP link utility U SInP l (l SInP ) is defined as the total aount of link usage by different virtual sensor nodes hosted on the substrate SInP node n SInP N SInP.Itis actually the dedicated channel utilization to a specific virtual sensor node: U l SInP = b(l vsn ). (2) The substrate SInP storage or eory utility U SInP (SInP ) is defined as the total aount of storage usage by different virtual sensor nodes hosted on the substrate SInP node n SInP N SInP. It is actually the eory utilization of different virtual sensor nodes: U SInP = s( vsn ). (3) The total utility of the processing power, link, and storage can be calculated by suing up (1), (2), and (3). Here α, β, and γ are the weighted values to express the node, link, and storage capacity, respectively, by a single utility: U T SInP =α c(n vsn )+β b(l vsn )+γ s( vsn ). (4) 4.4. Residual Resource Measureent. Residual resource anageent is perfored by easuring the available resources reaining after utilization. In this section we have given the atheatical forulation of the reaining resources of the SInP sensor node, corresponding link, and storage only. The residual capacity of the SInP sensor nodes is defined as the total processing capacity of the sensor nodes, which is explained in (5): R n SInP (n SInP )= c(n SInP ) U n SInP (n SInP ). (5) n N In a wireless sensor network, the counication is perfored by wireless links. By a link, we ean a wireless link between different SGR nodes. We allocate different channels of a particular wireless link to particular applications in the virtualization of the sensor network. Equation (6) represents the residual channel capacity in the underlying SInP: R l SInP (l SInP )= b(l SInP ) U l SInP (l SInP ). (6) l L There are two types of storage in the underlying SInP node: flash eory and SDRAM. In this atheatical odel we only consider the SDRAM, which is only physical eory shared by different applications in the VSN applications.equation (7) shows the total reaining residual storage for further applications: R s SInP (s SInP )= s( vsn ) U SInP ( SInP ). (7) M 4.5. VSN Node and Link Ebedding. In this work, the VSN node and link ebedding are very uch restricted to the SGR and the wireless link between different SGRs. Different VSNSP nodes share the sae or different SGRs of the SInP. The typical sharing depends on the storage liit of the SGR. For wireless link ebedding, we consider the efficient channel utilization. Individual VSNSPs provide particular services. For exaple, VSN-1 ay provide urgent data services and use channel-1, while VSN-2 ay provide suspicious data services and use channel-2 of the specific VSNSP. In this way, the sae SInP can be used by different VSNSPs. 5. Packets Classification and Scheduling in VSN This section discusses the packet classification and scheduling in VSN. Figure 5 depicts the packet classification and scheduling odule of the SGR in a VSN environent. It consists of different coponents such as the traffic classifier, scheduling, channel allocation, and link estiator. Brief descriptions of theechanissaregivenbelow. (i) The data packet is received by the IEEE interface. Then the data is sent to the MAC reception odule. (ii) All of the data fro the MAC reception passes through the traffic classifier odule. Based on the inforation provided in the data packet, such as reliability, delay deadline, and priority inforation fro the application layer, the data are classified as urgent, suspicious, oderate, and noral. (iii) The classified data are passed through the 4-to-16 decoder odule. Based on the availability of the VSN queues, the data are queued up. (iv) The scheduling odule sends the data to a specific path based on the priority and other requireents of the data packet and VSN queue. (v) The channel allocation is based on the link estiator inforation and scheduling requireents.

7 International Journal of Distributed Sensor Networks 7 To and fro applications Traffic classifier Urgent data Suspicious Moderate Noral data 4 to 16 decoders. VSN 1 VSN 2. VSN 16. Scheduling MAC reception CH allocation Link estiator PRR RSSI LQI CH 1 CH 2 CH Incoing traffic Outgoing traffic Figure 5: Protocol architecture of SGR. (vi) The link estiator odule depends on the PRR, RSSI, and LQI for link quality easureents. (vii) Finally, the data packets are transitted to the next VSN gateway. In the following sections we describe the individual coponents in detail Traffic Classifier. The traffic classifier receives different types of data packets fro the MAC reception odule. A specific VSN ay carry particular types of data packets or a cobination of different types of data based on the application requireent. The data packets consist of different vital signs of the patient, such as the cardiac data, glucose level, blood pressure, pulse rate, and teperature. The packet received fro the sensor nodes includes the data type, deadline, delay, and reliability requireents. Based on the inforation in the packet, the data are classified as urgent, suspicious, oderate, or noral. The traffic classification is context dependent. Urgent Data. This includes eergency traffic or other data types specified by the applications. These types of traffic require approxiately 100% reliability and a hard delay guarantee. It is usually event-triggered traffic and is generated whenever a life-threatening situation appears. For instance, when the heart rate and blood pressure of a patient exceed the danger liit, an eergency action is needed, which requires urgent transission with the highest reliability and lowest delay. Suspicious Data. This type of data requires a strict reliability requireent (>90%) but can tolerate a delay up to a certain liit, such as a edical iage like an X-ray or ultrasonography. On the other hand, soe data, such as a teleedicine video transission, ust eet a strict delay deadline but can tolerate soe packet loss. Moderate Data. Both delay and reliability guarantees are required. However, oderate data requires soft QoS rather than hard QoS. In this case, the reliability requireent is ore than 80%. Different types of edical applications, such as heart rate and SPO 2 continuously generate data that ust be delivered with oderate reliability and delay requireents. Noral Data. Thistypeoftrafficdoesnotrequireanystrict delay or reliability constraints. It consists of the regular data for patients such as the teperature, blood pressure, and glucose level. For noral data, less than 70% percent reliability is aintained during transission. The classified data are transitted over different VSNs according to their priority, reliability, and delay requireents. Data over different VSNs are forwarded to different users and applications through the dynaically allocated channels VSNs. Fro a technical point of view, all of the VSNs are the logical cobination of the CPU resources, storage, and the link of the SInP. A VSN is fored dynaically based on the requireent of the application level requests provided by different users. However, a typical VSNSP consists of 16 VSNs, due to the dedicated and available channels in the physical layer. These 16 channels of the particular VSNSP canbeallocatedbasedonthepriority,reliability,andend-toend delay requireents of the traffic. A particular application ay use ore than one VSN to ensure guaranteed service. In a specific VSN, there are ultiple counication paths by which the data ay be transitted Scheduling. The data scheduling is perfored based on the inforation provided by different coponents, such as the traffic classifier, VSN priority, and channel allocation

8 8 International Journal of Distributed Sensor Networks odule. The goal of this odule is to ensure applicationspecific reliability and end-to-end delay. Different applications have different reliability and end-to-end delay requireents Link Estiator. Link estiation is based on three paraeters: the packet reception rate (PRR), received signal strength indicator (RSSI), and link quality indicator (LQI). Based on these paraeters, the link estiator provides quality inforation regarding the particular link. The detailed atheatical derivations of these paraeters are given below. We estiate the current path state by using the link quality inforation, including the link quality indicator (LQI) and the received signal strength indicator (RSSI) [20, 21]. The RSSI is a function of the distance between two nodes and can be coputed as follows: RSSI (d) = RSSI (d 0 ) 10n log ( d d 0 ). (8) In (8), RSSI(d) isthereceivedsignalstrengthindb at adistanced frothesourcenode.rssi(d 0 ) is the received signal strength at a distance d 0 frothesourceandn is the attenuation exponent. The IEEE specification ensures that each incoing frae contains a link quality indicator (LQI) value. The LQI indicates the quality of the link at the tie of the frae reception. According to the standard, the LQI value ust be an integer that is uniforly distributed between 0 and 255, with 255 indicating the highest signal quality. The LQI is easured as follows: LQI =255+3 P r db. (9) Here, P r db is the power of a received frae expressed in decibel-illiwatts. If the coputed value is a fraction, thenaroundingoperationisperforedtoobtainaninteger. R ab represents the link quality and received signal strength between two nodes. RSSI fro node a to node b is represented by RSSI ab, and RSSI fro node b to node a is represented by RSSI ba. In this case, we consider syetric transission. The LQI values of nodes a and b are represented as LQI a and LQI b.thusr ab can be calculated as follows: R ab = RSSI ab RSSI ba LQI a LQI b. (10) To calculate the packet reception rate (PRR), each node estiates the link loss rate for every outgoing link using the weighted average loss interval ethod discussed in [6]. It uses the interval between loss events to estiate the loss rate of a link. We denote the interval between the th and ( + 1)th loss for the outgoing link of the ith path as l i,1 ().Then for the recent 1 nlosses, the average loss interval, l i,1, is l i,1 (i, n) = n =1 l i,1 () w n =1 w, (11a) l i,1 (0, n 1) = n 1 =0 l i,1 () w n =1 w, (11b) l i,1 = ax (l i,1 (i,n),l i(0,n 1) ), (11c) where l i,1 (0) is the interval since the ost recent loss and w is theweightgiventoeachlossinterval.wecoputetheaverage PRR of the first hop of the ith path, p i,1, using the average loss rate, p c i,1 =1/l i,1,as p i,1 =1 p c i,1. (11d) The counication nature of VSN enables the easureent of the success rate of a path by using passive inforation exchange. When a node forwards a packet in a path, it includes the success rate of the path in the packet. The success rate of the ith path of a node, P i (h i ),isgivenby P i (h i )= h i J=1 h i p i,j =p i,1 p i,j, (12) J=2 where p i,j is the success rate of the jth hop and p i,1 is the success rate of the first hop of the node. h i J=2 p i,j is the success rate of the path fro the downstrea node, and the node overhears this fro the forwarded packets of the downstrea node Channel Allocation. Channel allocation is a dynaic process by which the syste allocates a particular channel to the specific VSN. There are 16 channels in the PHY layer specification, starting at 2.4 GHz. Based on the link status fro the link estiator, this odule allocates the channels. The channel quality is coputed by the following equation: R Channel = n i=1 Ri ab, (13) n where R ab is the quality of the link and n is the nuber of links on the channel. With the help of the channel quality coputation, the channel allocator selects the channel as follows. Step 1. Periodically easure the channel quality with the R Channel and PRR values. Step 2. Define the scheduling probability P(a), which is the probability that traffic is assigned to channel I, representing the noralized value of R Channel relative to other channel values as shown: R Channel P i (a) = i k R. (14) Channel

9 International Journal of Distributed Sensor Networks 9 Weeasuretheprobabilityrangesforthechannelsusing the scheduling probability, P i (a). Thus, for each channel i,the probability range is defined as follows: i 1 ( k p i (a), i k p i (a)); i=1,2,...,n, where p (a) =0, i k=0 p i (a) =1. (15) Step 3. We use the probability ranges to follow the data to the channel in each interval of tie. A priority is assigned to the channel, referring to its R Channel.Achannelwithahigh R Channel hashigherpriority.thisisdynaicandchangeswith tie. The channel allocation odule selects a particular channel according to a generated rando nuber between 0 and 1. The value of the rando nuber falls into the range defined in (15). Thus the channel with index i related to the selected rangeischosentosendthedatapacket.theprobabilityrange is used as the priority. Lower-priority channels have a saller chance than higher-priority channels to follow data. 6. QoS Model in Delay Doain This section provides a atheatical odel for delay analysis for different types of packets and VSN. Different types of traffic require a certain delay guarantee. Here, we introduce the QoS differentiation odel for urgent, suspicious, oderate, and noral data in the delay doain. λ α, λ β, λ χ,andλ δ are the arrival rates, and α, β, χ,and δ are the service rates of urgent, suspicious, oderate, and noral data, respectively. λ isthetotalarrivalratethatindicatesthenuberofincoing packets per second at the SGR, and λ=λ α +λ β +λ χ +λ δ. = α + β + χ + δ denotes the nuber of packets that depart per second. The packet arrival rate to each SGR is a Poisson process. For a given SGR, the end-to-end path delay is coposed of the transission delay and queuing delay. The transission delay is avoided due to its negligence. We consider the M/M/1 queue odel with nonpreeptive priority. We consider the individual priorities, p α, p β, p χ, and p δ, for different classes of traffic. The traffic intensity at the SGR is coputed as follows: p= λ =p α +p β +p χ +p δ. (16) The probability that an urgent packet finds other packets in serviceisequaltotheratioofthetiespentbythesgr on the suspicious, oderate, and noral packets. These are calculated as λ β / = p β, λ χ / = p χ,andλ δ / = p δ. Little s law [22, 23] gives us a good approxiation regarding the queue behavior and a basis for predicting the perforance of the individual queues: E (n) =λe(t). (17) Here, E(n) is the average nuber of packets in the queue, λ is the packet arrival rate, and E(t) is the average delay tie per packet in the syste. Urgent Packet Processing. Urgent packets have the highest priority. For their processing, there are dedicated VSNs. This ensures alost negligible delay, which includes the short queuing delay. Suspicious Packet Processing. For delay-guaranteed suspicious packets, the processing is done using the sae ethod used for urgent packets. The packets that are not delay guaranteed face a longer queuing delay. The ean delay of such a packet depends on E(n) and the packets in service. This can be forulated as follows: E(t β )= E(n β) (p χ +p δ ), E(t β )= 1+(p χ +p δ ) (1 p β ), E(t β )= 1 + p β/ 1 p β. (18) Moderate Packet Processing. Thistypeofpackethastowaitfor any suspicious packets in service and the oderate packets in the ready queue. The delay can be calculated as follows: E(t χ )= E(n β) + E(n χ) p δ. (19) Since p=p α +p β +p χ +p δ,(14) can be represented as follows: E (t χ ) = E(n β) + E(n χ) (p p α p β p χ ). (20) Noral Packet Processing.Anoralpackethastowaitforthe suspicious and oderate packets in service and also for the noral packet in the queue. The delay can be calculated as follows: E(t δ )= E(n β) E(t δ )= E(n β) E(t δ )= E(n β) + E(n χ) + E(n χ) + E(n δ) + E(n δ) p δ, (p p α p β p χ p δ ), + E(n χ) + E(n δ) + 1. (21) The average delay of the arrived packet depends on the packets that are already buffered in the queue plus the packets in service.

10 10 International Journal of Distributed Sensor Networks 7. QoS Model in Reliability Doain This section provides a atheatical odel for reliability analysis for different types of packets and VSN. Reliability is a unitless quantity that can be defined as the ratio of the nuber of unique packets received by the gateway node tothenuberofuniquepacketssentbythesourcenode. In this paper, we consider both the link layer reliability and the network layer reliability. In a wireless network, MAC layer retransission is used to iprove the reliability. However, retransission increases the delay at each hop. Moreover, MAC layer retransission does not efficiently increase the reliability in a densely deployed WSN due to its increased ediu contention. Here we consider a VSNbasedapproachtoachievetherequiredreliability.Forurgent traffic, the VSN approach provides around 100% reliability by using dedicated paths over ultiple VSNs. Suspicious data with a reliability requireent of ore than 90% are transitted over ultiple paths of ultiple VSNs. Moderate data with a reliability requireent of ore than 80% are transitted over ultiple paths of a specific VSN. Finally, noral traffic with reliability of less than 70% is transitted over the available paths. This reliability-differentiated traffic is transitted with the help of the scheduling odule of the network layer and the channel allocation odule of the link layer. Let us assue that the nuber of unique data traffic sent bythesourcenodeisx s and the nuber of unique data traffic received by the gateway router is X r.thusthereliabilityisr= X r /X s. The reliability calculation follows the ultiplication law of probability: p (k) = k i=1 (1 p d i ). (22) Now we will address the probabilistic odel of reliability [5 7, 23] fordifferentcases,suchasultiplepathsover ultiplevsns,ultiplepathsoverasinglevsn,andasingle path over a specific VSN. (A) Multipath over ulti-vsn: in ultipath over ulti- VSN packet forwarding, if there are paths in n VSNs, the probability that at least one copy of a packet is successfully received by the SGR is p (, n) =1 [ 16 i=1 p (, n) =1 [ [1 p i (k i )]] [ [1 p j (k j )]], [ ] 16 i=1 [1 [ [ 1 [ [ k i=1 (1 p d i )]] (1 p d j ) ]], ]] (23) where p(, n) is the probability of success for ultipath and ultivsn packet forwarding with paths and n VSNs. p i (k i ) and p j (k j ) are the probabilities of success for the ith VSN and jth path, respectively, and p d i and p d j are the probabilities that a packet is dropped by the ith VSN and jth path, respectively. If X s packets are sent and X r packets are received by the gateway node with probability p(, n), then the nuber of total packets received by the SGR has a binoial distribution, and the probability ass function (pf )isgivenby: p[x r =t]=( X s t )[p(, n)]t [1 p (, n)] X s t. (24) The required reliability, R req,isachievedwhenthe nuber of unique packets received by the SGR is X r. This iplies that X r = R req X s.tofulfillthe requireent, X r shouldbegreaterthanorequalto R req, so the probability that the required reliability is et in ultipath over ulti-vsn packet forwarding, p VSN Path,is X s p VSN Path = ( X s t )[p(, n)]t [1 p (, n)] X s t. (25) t=x r (B) Multipath over single VSN: in ultipath over single VSN packet forwarding, if there are paths over a VSN, the probability that at least one copy of a packet is successfully received by the SGR is p () =1 [ [1 p j (k j )]], [ ] (26) p () =1 [ [ 1 (1 p d j ) ]], [ [ ]] where p() is the probability of success for ultipath for a specific VSN packet forwarding with paths. p i (k i ) and p j (k j ) are the probabilities of success for the ith VSN and jth path, respectively, and p d i and p d j are the probabilities that a packet is dropped by the ith VSN and jth path, respectively. Since X s packets are sent and X r packets are received by the gateway node with probabilityp(), then the nuber of total packets received by the SGR has a binoial distribution, and the pf is given by p[x r =t]=( X s t )[p()]t [1 p ()] X s t. (27) The required reliability, R req,isachievedwhenthe nuber of unique packets received by the SGR is X r. This iplies that X r = R req X s.tofulfillthe requireent, X r shouldbegreaterthanorequalto R req, so the probability that the required reliability is et in ulti-path over single VSN packet forwarding, p svsn Path,is X s p svsn Path = ( X s t )[p()]t [1 p ()] X s t. (28) t=x r

11 International Journal of Distributed Sensor Networks 11 (C) Single path over single VSN: in a single path over single VSN packet forwarding, to get the required reliability level, packets ust be retransitted since the failure probability is high. The probability denoted as p j (r) indicates that the jth hop successfully forwards a packet within r retransission attepts. In such a scenario, if there is a path in a VSN, the probability that at least one copy of a packet is successfully received by the SGR is p j (r) =1 (p j ) r. (29) The probability that a data packet is successfully received by the SGR in a single path over a specific VSN with a hop count s is p (s) = S [p j (r)] = S [1 (p j ) r ]. (30) Here, p(s) is the probability of success for a single path for single VSN packet forwarding. The required reliability, R req, is achieved when the nuber of unique packets received by the SGR is X r. This iplies that X r =R req X s. To fulfill the requireent, X r should be greater than or equal to R req,sothe probability that the required reliability is et with single path over single VSN packet forwarding, p svsn spath, is X s p svsn spath = 8. Perforance Evaluations ( X s t )[p(s)]t [1 p (s)] X s t. (31) t=x r In this section, we discuss the siulation environent and evaluation results. We have ipleented and evaluated the VSNware on the Iote2 sensor node. The Iote2 sensor node has a Marvel PXA27x ARM processor with 400 MHz clock speed, 32 MB Flash, and 32 MB SDRAM. We have selected Iote2 as the sensor node for its advanced features such asitseorysizeandcpuspeed.inthisevaluation,the sensor node runs Ebedded Linux as its operating syste. The detailed syste specifications are given in Table 2. We develop a virtual achine for wireless sensor network called VSNware. It is based on Ebedded Linux. The VSNware environent restricts access to all of the physical resourcesonthenode,thusensuringthatapplicationsare only allowed to access the hardware through the VSNware. VSNware is available in all SGR nodes. VSNware supports concurrent application execution and dynaic application deployent. The VSNware supports applications ipleented in high-level language, thereby enabling different applications fro health care scenarios to be executed and runinthevsnenvironent.wecoparetheproposed VSNapproachwithMMSPEED[5]andthetraditionalWSN approach. MMSPEED provides a virtual network of ultiple speed layers for a network-wide speed guarantee in ters of the reliability and tieliness. The traditional WSN approach Table 2: Syste specifications. Type Specifications Sensor node Iote2 CPU Marvel PXA27x ARM CPU speed 400 MHz Operating syste Ebedded Linux OS version VM VSNware Flash size 32 MB SDRAM size 32 MB Interface USB Bandwidth 250 Kbps Radio IEEE in this perforance evaluation process is used to eulate the exact scenario in the conventional ethod that is provided by the proposed VSN approach. In the following scenarios, utilization of VSNware is the technical point of a VSN-based syste evaluation. Here, we focus on different issues such as the eory utilization, CPU utilization, and execution ties of individual applications. Efficient eory and CPU utilizationaretheainconcernofthevsnapproach.the executiontieandcpuutilizationarerelatedtoeachother. We have copared the eory and CPU utilization of our proposed VSN schee to those of the MMSPEED and traditional approaches. In Figure 6, we plot the eory usage of the traditional, MMSPEED, and VSN approaches. The sensor virtualization version includes the overhead of the applications due to the additional eory usage of a single sensor node. However, the overhead is linear and increases slowly based on the nuber of applications being deployed. In coparison to the MMSPEED and traditional approaches, the proposed VSN approach provides better perforance. The perforance evaluation shows that the proposed VSN approach reduces the average eory utilization by 53% and 56% as copared to the MMSPEED and traditional approaches, respectively. In Figure 7, we plot the execution ties of different applications for different nubers of virtualized sensor nodes. The figure shows the execution ties of 3, 5, and 7 vital signsensing applications based on the virtualization of sensor network ethodology. The execution tie increases linearly based on the nuber of applications in a sensor node. In Figure 8, we plot the CPU utilization versus the nuber of applications in the traditional approach, MMSPEED approach, and in the virtualization of sensor network scenario. The CPU utilization increases linearly in all of the cases. In this scenario, the VSN approach uses CPU resources efficiently, since it executes different applications on the sae sensor node. The perforance evaluation result shows that the proposed VSN approach has average CPU utilization that is 56% and 60% lower than that of the MMSPEED and traditional approaches, respectively. In Figure 9, we depict the eory usage of different typical applications, such as edical iaging, cardiac, blood

12 12 International Journal of Distributed Sensor Networks Meory usage (KB) CPU usage (%) VSN approach MMSPEED Traditional approach Nuber of applications Figure 6: Coparative eory usage in VSN approach Nuber of applications VSN approach MMSPEED Traditional approach Figure 8: Coparative CPU usage in VSN approach. Execution tie (s) Nuber of nodes Sensing 3 vital signs Sensing 5 vital signs Sensing 7 vital signs Figure 7: Execution tie versus nuber of nodes. Meory usage (KB) Medical iage Cardiac data Blood sugar Applications VSN approach MMSPEED Traditional approach Blood pressure Teperature Figure 9: Coparative eory usage by applications. sugar, blood pressure, and teperature. Medical iage applications use ore eory than other applications. Since the total eory in the Iote2 sensor node is 32 MB, it can provide the environent for the execution and running of the typical applications. The figure also deonstrates the eory usage of different applications in ters of the MMSPEED schee and the traditional WSN and proposed VSN approaches. In Figure 10, wehavepresentedtheend-to-enddelays of different data flows. It shows that the average delay for the four classes of data packets increases with the increasing nuber of sensor nodes. It clearly shows that the average endto-end delay of an urgent packet is significantly lower than that of the suspicious, oderate, and noral data packets. In Figure 11, wehavepresentedtheend-to-enddelaysof different data flows with respect to the data rate. The data rate varies according to the priorities assigned to different data flows. Since the highest priority is assigned to the urgent class, it experiences the lowest delay with respect to the other classes of data packets. Suspicious and oderate data also experience iniu delay due to their priority and the VSN approach used in this schee.

13 International Journal of Distributed Sensor Networks E2E delay (s) Reliability Nuber of nodes Nuber of nodes Urgent Suspicious Moderate Noral Figure10:End-to-enddelayversusnuberofnodes. Urgent Suspicious Moderate Noral Figure 12: Reliability versus nuber of nodes. Delays (s) Urgent Suspicious Data rate (Kbps) Moderate Noral Figure 11: Delay versus data rate. Probability of packet drop Arrival rate Urgent Suspicious Moderate Noral Figure13:Packetdroppingprobabilityversusarrivalrate. Figure 12 shows the reliability of different data packets with respect to the nuber of nodes. We focused on designing our schee to ensure 100% percent reliability, but, practically, it ensures an average of approxiately 98% reliability. The reliability levels of the suspicious and oderate data are also significantly higher than that of the noral data. In Figure 13, we deonstrate the packet dropping probability versus the arrival rate. Packet dropping depends on the packet arrival rate at the queues. The figure shows that the urgent data has the lowest loss, followed by the suspicious, oderate, and noral traffic. 9. Conclusions In this paper, we propose a novel approach of a VSN-based packet delivery echanis to provide service differentiation and probabilistic QoS assurance in the delay and reliability doains. It also explores QoS-based data classification and scheduling schees for health care applications in a VSN environent. For the delay doain, we use ultiple layers based on VSN so that different data packets can dynaically choose the appropriate layers according to the delay requireents of individual data traffic. For the reliability doain, we use ultiple virtual layers as well as different paths within

14 14 International Journal of Distributed Sensor Networks the corresponding VSN. The perforance evaluations show thatthevsnscheeprovideslowend-to-enddelayandhigh reliability for the urgent data. It also significantly increases the perforance for the other data classifications. Our future interest is to ephasize the large-scale and federated sensor network platfor with ultiple applications sharing the sae physical resources that will facilitate the rapid deployent of theubiquitoushealthcaresyste. Acknowledgents This research was supported by the MSIP (Ministry of Science, ICT and Future Planning), Republic of Korea, under theitrc(inforationtechnologyresearchcenter)support progra supervised by the NIPA (National IT Industry Prootion Agency) (NIPA-2013-(H )). References [1] I. F. Akyildiz, W. Su, Y. Sankarasubraania, and E. Cayirci, A survey on sensor networks, IEEE Counications Magazine, vol. 40, no. 8, pp , [2] K. Akkaya and M. Younis, A survey on routing protocols for wireless sensor networks, Ad Hoc Networks, vol.3,no.3,pp , [3] M. M. Isla and E.-N. Huh, Sensor proxy obile IPv6 (SPMIPv6)-a novel schee for obility supported IP-WSNs, Sensors, vol. 11, no. 2, pp , [4] M. M. Isla and E.-N. Huh, A novel addressing schee for PMIPv6 based global IP-WSNs, Sensors,vol.11,no.9,pp , [5] E. Feleban, C.-G. Lee, and E. Ekici, MMSPEED: Multipath Multi-SPEED Protocol for QoS guarantee of reliability and tieliness in wireless sensor networks, IEEE Transactions on Mobile Coputing,vol.5,no.6,pp ,2006. [6] M. M. Ala, M. A. Razzaque, M. Maun-Or-rashid, and C. S. Hong, Energy-aware QoS provisioning for wireless sensor networks: analysis and protocol, Journal of Counications and Networks,vol.11,no.4,pp ,2009. [7] M. A. Razzaque, M. M. Ala, M. Maun-Or-rashid, and C. S. Hong, Multi-constrained QoS geographic routing for heterogeneous traffic in sensor networks, IEICE Transactions on Counications,vol.E91-B,no.8,pp ,2008. [8] M.M.Isla,M.M.Hassan,G.-W.Lee,andE.-N.Huh, Asurvey on virtualization of wireless sensor networks, Sensors, vol. 12, no. 2, pp , [9] M.M.Isla,J.H.Lee,andE.N.Huh, Anefficientodelfor sart hoe by the virtualization of wireless sensor network, International Journal of Distributed Sensor Networks, vol.2013, Article ID , 10 pages, [10] M. M. Isla, M. M. Hassan, and E.-N. Huh, Virtualization in wireless sensor network: challenges and opportunities, in Proceedings of the 13th International Conference on Coputer and Inforation Technology (ICCIT 10), Deceber [11] S. Kabadayi, A. Pridgen, and C. Julien, Virtual sensors: abstracting data fro physical sensors, in Proceedings of the International Syposiu on a World of Wireless, Mobile and Multiedia Networks (WoWMoM 06), pp , Buffalo- Niagara Falls, NY, USA, June [12] J.-H. Shin and D. Park, A virtual infrastructure for large-scale wireless sensor networks, Coputer Counications, vol. 30, no , pp , [13] A. P. Jayasuana, H. Qi, and T. H. Illangasekare, Virtual sensor networks a resource efficient approach for concurrent applications, in Proceedings of the 4th International Conference on Inforation Technology (ITNG 07), pp , April [14] S. Waharte, J. Xiao, and R. Boutaba, Overlay wireless sensor networks for application-adaptive scheduling in WLAN, in Proceedings of the 7th IEEE International Conference on High Speed Networks and Multiedia Counications (HSNMC 04), vol.3079oflecture Notes in Coputer Science, pp , Toulouse, France, [15] I.Leontiadis,C.Efstratiou,C.Mascolo,andJ.Crowcroft, Senshare: transforing sensor networks into ulti-application sensing infrastructures, in Proceedings of the 9th European Conference on Wireless Sensor Networks,February2012. [16] C. Efstratiou, I. Leontiadis, C. Mascolo, and J. Crowcroft, Deo abstract: a shared sensor network infrastructure, in Proceedings of the 8th ACM International Conference on Ebedded Networked Sensor Systes (SenSys 10), pp , Zurich, Switzerland, Noveber [17] P. Levis and D. Culler, Maté: a tiny virtual achine for sensor networks, in Proceedings of the 10th International Conference on Architectural Support for Prograing Languages and Operating Systes (ASPLOS 02), pp , New York, NY, USA, October [18] Y.Yu,L.J.Rittle,V.Bhandari,andJ.B.LeBrun, Supportingconcurrent applications in wireless sensor networks, in Proceedings of the 4th International Conference on Ebedded Networked Sensor Systes (SenSys 06), pp , Noveber [19] N. M. Mosharaf, K. Chowdhury, M. R. Rahan, and R. Boutaba, Virtual network ebedding with coordinated node and link apping, in Proceedings of the IEEE 28th Conference on Coputer Counications (INFOCOM 09), pp , April [20] M. Chowdhury, M. R. Rahan, and R. Boutaba, ViNEYard: virtual network ebedding algoriths with coordinated node and link apping, IEEE/ACM Transactions on Networking, vol. 20,no.1,pp ,2012. [21] A. H. Shuaib and A. H. Aghvai, A routing schee for the IEEE enabled wireless sensor networks, IEEE Transactions on Vehicular Technology, vol.58,no.9,pp , [22] J. D. C. Little and S. C. Graves, Little s Law, Massachusetts Institute of Technology, in Building Intuition: Insights Fro Basic Operations Manageent Models and Principles,D.Chhajed and T. J. Lowe, Eds., Springer Science, Business Media, LLC, [23] A. Papoulis and S. U. Pillai, Probability, Rando Variables and Stochastic Processes, McGraw-Hill, 4th edition, 2002.

15 International Journal of Rotating Machinery Engineering Journal of The Scientific World Journal International Journal of Distributed Sensor Networks Journal of Sensors Journal of Control Science and Engineering Advances in Civil Engineering Subit your anuscripts at Journal of Journal of Electrical and Coputer Engineering Robotics VLSI Design Advances in OptoElectronics International Journal of Navigation and Observation Cheical Engineering Active and Passive Electronic Coponents Antennas and Propagation Aerospace Engineering International Journal of International Journal of International Journal of Modelling & Siulation in Engineering Shock and Vibration Advances in Acoustics and Vibration