Prof. U. S. Tiwary IIIT-Allahabad

Transcription

1 Data Aggregation in Cluster-based Wireless Sensor Networks A Thesis Submitted in partial fulfillment of the requirements for the award of the degree of MASTER OF TECHNOLOGY in INFORMATION TECHNOLOGY (Specialization: WIRELESS COMMUNICATION & COMPUTING) by Pranay Tiwari (IWC ) Under the Guidance of: Prof. U. S. Tiwary IIIT-Allahabad INDIAN INSTITUTE OF INFORMATION TECHNOLOGY ALLAHABAD (INDIA) July, 2008

2 INDIAN INSTITUTE OF INFORMATION TECHNOLOGY Allahabad (Deemed University) (A Centre of Excellence in Information Technology Established by Govt. of India) Date: I HEREBY RECOMMEND THAT THE THESIS PREPARED UNDER OUR SUPERVISION BY Pranay Tiwari ENTITLED Data Aggregation in Cluster-based Wireless Sensor Network BE ACCEPTED IN PARTIAL FULFILMENT OF THE REQUIREMENTS FOR THE DEGREE OF MASTER OF TECHNOLOGY IN INFORMATION TECHNOLOGY FOR EXAMINATION Prof. U. S. Tiwary (THESIS ADVISOR) COUNTERSIGNED Prof. U. S. Tiwary DEAN (ACADEMIC)

3 INDIAN INSTITUTE OF INFORMATION TECHNOLOGY Allahabad (Deemed University) (A Centre of Excellence in Information Technology Established by Govt. of India) CERTIFICATE OF APPROVAL* The foregoing thesis is hereby approved as a creditable study in the area of Information Technology carried out and presented in a manner satisfactory to warrant its acceptance as a pre-requisite to the degree for which it has been submitted. It is understood that by this approval the undersigned do not necessarily endorse or approve any statement made, opinion expressed or conclusion drawn therein but approve the thesis only for the purpose for which it is submitted. COMMITTEE ON FINAL EXAMINATION FOR EVALUATION OF THE THESIS *Only in case the recommendation is concurred in

4 INDIAN INSTITUTE OF INFORMATION TECHNOLOGY Allahabad (Deemed University) (A Centre of Excellence in Information Technology Established by Govt. of India) DECLARATION This is to certify that this thesis work entitled Data Aggregation in Cluster-based Wireless Sensor Networks which is submitted by me in partial fulfillment of the requirement for the completion of M.Tech. in Information Technology specialization in Wireless Communication and Computing to Indian Institute Of Information Technology, Allahabad comprises only my original work and due acknowledgement has been made in the text to all other materials used. Pranay Tiwari M.Tech. (WCC) IWC200614

5 Abstract The rapid advancement of hardware technology has enabled the development of small, powerful, and inexpensive sensor nodes, which are capable of sensing, computation and wireless communication. This revolutionizes the deployment of wireless sensor network for monitoring some area and collecting regarding information. However, limited energy constraint presents a major challenge such vision to become reality. We consider energy constrained wireless sensor network deployed over a region. The main task of such a network is to gather information from node and transmit it to base station for further processing. So the aim of any data forwarding protocol is to conserve energy to maximize the network lifetime. Sensor nodes are capable of performing innetwork aggregation of data coming from more than one source. In this thesis we have concentrated on energy consumption issue and aim to develop an energy efficient data aggregation protocol. To provide energy efficiency we have considered a cluster-based wireless sensor network. Our protocol executes on each cluster independently and provides an energy efficient data aggregation in a cluster and hence maximize network lifetime for whole network.

6 Acknowledgements Before, I get into thick of things; I would like to add a few heartfelt words for the people who were part of my thesis in numerous ways, people who gave unending support right from the beginning. During this period, the faculty members and my batch mates took keen interest and participated actively. They are very efficient and qualified in their respective disciplines. I express my sincere gratitude to my thesis supervisor Prof. U.S. Tiwary, Indian Institute of Information Technology-Allahabad for all his affectionate encouragement and guidance during the entire Thesis. His views and inputs are very helpful throughout the process. I would like to thank Dr. Shirshu Verma, Indian Institute of Information Technology-Allahabad, who suggested many related points and is always very constructive and helpful. I would like to thank Dr. M.D. Tiwari, Hon ble Director, Indian Institute of Information Technology-Allahabad for the facilities and environment for research. Not the least I would like to appreciate the support and suggestions of my dear friend Neelam who continuously encouraged me during the progress of my thesis work and my M.Tech friends Anuraag, Koustubh, Lalit, and Satish for everyday chatting we had on several topics. Lastly I would like to thank my family for their love, support and encouragement that they have given me throughout my life, helping me to persevere in my studies. In addition, I thank almighty god who made me a normal human being and gave me such strength.

7 List of Figures Figure 2.1 Simplified Schematic of Directed Diffusion 12 Figure 2.2 E-Span protocol 14 Figure 2.3 Example of Aggregation path over ring structure 15 Figure 2.4 Illustration of Two-phase clustering 17 Figure 2.5 Data transmission using ESPDA 18 Figure 3.1 A typical scenario of data aggregation in a cluster 23 Figure 3.2 Parent selection procedure 24 Figure 4.1 Random nodes scenario in NAM 29 Figure 4.2 Snapshot of console 32 Figure 4.3 Data transfer between node 0 and 10 through node 1 & 2 33 Figure 4.4 Energy level of nodes drop to first threshold 34 Figure 4.5 Energy level of nodes drop to second threshold 35 Figure 4.6 Residual energy of source as a function of time 36 Figure 4.7 Throughput as function of time 37 Figure 4.8 Effect of network density 38 Figure 4.9 Packet delivery ratio of the network 39 Figure A.1 The basic structure of NS-2 43

8 Table of Contents CERTIFICATE OF APPROVAL DECLARATION ABSTRACT ACKNOWLEDGEMENTS LIST OF FIGURES Chapter 1 INTRODUCTION Wireless Sensor Networks Sensor Network Challenges Wireless Sensor Network vs. Traditional Wireless Network Clustering in WSN Motivation Problem Definition Report Organization... 8 Chapter 2 DATA AGGREGATION: AN OVERVIEW In-Network Aggregation Tree-Based Approaches Multi-path Approaches Cluster-based Approaches Simulation Tools Summary. 19 Chapter 3 ENERGY-AWARE BALANCED IN-NETWORK AGGREGATION 3.1 Introduction System & Energy Model Protocol Description Configuration Packet Flow Data Packet Flow Issues Summary

9 Chapter 4 PERFORMANCE ANALYSIS Simulation Analysis Simulation Setup Simulation Run Simulation Results Chapter 5 CONCLUSION & FUTURE WORK Appendix A Network Simulator References... 47

10 Chapter 1 Introduction Before giving any outline of this thesis, we are going to introduce you to the world of wireless sensor networks. We will briefly describe challenges and application of wireless sensor networks. We will also provide you the motivation behind the selection of the data aggregation issue for this thesis and also describe what exactly the data aggregation is. Finally we will conclude this with providing outline of the thesis report. 1.1 Wireless Sensor Networks A wireless sensor network is a wireless network consisting of tiny devices which monitor physical or environmental conditions such as temperature, pressure, motion or pollutants etc. at different regions. The tiny device, known as sensor node, consists of a radio transceiver, microcontroller, power supply, and the actual sensor. Initially sensor network were used for military applications but now they are widely used for civilian application area including environment and habitat monitoring, healthcare application and so on. Normally sensor nodes are spatially distributed throughout the region which has to be monitored; they self-organize in to a network through wireless communication, and collaborate with each other to accomplish the common task. With the going time, sensor nodes are becoming smaller, cheaper, and more powerful which enable us to deploy a large-scale sensor network. Basic features of sensor networks are self-organizing capabilities, dynamic network topology, limited power, node failures & mobility of nodes, short-range M.Tech. - Thesis (July 2008) Indian Institute of Information Technology, Allahabad 1

11 broadcast communication and multi-hop routing, and large scale of deployment [12]. The strength of wireless sensor network lies in their flexibility and scalability. The capability of self-organize and wireless communication made them to be deployed in an ad-hoc fashion in remote or hazardous location without the need of any existing infrastructure. Through multi-hop communication a sensor node can communicate a far away node in the network. This allows the addition of sensor nodes in the network to expand the monitored area and hence proves its scalability & flexibility property. Presently there are different types of commercially available sensor nodes. University of California at Berkeley has developed Mica mote which is a special purpose sensor node. Other special purpose sensor nodes available are Spec, Rene, Mica 2, Telos etc. Some high bandwidth sensor nodes available are BTNode, Imote 1.0, Stargate, Inryonc Cerfeube etc. [13] Sensor Network Challenges Wireless sensor network promise a wide variety of application and to realize these application in real world, we need more efficient protocols and algorithms. Designing a new protocol or algorithm address some challenges which are need to be clearly understood. These challenges are summarized below: Physical Resource Constraints: The most important constraint imposed on sensor network is the limited battery power of sensor nodes. The effective lifetime of a sensor node is directly determined by its power supply. Hence lifetime of a sensor network is also determined by the power supply. Hence the energy consumption is main design issue of a protocol. Limited computational power and memory size is another constraint that affects the amount of data that can be stored in individual sensor nodes. So the protocol should be simple and light-weighted. Communication delay in sensor network M.Tech. - Thesis (July 2008) Indian Institute of Information Technology, Allahabad 2

12 can be high due to limited communication channel shared by all nodes within each other s transmission range. Ad-hoc Deployment: Many application or most of them requires the ad-hoc deployment of sensor nodes in the region. Sensor nodes are randomly deployed over the region without any infrastructure which requires the system to be able to cope up with random distribution and form connection between the nodes. As an example, for fire detection in a forest the nodes typically would be dropped in to the forest from a plane. Fault-Tolerance: In a hostile environment, a sensor node may fail due to physical damage or lack of energy (power). If some nodes fail, the protocols that are working upon must accommodate these changes in the network. As an example, for routing or aggregation protocol, they must find suitable paths or aggregation point in case of these kinds of failures. Scalability: In a region, depending upon the application, the number of sensor nodes deployed could be in order of hundreds, thousands or more. The protocols must scalable enough to respond and operate with such large number of sensor nodes. Quality of Service: Some sensor application are very time critical which means the data should be delivered within a certain period of time from the moment it is sensed, otherwise the data will be careless. So this could be a QOS parameter for some applications. M.Tech. - Thesis (July 2008) Indian Institute of Information Technology, Allahabad 3

13 1.1.2 Wireless Sensor Networks vs. Traditional Wireless Networks Though many existing protocol, techniques and concepts from traditional wireless network, such as cellular network, mobile ad-hoc network, wireless local area network and Bluetooth, are applicable and still used in wireless sensor network, but there are also many fundamental differences which lead to the need of new protocols & techniques. Some of the most important characteristic differences are summarized below: Number of nodes in wireless sensor network is much higher than any traditional wireless network. Possibly a sensor network has to scale number of nodes to thousands. Moreover a sensor network might need to extend the monitored area and has to increase number of nodes from time to time. This needs a highly scalable solution to ensure sensor network operations without any problem. Due to large number of sensor nodes, addresses are not assigned to the sensor nodes. Sensor networks are not address-centric; instead they are data-centric network. Operations in sensor networks are centered on data instead of individual sensor node. As a result sensor nodes require collaborative efforts. Wireless sensor networks are environment-driven. While data is generated by humans in traditional networks, the sensor network generate data when environment changes. As a result the traffic pattern changes dramatically from time to time. Another characteristic unique to wireless sensor network is the correlated data problem. Data collected by neighboring sensor nodes are often quite similar which makes possible to the development of routing and aggregation techniques that can reduce redundancy and improve energy efficiency. It also been observed M.Tech. - Thesis (July 2008) Indian Institute of Information Technology, Allahabad 4

14 that the environmental quantities changes very slow and some consecutive readings sense temporally correlated data. This advantageous feature can be exploited to develop an energy efficient data gathering and aggregation techniques Clustering in WSN It is widely accepted that the energy consumed in one bit of data transfer can be used to perform a large number of arithmetic operations in the sensor processor [13]. Moreover in a densely deployed sensor network the physical environment would produce very similar data in close-by sensor nodes and transmitting such data is more or less redundant. Therefore, all these facts encourage using some kind of grouping of nodes such that data from sensor nodes of a group can be combined or compressed together in an intelligent way and transmit only compact data. This can not only reduce the global data to be transmitted and localized most traffic to within each individual group, but reduces the traffic and hence contention in a wireless sensor network. This process of grouping of sensor nodes in a densely deployed large-scale sensor network is known as clustering. The intelligent way to combined and compress the data belonging to a single cluster is known as data aggregation. There are some issues involved with the process of clustering in a wireless sensor network. First issue is, how many clusters should be formed that could optimize some performance parameter. Second could be how many nodes should be taken in to a single cluster. Third important issue is the selection procedure of cluster-head in a cluster. Another issue that has been focused in many research papers is to introduce heterogeneity in the network. It means that user can put some more powerful nodes, in terms of energy, in the network which can act as a cluster-head and other simple node work as clustermember only. Considering the above issues, many protocols have been proposed which deals with each individual issue. M.Tech. - Thesis (July 2008) Indian Institute of Information Technology, Allahabad 5

15 1.2 Motivation Wireless sensor network are new age technology which can be used at a place which is possibly hostile & human inaccessible. Wireless sensor network is a class of wireless network which consists of thousands of densely deployed sensor nodes which can be used for a number of applications. Sensor nodes are tiny devices which are composed of a sensing unit, a radio, a processor & a limited battery power. A network of thousands of sensor nodes could be setup for many applications such as environmental monitoring, health monitoring, disaster management, industrial areas, military application and many more. In wireless sensor network, there are so many challenges & issues as above already been discussed. The main challenges are how to provide maximum lifetime to network and how to provide robustness to network. As sensor network totally rely on battery power, the main aim for maximizing lifetime of network is to conserve battery power or energy. In sensor network, the energy is mainly consumed for three purposes: data transmission, signal processing, and hardware operation. It is said in [4] that 70% of energy consumption is due to data transmission. So for maximizing the network lifetime, the process of data transmission should be optimized. The data transmission can be optimized by using efficient routing protocols and effective ways of data aggregation. Routing protocols have their own ways to save energy of nodes in the network by providing or creating an optimal route from sensor nodes to base station or sink. Data aggregation plays an important role in energy conservation of sensor network. Data aggregation methods are used not only for finding an optimal path from source to destination but also to eliminate the redundancy of data, since transmitting huge volume of raw data is an energy intensive operation, and thus minimizing the number of data transmission. Also multiple sensors may see the same phenomenon, albeit from different view and if this data can be reconciled into a more meaningful form as it passes through the network, it becomes more useful to an application. One more benefit of data M.Tech. - Thesis (July 2008) Indian Institute of Information Technology, Allahabad 6

16 aggregation is that if data is processed as it is passed through the network, it may be compressed thus occupying less bandwidth. This also reduces the amount of transmission power expended by nodes. Hence data aggregation can be considered as a very challenging problem in wireless sensor network. 1.3 Problem Definition Data aggregation protocols aims at eliminating redundant data transmission and thus improve the lifetime of energy constrained wireless sensor network. In wireless sensor network, data transmission took place in multi-hop fashion where each node forwards its data to the neighbor node which is nearer to sink. That neighbor node performs aggregation function and again forwards it on. But performing data forwarding and aggregation in this fashion from various sources to sink causes significant energy waste as each node in the network is involved in operation. Since closely placed nodes may sense same data, above approach cannot be considered as energy efficient. An improvement over the above approach would be clustering where each node sends data to cluster-head (CH) and then cluster-head perform aggregation on the received raw data and then send it to sink. Performing aggregation function over cluster-head still causes significant energy wastage. In case of homogeneous sensor network cluster-head will soon die out and again re-clustering has to be done which again cause energy consumption. We would like to present an algorithm that performs data aggregation within a cluster and thus reducing the load of aggregation at cluster-head to provide energy efficiency for maximizing network lifetime. M.Tech. - Thesis (July 2008) Indian Institute of Information Technology, Allahabad 7

17 1.4 Report Organization With the end of chapter 1, the whole report is organized as follows: Chapter 2 gives a detailed overview of data aggregation. This chapter also presents the literature survey that had been done. Chapter 3 introduces and describes the new proposed protocol for data aggregation in cluster-based wireless sensor networks. Chapter 4 will present the performance analysis of the proposed protocol. It will also provide you the comparison results. Conclusion is given in the last chapter 5 and scope of future enhancements is also incorporated. Appendix A describes the network simulator NS-2 and its usability to simulate proposed protocol. Materials (e.g. URLs, books, and research papers) used and studied are given in Reference. M.Tech. - Thesis (July 2008) Indian Institute of Information Technology, Allahabad 8

18 Chapter 2 Data Aggregation: An Overview Wireless sensor network have recently been the focus of many research efforts and emerged as an important new area in wireless technology. As chapter 1 already discuss about the problem of data aggregation, this chapter presents an overview of data aggregation. This chapter first discusses in-network aggregation and then different approaches that are widely used for data aggregation. 2.1 In-Network Aggregation In a typical sensor network scenario, different node collect data from the environment and then send it to some central node or sink which analyze and process the data and then send it to the application. But in many cases, data produced by different node can be jointly processed while being forwarded to the sink node. So in-network aggregation deals with this distributed processing of data within the network. Data aggregation techniques explore how the data is to be routed in the network as well as the processing method that are applied on the packets received by a node. They have a great impact on the energy consumption of nodes and thus on network efficiency by reducing number of transmission or length of packet. Elena Fosolo et al in [7] defines the in-network aggregation process as follows: In-network aggregation is the global process of gathering and routing information through a multi-hop network, processing data at intermediate nodes with the objective of reducing resource consumption (in particular energy), thereby increasing network lifetime. M.Tech. - Thesis (July 2008) Indian Institute of Information Technology, Allahabad 9

19 There are two approaches for in-network aggregation: with size reduction and without size reduction. In-network aggregation with size reduction refers to the process of combining & compressing the data packets received by a node from its neighbors in order to reduce the packet length to be transmitted or forwarded towards sink. As an example, consider the situation when a node receives two packets which have a spatial correlated data. In this case it is worthless to send both packets. Instead of that one should apply any function like AVG, MAX, MIN and then send a single packet. This approach considerably reduces the amount of bits transmitted in the network and thus saving a lot of energy but on the other hand, it also reduces the precision of value of data received. Innetwork aggregation without size reduction refers to the process merging data packets received from different neighbors in to a single data packet but without processing the value of data. As an example, two packets may contain different physical quantities (like temperature & humidity) and they can be merged in to a single packet by keeping both values intact but keeping a single header. This approach preserves the value of data and thus transmit more bits in the network but still reduce the overhead by keeping single header. This of the two approaches to use depends on many factors like the type of application, data rate, network characteristics and so on. There is also a trade-off between energy consumption and precision of data for the two approaches. Most of the work done till now on in-network aggregation mainly deals with problem of forwarding packets from source to sink, to facilitate aggregation therein. Actually the main idea behind were to enhance existing routing protocols such that they can efficiently aggregate data. So till now, most of the data aggregation techniques fall under three categories. They are tree-based approaches, multi-path approaches, and cluster-based approaches. There also some hybrid approaches that combines any of the three techniques above. So, all the three approaches will be described in coming sections with giving details of some of the main techniques by different authors. M.Tech. - Thesis (July 2008) Indian Institute of Information Technology, Allahabad 10

20 2.2 Tree-Based Approach The simplest way to aggregate data is to organize the nodes in a hierarchical manner and then select some nodes as the aggregation point or aggregators. The treebased approach perform aggregation by constructing an aggregation tree, which could be a minimum spanning tree, rooted at sink and source nodes are considered as leaves. Each node has a parent node to forward its data. Flow of data starts from leaves nodes up to the sink and therein the aggregation done by parent nodes. The way this approach operates has some drawbacks. As we know like any wireless network the wireless sensor networks are also not free from failures. In case of packet loss at any level of tree, the data will be lost not only for a single level but for whole related sub-tree as well. In spite of high cost for maintaining tree structure in dynamic networks and scarce robustness of the system, this approach is very much suitable for designing optimal aggregation technique and energy-efficient techniques. S. Madden et al. in [14] proposed a data-centric protocol which is based on aggregation tress, known as Tiny Aggregation (TAG) approach [14]. TAG works in two phases: distribution phase and collection phase. In distribution phase, TAG organizes nodes in to a routing tree rooted at sink. The tree formation starts with broadcasting a message from sink specify level or distance from root. When a node receive this message it sets its own level to be the level of message plus one and elect parent as node from which it receives the message. After that, node rebroadcast this message with its own level. This process continues until all nodes elect their parent. After tree formation, sink send queries along structure to all nodes in the network. TAG uses database query language (SQL) for selection and aggregation functions. In collection phase, data is forwarded and aggregated from leaves nodes to root. A parent node has to wait for data from all its child node before it can send its aggregate up the tree. Apart from the simple aggregation function provided by SQL (eg: COUNT, MIN, MAX, SUM, and AVG), TAG also partitions aggregates according to the duplicate sensitivity, exemplary and summary, and monotonic properties. Though TAG periodically refresh tree structure of M.Tech. - Thesis (July 2008) Indian Institute of Information Technology, Allahabad 11

21 network but as most of the tree-based schemes are inefficient for dynamic network, so TAG may be. C. Intanagonwiwat et al. in [3] proposed a reactive data-centric protocol for applications where sink ask some specific information by flooding, known as directed diffusion paradigm. The main idea behind directed diffusion paradigm is to combine data coming from different source and en-route them by eliminating redundancy, minimizing the number of data transmission; thus maximizing network lifetime. Directed diffusion consists of several elements: interests, data messages, gradients, and reinforcements. Figure 2.1 Simplified schematic for directed diffusion. (a) Interest propagation. (b) Initial gradients setup. (c) Data delivery along reinforced path [3]. The base station (BS) requests data by broadcasting an interest message which contains a description of a sensing task. This interest message propagates through the network hop-by-hop and each node also broadcast interest message to its neighbor. As interest message propagates throughout the network, gradients are setup by every node within the network. The gradient direction is set toward the neighboring node from which the interest is received. This process continues until gradients are setup from source node to base station. Loops are not checked at this stage but removed at later stage. After this path of information flow are formed and then best path are reinforced to prevent further flooding according to a local rule. Data aggregation took place on the way of different paths from different sources to base station or sink. The base station periodically refresh & resend the interest message as soon as it start to receives data from sources to provide reliability. The problem with directed diffusion is that it may not be applied to applications (e.g. environmental monitoring) that require continuous data delivery to base M.Tech. - Thesis (July 2008) Indian Institute of Information Technology, Allahabad 12

22 station. This is because query driven on demand data model may not help in this regard. Also matching data to queries might require some extra overhead at the sensor nodes. Mobility of sink nodes can also degrade the performance as path from sources to sinks cannot be updated until next interest message is flooded throughout the network. To cope up with above issue if introduce frequent flooding then also too much overhead of bandwidth and battery power will be introduced. Furthermore, exploratory data follow all possible paths in the network following gradients which lead to unnecessary communications overhead. M. Lee et al. in [2] proposed a new low-control-overhead data dissemination scheme, which they called as pseudo-distance data dissemination (PDDD), for efficiently disseminating data from all sensor nodes to mobile sink. Some assumption have been made, they are: (1) all source nodes maintain routes to mobile sink node, (2) no periodically messaging for topological changes due to mobile sink node, (3) all link are bi-directional and no control messages are lost, (4) mobile sink nodes have unlimited battery power, so no need to care about battery efficiency of sink node, and (5) network partitioning is not considered. Data dissemination process is influenced by directed diffusion [3]. Though mobile sink periodically broadcast interest message, sensor nodes do not send exploratory data and do not wait reinforcement message because each sensor node already has routes to the sink node. After getting interest message, adjacent nodes set a parent-child relationship using pseudo-distance of each node and finally a partial ordered graph (POG) has been build. Optimal data dissemination is achieved in terms of path length by forwarding packets to a parent node until topology is unchanged. Then each sensor node is assigned a level for a corresponding sink node with pseudo-distance. In order to overcome the shortcoming of POG, author used totally ordered graph (TOG) in place of POG. The problem identified in this approach is that due to mobility of sink node all sensor nodes have to maintain routes and for any change in topology nodes have to again change route accordingly which led to energy waste. Marc Lee et al. in [8] proposed an energy-aware spanning tree algorithm for data aggregation, referred as E-Span. E-Span is a distributed protocol in which source node that has highest residual energy is chosen as root. Other source nodes choose their parent M.Tech. - Thesis (July 2008) Indian Institute of Information Technology, Allahabad 13

23 based on residual energy and distance to the root. The protocol uses configuration message to exchange information of node i.e., residual energy and distance to the root. Each node performs single-hop broadcast operation to send packets. Single-hop broadcast refers to the operation of sending a packet to all single-hop neighbors [8]. Figure 2.2 E-Span protocol [8]. 2.3 Multi-path Approach One of the main drawbacks of tree-based approach is the scarce robustness of the system. To overcome this drawback, a new approach was proposed by many researchers. Instead of sending partially aggregated data to a single parent node in aggregation tree, a node could send data over multiple paths. The idea behind is that each node can send the data to its possibly multiple neighbors by exploiting the wireless medium characteristic. Hence data will flow from sources to sink along multiple paths and aggregation can be performed by each intermediate node. Clearly schemes using this approach will make the system robust but with some extra overhead. One of the aggregation structures that fit well with this approach is ring topology, where network is divided in to concentric circles with defining levels according to the hop distance from sink. S. Nath et al. in [15] presented a data aggregation technique using multi-path approach, known as synopsis diffusion. Synopsis diffusion works in two phases: distribution of queries and data retrieval phase. During distribution of queries phase, a node sends a query in the network. The network nodes then form a set of rings around the M.Tech. - Thesis (July 2008) Indian Institute of Information Technology, Allahabad 14

24 querying node. The node which is i hop away from querying node is considered is ring R i. In the second phase, aggregation starts from outermost ring and propagate level by level towards the sink. Here a source node can have multiple paths towards sink. Figure 2.3 Examples of aggregation paths over a ring structure [7]. L. Gatani et al. in [5] describe a new strategy for data gathering in wireless sensor network that consider both issues: energy efficiency and robustness. Authors first say that single path to connect each node to the base station is simple & energy-saving approach but expose a high risk of disconnection due to node/link failures. But multi-path approach would require more nodes to participate with consequent waste of energy. Authors present a clever use of multi-path only when there is loss of packet which is implemented by smart caching of data at sensor nodes. Authors also argue that in many practical situation data may be gathered only from a particular region, so they use a different approach that relies on a spanning tree and provides alternative paths only when a malfunctioning is detected. Algorithm adopts a tree-based approach for forwarding packets through the network. In the ideal situation when no failures occur, this is certainly the best choice, as the minimum number of nodes is engaged in the transmission phase. In the presence of link or node failures, the algorithm will discover alternative M.Tech. - Thesis (July 2008) Indian Institute of Information Technology, Allahabad 15

25 paths, so as ensure the delivery of as many packets as possible within the time constraints. The problem with this approach is that it may cause the arising of hot spots and nodes along preferred paths will consume their energy resources quickly, possibly causing disconnection in the network. 2.4 Cluster-Based Approach We talked about hierarchical organization of the network in tree-based approach. Another scheme to organize the network in hierarchical manner is cluster-based approach. In cluster-based approach, whole network is divided in to several clusters. Each cluster has a cluster-head which is selected among cluster members. Cluster-heads do the role of aggregator which aggregate data received from cluster members locally and then transmit the result to sink. The advantages and disadvantages of the cluster-based approaches is very much similar to tree-based approaches. K. Dasgupta et al. in [16] proposed a maximum lifetime data aggregation (MLDA) algorithm which finds data gathering schedule provided location of sensors and base-station, data packet size, and energy of each sensor. A data gathering schedule specifies how data packet are collected from sensors and transmitted to base station for each round. A schedule can be thought of as a collection of aggregation trees. In [6], they proposed heuristic-greedy clustering-based MLDA based on MLDA algorithm. In this they partitioned the network in to cluster and referred each cluster as super-sensor. They then compute maximum lifetime schedule for the super-sensors and then use this schedule to construct aggregation trees for the sensors. W. Choi et al. in [1] present a two-phase clustering (TPC) scheme. Phase I of this scheme creates clusters with a cluster-head and each node within that cluster form a direct link with cluster-head. Phase I of this scheme is similar to various scheme used for clustering but differ in one way that the cluster-head rotation is localized and is done based on the remaining energy level of the sensor nodes which minimize time variance of sensors and this lead to energy saving from unnecessary cluster-head rotation. In phase II, M.Tech. - Thesis (July 2008) Indian Institute of Information Technology, Allahabad 16

26 each node within the cluster searches for a neighbor closer than cluster-head which is called data relay point and setup up a data relay link. Now the sensor nodes within a cluster either use direct link or data relay link to send their data to cluster head which is an energy efficient scheme. The data relay point aggregates data at forwarding time to another data relay point or cluster-head. In case of high network density, TPC phase II will setup unnecessary data relay link between neighbors as closely deployed sensor will sense same data and this lead to a waste of energy. Figure 2.4 Illustration of Two Phase Clustering [1]. H. Cam et al. in [4] present energy efficient and secure pattern based data aggregation protocol which is designed for clustered environment. In conventional method data is aggregated at cluster-head and cluster-head eliminate redundancy by checking the content of data. This protocol says that instead of sending raw data to cluster-head, the cluster members send corresponding pattern codes to cluster-head for data aggregation. If multiple nodes send the same pattern code then only one of them is finally selected for sending actual data to cluster-head. For pattern matching, authors present a pattern comparison algorithm. M.Tech. - Thesis (July 2008) Indian Institute of Information Technology, Allahabad 17

27 Figure 2.5 Data transmission using ESPDA [4]. 2.5 Simulation Tools We have plenty of simulation tools or simulators for simulating wireless networks. The simulators which are most popular are NS-2, OPNET, OMNet++, J-Sim, GlomoSim, Qualnet, TOSSIM and so on. Since wireless sensor networks are special type of wireless networks, most of the simulators available are not enough supported for simulating a wireless sensor network scenario. The literature shows that the simulators which are mostly used for wireless sensor network are NS-2, J-Sim, GlomoSim, OPNET, TOSSIM, PROWLER and even MATLAB is also used. NS-2 is the most popular and powerful simulator. NS-2 is an object-oriented discrete time event simulator and its modular design made it to be extensible. The detail of NS-2 is provided in appendix A. To simulate sensor network, there had been made attempt to put some add-ons. The most appreciable extension to NS-2 for wireless sensor networks was developed in 2004 by Ian Downward of Naval Research Laboratory (NRL) [18]. In this extension, they had created phenomenon packet which trigger event for sensor nodes. They also designed sensor agent, sensor application and other application to support wireless sensor network. But there has been no improvement or work done since This is the reason why all research is going by using NS-2 without any extension support. Other simulators like J-Sim, GlomoSim, OPNET, OMNet++ are M.Tech. - Thesis (July 2008) Indian Institute of Information Technology, Allahabad 18

28 exceptionally used by researchers. Another simulators which is MATLAB based, known as PROWLER, is sometimes used for simulating routing protocol or some MAC issues in wireless sensor network. MATLAB is also used for simulating some physical layer issues. But still there is no efficient simulator which is purely dedicated to wireless sensor network. More details for comparative study on simulator for wireless sensor network could be found in [19]. 2.6 Summary In-network data aggregation is the process of gathering data from all nodes and processes them at intermediate node while forwarding towards sink. Tree-based approaches construct a minimum spanning aggregation tree rooted at sink and covering all nodes in the network. In multi-path approaches a node choose more than one parent for forwarding data so as to have multiple paths towards sink to make the system robust. Cluster-based approaches organize the network in to several clusters each with a cluster-head which responsible for aggregating data locally before forwarding towards sink. Many simulators are used by researchers but NS-2 is the most efficient and widely used. M.Tech. - Thesis (July 2008) Indian Institute of Information Technology, Allahabad 19

29 Chapter 3 Energy-aware Balanced In-Network Aggregation After the detailed description of wireless sensor networks, problem of data aggregation for maximizing network lifetime, and existing aggregation techniques in chapter 1 and 2, we now present our proposed protocol for solving data aggregation problem. This chapter first gives you an introduction of our protocol Energy-aware Balanced In-Network Aggregation (E-BINA) and then briefly describes the system and energy model that has been considered for designed protocol. After this a detailed description of the algorithm will be presented. As there are some issues involved with every protocol, so with our proposed protocol also. These issues will be presented in brief and finally we summarize this chapter. 3.1 Introduction The three broad categories of data aggregation techniques that we have described in previous chapter are: tree-based approach, multi-path approach, and cluster-based approach. Focusing on cluster-based approach, we found that the existing protocols assume a sensor network which is divided in to several clusters. Depending upon the protocol operation, each cluster-head receives the data packets from some clustermember or from all cluster-member nodes directly and then cluster-head perform aggregation operation. Taking the advantageous features of tree-based approach, we have designed our protocol which takes the merits of both cluster-based and tree-based approach. E-BINA assumes a cluster-based wireless sensor network and applies tree-based approach inside each cluster. When a cluster is formed and cluster-head selected, it consider cluster-head M.Tech. - Thesis (July 2008) Indian Institute of Information Technology, Allahabad 20

30 as root and construct an aggregation tree over cluster-member nodes. The process of aggregation tree construction requires the sensor nodes to reduce their transmission power as sensor nodes now have to send their data packets to the neighbor node which is selected as parent. Energy consumed in wireless transmission is directly proportional to the square of the distance between nodes in communication [13]. Since cluster-member node now sends their data packets to the neighbor node instead of cluster-head, the transmission distance is reduced and hence the energy consumption of the sensor node. Likewise, overall energy consumption of sensor nodes in a cluster is reduced and so for the whole sensor network. Hence overall network lifetime will be increased. Energy-aware Balanced In-Network Aggregation (E-BINA) protocol is energyaware as it has taken the residual energy of sensor node in to consideration while constructing the aggregation tree. The protocol also balances the network load by selecting different parent for a node according to the energy level remain in the sensor node during the aggregation tree construction process. Each parent node performs aggregation of data packet that it receives from its child nodes and hence the protocol justifies the in-network aggregation concept. 3.2 System & Energy Model Consider a homogeneous network of n sensor nodes and a base station or sink node distributed over a region. The location of the sensors and the base station are fixed and known priori. Each sensor produces some information as it monitors its vicinity. We assume that the whole network is divided in to several clusters; each cluster has a clusterhead (CH). The clustering and the selection of cluster-head (CH) can be done by using any existing protocol like LEACH, such that cluster-head (CH) is maximum k-hop away from any node in cluster. We also assume that after the formation of cluster the transmission power of all nodes is adjusted in such a way that they can perform single hop broadcast. Single hop broadcast refers to the operation of sending a packet to all single-hop neighbors [8]. M.Tech. - Thesis (July 2008) Indian Institute of Information Technology, Allahabad 21

31 Our energy model for the sensors is based on the first order radio model described in [17]. A sensor consumes Eelec = 50nJ/bit to run the transmitter or receiver circuitry and Eamp = 100pJ/bit/m 2 for the transmitter amplifier. Thus, the energy consumed by a sensor i in receiving a l-bit data packet is given by, E Rxi = E elec. l (1) while the energy consumed in transmitting a data packet to sensor j is given by, E Txi,j = E elec. l + E amp. d 2 i,j. l (2) where di,j is the distance between nodes i and j. 3.3 Protocol Description In a cluster-based wireless sensor network, our algorithm is designed to provide energy-aware in-network data aggregation in a cluster. Each cluster uses this algorithm independently. In a cluster, the nodes can be categorized as: one cluster-head (CH) and other cluster member node. Function of cluster-head (CH) 1. Receive a query from base station. 2. Cluster-head (CH) sends configuration packets to all single-hop neighbors. 3. Receive data packets from all single hop neighbors. 4. Finally aggregate the data packets received and route it to base station. Function of cluster member 1. Receive configuration packets from neighbor nodes. 2. Update and forward configuration packets to all single-hop neighbors. 3. Receive data packets from neighbor nodes. 4. Aggregate all data packets by applying redundancy factor and send it to selected parent node. M.Tech. - Thesis (July 2008) Indian Institute of Information Technology, Allahabad 22

32 The algorithm works in two phases: Configuration packet flow and Data packet flow that are described below. Initial cluster position Flow of configuration packets Flow of data packets through selected parent Figure 3.1 A typical scenario of data aggregation in a cluster Configuration Packet Flow Initially cluster-head broadcast configuration packet to all its neighbors. Configuration packet contains the following fields: Node Id location of node that each node know in prior Hop Distance distance from cluster-head in terms of hop count (set zero for CH) Residual Energy current energy in node Each node upon receiving the broadcast configuration packet that is originated from cluster-head adds the sender of the packet in the list of its possible parents with its node id, hop distance, residual energy. After this the node again broadcast the configuration packet to all its neighbors by updating node id to its own id, incrementing hop distance by one and its own residual energy. This process continues until all the M.Tech. - Thesis (July 2008) Indian Institute of Information Technology, Allahabad 23

33 nodes in cluster receive configuration packet. All nodes that broadcast the configuration packet do so by predefined and common signal strength that is know to all the nodes Data Packet Flow When all nodes receives configuration packets, each node now select the parent to which it can forward the data packet. The parent selection procedure is shown in fig Each node looks in to the list of all its possible parents. The neighbor node which has least hop distance, ie closest to cluster-head, is selected as parent by a node. In case when two neighbor nodes have the least but equal hop distance, the node checks the residual energy of two neighbor nodes. The neighbor node that has greater residual energy is now selected as parent. In both the cases, node also calculate the difference of residual energy of two neighbor nodes, which have least hop distance, and checks whether this difference is less than the threshold or not. If it is then the node selects the parent as usual. But if it is not then the node selects other neighbor node as its parent. Define: E r [i]: residual energy of node i d h [i]: distance from CH in terms of hop count of node i E d ij: difference of residual energy of two nodes i & j t e : threshold of residual energy difference of two nodes i & j for balancing load nid: id of a node S: set of configuration packets received by node i ParentSelection(nid) 1 select two nodes j & k such that d h [j] & d h [k] is minimal in S 2 if (d h [j] < d h [k] AND E d jk < t e ) 3 then return nid of node j 4 else if (d h [j]==d h [k] AND E d jk < t e ) 5 then return nid of node with max(e r [j], E r [k]) 6 endif 7 else return nid of node k 8 endif Figure 3.2 Parent selection procedure. M.Tech. - Thesis (July 2008) Indian Institute of Information Technology, Allahabad 24

34 This allows a node that has more available resources to be selected as a parent node. This also balances the consumption of energy of nodes in the cluster and leads to die out of nodes nearly at same time. After selecting the parent node, each node now forwards its data to its parent. When a parent node receives multiple data packets from its neighbor nodes, it performs aggregation operation by eliminating redundancy in the data. Each parent node checks the equation below: V Ni V Nj < K (3) where, V Ni data value of node i V Nj data value of node j K redundancy factor If this equation satisfies, the parent node perform aggregation by applying any aggregation functions like MIN, MAX, and AVG on the values of data packet and send only one packet while discarding other packets. But if this equation do not satisfies, the parent performs aggregation by simply concatenating two data packet in to one keeping value of both packets intact. The selection of value for redundancy factor (K) has a trade-off between precision and energy consumption. If the application wants more precision, it should select a low value for redundancy factor otherwise a high value. Selecting high value for K means sending only one value thus less number of bits needs to be transmitted and hence low energy consumption. M.Tech. - Thesis (July 2008) Indian Institute of Information Technology, Allahabad 25

35 3.4 Issues E-BINA significantly reduces the energy consumption of all nodes in the cluster by reducing the transmission power of all nodes. The one issue that arises in our designed protocol is that after the formation of cluster and selection of cluster-head, all nodes have to reduce their transmission power. All nodes have to reduce their transmission power in such a way that they could only reach their single-hop distance neighbors. This operation requires some kind of synchronization among all nodes. The nodes have to program before to perform the above task. For this, the programming task needs little extra effort. Now when cluster-head received all data packets and aggregated them, it has to now increase its transmission power so that it can transmit the final aggregated data up in the cluster-head hierarchy towards the sink. Another issue that remains with any tree based approach is robustness of the system. In case of failure of any intermediate node in the tree hierarchy during operation will lead to the loss of data. Though E-BINA requires all nodes to adjust their transmission power after the deployment and requires extra effort for programming before, it conserves a significant amount of energy. So in the presence of the above issue, E-BINA outperforms when we try to maximize the network lifetime. 3.5 Summary E-BINA takes the advantageous features of cluster-based and tree-based approaches. Instead of sending data directly to cluster-head, nodes form an aggregation tree in each cluster. In aggregation tree, the selection of parent is based on two factors: hop distance and residual energy of node. Energy model is based on first order radio model. M.Tech. - Thesis (July 2008) Indian Institute of Information Technology, Allahabad 26

36 Protocol requires all nodes to adjust their transmission power. Protocol defines two types of packets: configuration packet and data packet. Each node eliminates redundancy in the data by satisfying equation (3). In-network aggregation is performed during data flow towards cluster-head. Issue: Adjustment of transmission power after deployment requires extra programming effort. M.Tech. - Thesis (July 2008) Indian Institute of Information Technology, Allahabad 27

37 Chapter 4 Performance Analysis In previous chapter we have explained our proposed protocol for data aggregation in cluster-based wireless sensor networks, called E-BINA. Now in this chapter we are going to analyses the performance of our protocol. We will show how our protocol outperforms in terms of energy efficiency in comparison to conventional protocol for data aggregation in cluster-based wireless sensor networks by presenting simulation analysis with different simulation parameters taken and results obtained. 4.1 Simulation Analysis Though many simulation tools are available for wireless sensor networks as discussed in chapter 2, we have chosen Network Simualtor-2 (NS-2) [20], in particular NS , as our tool to simulate the proposed protocol Simulation Setup A square field of 160m X 160m is taken where 11 nodes are randomly deployed. One node is designated as cluster-head (CH) and one node is designated data source. Command: set val(sc) "/root/desktop/mov1" M.Tech. Thesis (July 2008) Indian Institute of Information Technology, Allahabad 28

38 set val(cp) "" #setdest is found in "/root/ns-allinone-2.29/ns-2.29/indeputils/cmu-#scen-gen/setdest" exec./setdest -n 10 -M 0.1 -p 21 -x 160 -y 160 -t 20 > mov1 & puts "loadin RANDom sceanrio" source $val(sc) The snapshot of node scenario in NAM is shown below. The three colors of node show the energy levels of nodes. The initial color of nodes is green. When energy drop to first threshold level the color turns to yellow and when drop to second threshold level the color turns to red. After this level a node is dead and color is red. Figure 4.1 Random nodes scenario in NAM. M.Tech. Thesis (July 2008) Indian Institute of Information Technology, Allahabad 29

39 Energy model is ON. Transmit power, Receive power, Idle power, Sleep power, Transition power, and Initial Energy of nodes is set accordingly. Also transmission range is set by controlling the transmit power and receiving threshold of antenna of nodes. All other parameters are taken default values. Command: set val(energymodel) EnergyModel ;# energy model is on set val(initialenergy) 1 ;# Initial energy in Joules set val(sleeppower) 0.0 ;# sleep power in Watt set val(tp) ;# transition power consumption(watt) in state transition from sleep to idle (active) set val(tt) ;# transition time(second) use instate transition from sleep to idle (active) set val(ip) ;# idle power M.Tech. Thesis (July 2008) Indian Institute of Information Technology, Allahabad 30

40 Transmit power, Receive power, transmit power & receiving threshold of antenna of nodes are set with different values to use different transmission range of nodes and show the comparison between E-BINA and conventional protocol. Case 1: E-BINA set val(rxpower) ;# receive power in Watt set val(txpower) 0.66 ;# transmit power in Watt Phy/WirelessPhy set Pt_ e-4 ;# 40m Phy/WirelessPhy set RXThresh_ e-10 Case 2: Conventional protocol set val(rxpower) 1.0 ;# receieve power in Watt set val(txpower) 2.0 ;# transmit power in Watt Phy/WirelessPhy set Pt_ 7.214e-3 ;# 100m Phy/WirelessPhy set RXThresh_ e-10 Other values that could be used are: #Phy/WirelessPhy set Pt_ e-3 #Phy/WirelessPhy set Pt_ ;# Transmission range 50m, ;# 250m The value of RXThresh_ is obtained by executing threshold.cc defined in "/root/nsallinone-2.29/ns-2.29/indep-utils/propogation. Snapshot is given below: M.Tech. Thesis (July 2008) Indian Institute of Information Technology, Allahabad 31

41 Figure 4.2 Snapshot of console Some other parameters used are: Channel Type Wireless channel Propagation Model Two Ray Ground MAC Type Network Interface Type Phy/WirelessPhy Interface Queue Type Queue/DropTail/PriQueue Antenna Model Antenna/OmniAntenna Routing Protocol AODV Simulation Time 20 sec Parameters set for data transfer are: Cluster-head Source node Traffic Type Packet Size node 10 with UDP agent attached node 0 with UDP agent attached CBR with a rate of 5 packets / second 136 bytes M.Tech. Thesis (July 2008) Indian Institute of Information Technology, Allahabad 32

42 4.1.2 Simulation Run Case 1: E-BINA For E-BINA we set transmission range of 40m such that a node sends its data to its single-hop neighbor and data is forwarded in a multi-hop fashion. Figure 4.3 shows the data transfer between node 0 and node 10. Node 1 and 2 are relay nodes. Since the transmission range is set to 40m, node 0 can only send its data to node 1 and so other nodes. Figure 4.3 Data transfer between node 0 and 10 through node 1 & 2. M.Tech. Thesis (July 2008) Indian Institute of Information Technology, Allahabad 33

43 Figure 4.4 shows that after some time energy level of node 2, 1, 0, and 10 dropped to first threshold level and hence the color of nodes turns to yellow. Figure 4.4 Energy level of nodes drop to first threshold. Figure 4.5 shows that after some more time energy level of node 1, 2, 0, and 10 dropped to second threshold level and hence the color of nodes turns to red. M.Tech. Thesis (July 2008) Indian Institute of Information Technology, Allahabad 34

44 Figure 4.5 Energy level of nodes drop to second threshold. Case 2: Conventional Protocol In conventional method, all nodes in a cluster send their data directly to cluster-head. For this reason we set transmission range of nodes to be 100m so that source node 0 can send data directly to node 10. To able to have a large transmission range the transmitting and receiving power of nodes are more than double of as in case 1. The data transfer start between node 0 and node 10 directly. As happened in last case, again after some time energy level of node 0 and node 10 decreased to first threshold level and color of nodes change from green to yellow and then energy level of both nodes go down to second threshold level and nodes turn to red. M.Tech. Thesis (July 2008) Indian Institute of Information Technology, Allahabad 35

45 4.1.3 Simulation Results A. Conserving Energy We determine residual energy of the source node, which is defined as the remaining energy of a node and considered that as the metric to prove energy efficiency of our proposed protocol. We used this metric to show the impact of transmission power on energy reduction. Figure 4.9 shows the significant reduction in energy consumption by using E-BINA when compared with conventional protocol. This shows the benefit of sending data in a multi-hop fashion towards cluster-head. E-BINA Conventional protocol Figure 4.6 Residual energy of source as a function of time. M.Tech. Thesis (July 2008) Indian Institute of Information Technology, Allahabad 36

46 B. Throughput We have also measured the throughput of the receiving node i.e. cluster-head node 10 in our scenario for both the cases. Throughput of a node is defined as the average rate of successful message delivery over a communication channel. Figure 4.10 show that E- BINA achieves high throughput in comparison with conventional protocol. E-BINA Conventional protocol Figure 4.7 Throughput as function of time. M.Tech. Thesis (July 2008) Indian Institute of Information Technology, Allahabad 37