Efficient Data Forwarding Techniques in Wireless ... - IEEE Xplore

Efficient Data Forwarding Techniques in Wireless Sensor Networks Zeenat Rehena

Sarbani Roy

Deptt. of CSE, Aliah University Salt Lake, Kolkata, India [email protected],

Abstract— Energy consumption is the key design criterion for routing data in WSN. However, some applications of WSN like disaster management, battlefield control etc. demand fast data delivery. In this paper, three data forwarding techniques are proposed. Here, the source nodes or intermediate nodes select a next node to forward the data to the destination based on different criteria. The process repeats until data reach the destination. In the first technique, neighbor node nearer to the sink is chosen considering its distance from sink node as a criterion. In the second technique remaining energy of the neighboring nodes is used as a criterion to select the next node. The combination of these two criteria is considered in the third technique based on Multiple Criteria Decision Analysis. A comparative study of the performance of these techniques has been carried out and results are presented in this paper. Accommodate Keywords—Data Forwarding; Analysis; Wireless Sensor Network

I.

Nandini Mukherjee

Deptt. of CSE Jadavpur University, Jadavpur, Kolkata, India [email protected],

Multiple-Criteria

Decision

INTRODUCTION

With the advancement of wireless communication technologies, Wireless Sensor Networks (WSNs) have become useful for many applications. In particular, the monitoring applications covering large geographical areas are much benefited by the use of WSNs. However, in large-scale WSNs, data may need to follow a long multi-hop communication path and each intermediate node has to receive and transmit the packets. The residual energy of the intermediate nodes reduces due to such communication activities. Moreover, overall energy consumption for packet routing depends on the number of hops in the path. Therefore, it is important that the network must be designed in such a way that the average number of hops for routing the data from source to sink remains as minimum as possible. For many applications, timeliness of data delivery to the sink is also one of the essential requirements. An example of such applications is Disaster management. Although data transmission is fast when flooding routing protocol is used, energy consumption due to this protocol is high. In this paper we focus on designing routing algorithms for a partitioned network for fast and energy efficient data delivery. It is assumed that a sink is placed in every partition and all nodes in the partition know the location of the sink. The

c 978-1-4673-4529-3/12/$31.00 2012 IEEE

Deptt. of CSE Jadavpur University Jadavpur, Kolkata, India [email protected]

paper discusses three algorithms in particular and compares their effectiveness. First we propose a routing algorithm for forwarding data from source to sink by selecting a next node based on its distance from the sink. As the next node which is nearer to the sink is selected, minimum energy is consumed in this case. But, it will tend to use the same routing path again and again and thus the energy of the nodes on this path reduces quickly resulting route hole problem [12]. So, another data forwarding mechanism based on maximum remaining energy of each node is proposed. Here, the objective is to select the next node on the basis of its remaining energy such that the energy usage is equally distributed among the sensor nodes. Finally a third data forwarding mechanism is proposed which is based on multiple criteria analysis. In this mechanism two criteria, i.e. distances from the sink node and remaining energy of nodes are considered. A comparative study of these three mechanisms has been made. The results of this study are also presented. The rest of the paper is organized as follows. Section II provides a review of the related works. Section III, Section IV and Section V present the three algorithms, i.e. minimum distance based next node selection, maximum remaining energy based next node selection, and multiple criteria based data forwarding respectively. Section VI presents the simulation results of comparing these three algorithms and analyzes the results. Multiple sensing nodes for single event occurrence and hop counts required for these three algorithms when compared with restricted flooding algorithm are discussed in section VII. Section VIII concludes the paper. II.

RELATED WORK

Extensive research work has been carried out on data routing techniques. In [7], the authors proposed Multiple Sink Dynamic Destination Geographic Routing (MSDDGR). It is based on greedy forwarding scheme. When a packet needs to be sent, the sender selects the nearest sink as the current destination. Also if the intermediate node sees another sink is nearer to it, then the current destination node is changed and the new sink node is selected as the destination. The authors in [8] address the problem of efficient routing the data from multiple sources to multiple sinks. The authors first define a mathematical model to derive an optimal solution. In the second phase, it is assumed that the initial state of the system is

449

such that a tree exists for each sink connecting it to all the relevant sources. The path from source to sink is changed periodically adapting the different sink-rooted trees. The adaptation consists of selecting a different neighbor as the parent towards a given sink. In [13], the authors propose a protocol called MRMS (Multipath Routing in large scale sensor networks with Multiple Sink nodes) which incorporates multiple sink nodes. In MRMS, a primary path is created with minimum path cost. It also saves the other paths from different sinks. Thus, when the primary path is not reachable or if the residual energy of the sensors along the path falls below a certain threshold, another path is selected. A scalable multipath routing approach called Neighbor Sink Nexus (NSN) routing algorithm is presented in [6]. It is based on Set Cover problem. This paper claims that set cover method gives balanced and better performance in terms of energy efficiency and reduced latency for multiple sink WSNs. Malakooti et al. in [2] propose a Distributed Composite Multiple Criteria Routing approach. In this work two mechanisms, such as global multiple criteria routing and distributed multiple criteria routing are discussed. The multiple criteria like energy, delay and bit error rate are applied to the Distance Vector routing protocol to find the best possible path from any source to destination. Li et al. in [9] propose a datacentric multi-criteria routing algorithm called MCR. Here criteria like energy remaining at the sensor nodes, power consumption model, and group membership of each node are considered. Tabatabaei in [10] also uses the multiple criteria approach for mobile ad hoc networks with AODV routing protocol. In the aforementioned works the authors do not consider a partitioned network topology. Instead, the source nodes send data to the nearest sink(s) or save multiple paths and choose the optimal one. Based on the requirements of the routing mechanisms, grouping or clustering of nodes is performed in the network. On the other hand, the particular focus of the research work presented in this paper is on proposing routing mechanisms for a partitioned multiple-sink network. The partitioning algorithms are mentioned in [14]-[15] and in this paper only routing mechanisms are discussed. Further, the application area considered for our work is disaster management and data transmission time needs to be minimized for such application. Therefore, the multi-criteria routing algorithm proposed in this paper considers remaining energy of a sensor node and distance of the node from sink so that the distribution of data traffic is balanced and less number of hops are needed for data transmission. III.

DATA ROUTING BASED ON MINIMUM DISTANCE

In this section the first routing algorithm based on minimum distance is discussed. The goal of the algorithm, called NNMD (Next Node with Minimum Distance), is to minimize the number of hops a packet needs to travel within the network to reach a sink node. The following assumptions are made for the proposed scheme: 1) All nodes know their location information, the id and location of the sink in that partition

When a source node requires disseminating its sensed data to the sink, it invokes the NNMD routing algorithm. The route is created on the fly and data is transmitted. The algorithm works in two phases. 1) Generation of Eligible Neighbor Set, and 2) Selection of Next Forwarding Node. The phases are described as follows: A. Generation of Eligible Neighbor Set

S = {s1 , s 2 ,..., s n }is the set of sensor nodes and SI = {SI 1 , SI 2 ,..., SI m }is the set of sinks. NNs k Let us consider

is the list containing 1-hop neighbors of any sensor node s k ∈ S . ENSs k is the eligible neighbor set of s k from which the next forwarding node is to be chosen i.e. ENSs k ⊆ NNs k . Also SI m (where SI m ∈ SI ) is the sink in the partition which contains

s k (where s k ∈ S ).

Whenever a source node senses any critical data (event) from the environment, it first creates its own eligible neighbor set ( ENSs k ). To do this, a source s k broadcasts a Request packet which is received by its 1-hop neighbor nodes. The structure of the Request packet is presented in Fig. 1 where s _ id and d _ id represent id of the source node and id of the destination node (i.e. the sink node). The dist field represents Euclidean distance between the source and destination node. Upon receiving the Request packet, the neighbor nodes of s k calculate their own distances from the sink node as each node knows own and corresponding sink location. Then they compare their distances with the distance between the source and destination (sink), i.e. the value of dist in the Request packet. If any neighbor node finds that its distance is less than the value of dist , it sends back a Reply packet with its calculated distance to the source node s k . Fig. 2 shows the structure of the Reply packet. Here own _ id represents the id of the neighbor node which sends the Reply packet to the source node and new _ dist is the calculated distance value to the sink from the neighbor node. The eligible neighbor set ( ENSs k ) of node all the nodes from

NNs k those sent Reply packets to the

source s k . The data packets are then only forwarded to a node which belongs to ENSs k . Selection of this node is discussed in the next section. If the ENSs k is empty, i.e. there are no neighbor nodes closer to the destination or the 1-hop neighbor nodes are dead, the data packets are dropped.

Fig. 1. Request Packet Structure

2) Each node maintains a list of 1-hop neighbor nodes

450

s k consists of

2013 3rd IEEE International Advance Computing Conference (IACC)

Fig. 2. Reply Packet Structure

B. Selection of Next Forwarding Node

neighbor nodes which are closer to the destination or the neighbor nodes have become dead due to energy depletion. One such situation is explained in Fig. 4. Here node S1 has data and node S3 is a dead node. At this point when node S1 wants to transmit its data to sink, it finds an empty ENS because no other nodes are there to forward the data to the sink.

After getting ENSs k , the source node selects a next forwarding node s i as follows:

min {dist (s i , SI m ) : s i ∈ ENS (s k , SI m ); s i ≠ s k } Now the node

s i becomes the new forwarding node.

These two phases are repeated until the data packets reach the sink node. Fig. 3 shows the NNMD routing steps. Initially in Fig. 3(a), node a has data and it broadcasts a Request packet. All the nodes within its vicinity listen to the Request packet. In Fig. 3(b), the nodes which are nearer to the sink (i.e. nodes e and d) reply back to the source node a. Node a first builds its eligible neighbor set with e and d, selects node d as the next forwarding node as node d has minimum distance to sink (Fig. 3(c)). Now node d receives data for transmission. Node d broadcasts the Request packet as shown in Fig. 3(d). The process repeats until the data reach the sink. The remaining steps are shown in Fig. 3(e, f, g, h).

Fig. 4.

Occurrence of Route hole Problem

IV.

DATA ROUTING BASED ON MAXIMUM REMAINING ENERGY It can easily be observed that NNMD algorithm reduces overall energy consumption within the network. But, there is possibility of selecting the same path repeatedly for the same source node. Therefore, energy consumption of the nodes on that path increases. To balance the energy consumption among the nodes we propose to consider the residual energy of the nodes while selecting the next forwarding node. In this section we describe Next Node with Maximum Remaining Energy (NNMRE) Routing protocol. In NNMD, the node which is closest to the sink is chosen as the next forwarding node. But in NNMRE the node is chosen if it has maximum remaining energy among the neighboring nodes. Like the previous one, this algorithm also has two phases: 1) Generation of Eligible Neighbor Set, and 2) Selection of Next Forwarding Node. The first phase is same for both algorithms. Only difference is in the selection phase for the next forwarding node.

Fig. 3. NNMD Route Discovery Phases for Data Forwarding: (a) node a broadcasts request packet; (b) node e & d are nearer to sink and send reply packet; (c) node a chooses node d as next forwarding node and sends data packet; (d)-(e)-(f) repeat steps (a)-(b)-(c); (h) shows final route from source node a to sink.

C. Route Hole Problem A routing hole is created in a region within a sensor network where either nodes are not available or the available nodes cannot participate in actual routing of data due to various possible reasons. The NNMD algorithm suffers from the route hole problem because it may return an empty ENS for any sensor node in the network. An empty ENS is returned when a node has no

In the first phase, in addition to the neighbors’ distances from the sink node, the neighbor nodes are also asked to send their remaining energy value. The structure of the Reply packet for this algorithm is depicted in Fig. 5. An extra field, rem_energy, storing the remaining energy of the node, is added to the Reply packet.

Fig. 5. Reply Packet Structure

ENSs k in the similar manner, the source node selects a next forwarding node s i as follows: After creating

max{rem _ energy(si , SI m ) : si ∈ ENS(sk , SI m ); si ≠ sk } If more than one node has equal maximum remaining energy

2013 3rd IEEE International Advance Computing Conference (IACC)

451

then one of them which is closest to source node is chosen. By choosing a node closest to the source node, the energy consumption for data transmission can be minimized. V.

MULTIPLE CRITERIA DECISION BASED ROUTING

ii) Measure all criteria on similar numerical scales and normalize each criterion using equation (2) iii) Assign weights of importance to the criteria and calculate utility for each alternative using the equation (1) iv) Rank alternatives in descending order of U

A. MCDA Method In NNMD, our contribution is to minimize the number of hops required to reach the sink node. Thus the end-to-end delay is reduced and data flow very fast. On the other hand, in NNMRE we consider residual energy of the node, thereby reducing the possibility of occurrence of energy hole problem. However, in this approach, more hops may be necessary to reach data to the sink. Thus delay for delivering data to the sink may increase and also energy consumption for data transmission may increase for use of longer route. Our objective is to reduce the delay, as well as distribute the data traffic among the sensors so that longevity of the network can be increased. With this objective, our third algorithm is based on Multiple Criteria Decision Analysis (MCDA) (Malakooti et al., 2006; Li et al., 2005 and Tabatabaei, 2009). The MCDA method makes the decision from various alternatives. These alternatives are ranked on the basis of different criteria used in the MCDA method. It is usually specified in terms of normalized weights for each criterion, as well as normalized scores for all options or alternatives relative to each of the criteria. Let us consider that the MCDA has a set of n alternatives with m criteria and the utility U for each alternative Oi is defined as follows: m

U (Oi ) = ¦ Z k (Oi ) ∗ Wk k =1

for i = 1,2, , n

(1)

Where Z k (Oi ) is the normalized score of alternative Oi under criterion

C k and W k is weight of importance for the

criterion C k .

Z k (Oi ) is calculated according to the proportional scoring method as follows: Z k (Oi ) = (Oik − Ok min ) / (Ok max − Ok min )

(2)

and the absolute value is taken. The following steps are used to calculate the utility function

U for each alternative Oi

alternative with the highest

(Oi ) . The

U is the best alternative.

B. Routing Mechanism Our proposed MCDA based routing algorithm known as MCDR (Multiple Criteria Decision Routing) considers the following two criteria: a. Remaining energy of a node, and b. Distance of the node from sink We use the same scale to measure both the criteria, i.e. a 0 to 100 scale. In this paper we consider equal weight value for each criterion, i.e. 50% for remaining energy and 50% for distance from sink. These two steps are explained below. 1) Measurement of all criteria on similar numerical scales: a) Remaining Energy of a node: It is assumed that the worst value of remaining energy of a node is 0.0005 J (for one receive and one transmit according to the First order radio model [11] and best or maximum remaining energy of a node is 0.5 J (here initial energy of a node). b) Distance of the node from sink: When the Euclidean distances from sink to the nodes (alternatives) are measured, maximum value is considered to be the distance from a source node to the sink node within its partition. This maximum value is taken as the worst value. The best value is considered to be 0 (unit is m). Thus if a node is far from the sink, it will get less priority and if it is closer to the sink, it will get higher priority. In order to obtain the scores for other intermediate values (both for remaining energy and distance), we use the proportional scoring technique as mentioned earlier. The following example describes the MCDR algorithm. Fig. 6 shows that node 1 (source node) has three neighbor nodes and these nodes comprise the eligible neighbor set of node 1. In this example the maximum distance from source to sink is assumed to be 70 m (i.e. from node 1 to sink). Table I shows the distances from the sink and remaining energy values of nodes 2, 3 and 4.

.

i) For each criterion, find the range of criterion values:

Ok min = min{Oik i = 1,2,..., n} for each criterion C k , k = 1,2,..., m and

Fig. 6. Example

Ok max = max{Oik i = 1,2,..., n} for each criterion C k , k = 1,2,..., m

TABLE I. Alternatives Criteria

452

VALUES AND CRITERIA Node-2

2013 3rd IEEE International Advance Computing Conference (IACC)

Node-3

Node-4

C1(distance in m)

52

50

55

C2 (remaining energy in J)

0.4

0.45

0.42

Above values are normalized between 0 and 100. The normalized values (scores) obtained in this way which are shown in Table II. TABLE II.

NORMALIZED VALUE

Alternatives Criteria C1 C2

Node-2

Node-3

Node-4

25.71 79.97

28.57 89.98

21.42 83.98

nodes are placed at each corner of the region. For the experiment purpose we ignore the effect of signal collision and interference due to the wireless channel. A. Simulation Environment A 100 node field is used and generated by randomly placing the nodes in a 200 m x 200 m square area. We assume that the area contains homogeneous sensor nodes with a communication range of 45m. The first order radio model [11] is used here. The simulation parameters used are shown in Table IV. TABLE IV.

2) Weighted values of the criteria: The utility U for each alternative Oi (nodes 2, 3 and 4 in the example) is calculated using the following equation:

Parameters

m

U (Oi ) = ¦ Z k (Oi ) ∗ Wk k =1

Here,

U (O2 ) = Z1 (O2 ) ∗ W1 + Z 2 (O2 ) ∗ W2 U (O3 ) = Z 1 (O3 ) ∗ W1 + Z 2 (O3 ) ∗ W2 and

U (O4 ) = Z1 (O4 ) ∗ W1 + Z 2 (O4 ) ∗ W2 The utility for each alternative is shown in Table III and node 3 is chosen as the next forwarding node of source node 1. C. Algorithm The routing algorithm is described in this section. This algorithm is run by each node (either the source or any intermediate node) on a path that has some data to disseminate. The first step is similar to the previous two algorithms. Selection of the next node is made using the two criteria and the values of the neighbor nodes received by the node which is running the algorithm. After receiving the values, the node calculates the utility value for each alternative (nodes in the eligible neighbor set) and chooses the best node among the alternatives which has the highest utility. TABLE III.

UTILITY VALUE

Alternatives Node-2

Node-3

Node-4

Criteria C1

12.85

14.28

10.71

C2

39.98

44.99

41.99

Utility U(Oi)

52.83

59.27

52.7

VI.

SIMULATION ENVIRONMENT AND RESULTS

A comparative study of the proposed routing algorithms and the restricted flooding routing protocol has been made in a simulation environment. Simulations have been done using Matlab. A square region is considered for deployment of the wireless sensor network. Sensor nodes are deployed randomly and all sensor nodes have the same capacity. The entire network is partitioned into four smaller sub-networks. Sink

SIMULATION PARAMETERS Values

Transmit and Receive Electronics

50 nJ/bit

Transmit Amplifier

100 pJ/bit/m2

Data Message

4000 bits

Control Message

1000 bits

Initial Node Energy

0.5 J

Communication Range

45 m

B. Performance Metrics Performance metrics as described in this section are used to evaluate the performances of the proposed routing algorithms and to compare them with the restricted flooding routing protocol. The restricted flooding routing protocol [14] is taken as the lowest bound. In this paper, we also present a term round which is defined below: Round: A single event (a packet) reaches sink(s) either from single source node or from multiple source nodes. Following performance metrics are used for evaluation: •

Energy consumption: Total energy consumption for transmitting a packet to sink from source nodes successfully.

•

Rounds before first node failure: Number of rounds before the first node dies.

•

Number of dead nodes: Number of dead nodes increases as number of rounds increases. This metric provides an indication of the expected lifetime of the sensor network.

C. Simulation Results First the NNMD algorithm is compared with restricted flooding in multi-sink partitioned network. Fig. 7 and Fig. 8 show the simulation results of these routing algorithms respectively. The point marked with ‘*’ is the event occurrence place within the network. When flooding is used, every node is engaged in transmitting and receiving data packets. On the other hand, in NNMD only some selected nodes on a path are used to transmit data from source node to sink. Fig. 9 and Fig. 10 show data routing for multiple event occurrence in case of NNMD and NNMRE respectively. When two events occur at different times at the same place within the monitoring region

2013 3rd IEEE International Advance Computing Conference (IACC)

453

and sensed by a single source node (event occurrence is marked by '*' symbol), the data is routed towards the sink with which the source node is associated. As expected in NNMD, the same route is used for these two events. While in case of NNMRE, two different paths are used for routing data. Similarly, in case of MCDR different paths are followed for two events to reach the data to the sink. This situation is depicted in Fig.11. D. Performance Results First we consider metric 1. Fig.12 shows energy consumption for routing a single data packet to the sink. From Fig. 12 it is clear that the energy consumption in case of each of the proposed routing algorithms is less compared to Flooding routing algorithm. Among them, lowest energy is consumed for NNMD. Since NNMD needs less number of hops to reach sink less energy is consumed here. On the other hand almost same amount of energy is consumed in case of MCDR. Energy consumption in case of NNMRE is higher than NNMD and MCDR. Fig. 13 depicts the number of rounds needed in each case before the first node dies. For getting this result random occurrences of events are considered. As expected MCDR performs better here since it considers residual energy of the node as well as minimum distance of the node from the sink. NNMD performs better compared to NNMRE, but not as good as MCDR. Fig. 14 also shows number of rounds in each case until the first node dies. However, here same source node is considered instead of random occurrence of events. From Fig. 14 it is clear that the number of rounds is almost same in case of NNMRE as well as MCDR and both perform better than the other two. Since data routing depends on the remaining energy of the nodes in case of these two algorithms, data traffic is distributed among the nodes which increases the network lifetime.

Fig. 8. Data routing using Flooding

Fig. 9. Same Routes from source node to sink node in NNMD

Fig. 10. Different Routes from source node to sink node in NNMRE

Fig. 7. Data routing using NNMD

Fig. 11. Different Routes from source node to sink node in MCDR

454

2013 3rd IEEE International Advance Computing Conference (IACC)

and Max-Hop) is shown on the vertical axis. For each event occurrence, NNMD and MCDR need almost equal number of Min-Hop and Max-Hop and these are low compared to NNMRE. A. Discussion Analyzing all the above mentioned performance metrics it is concluded that MCDR shows better performance for achieving longer lifetime. Also number of hops (both Max-Hop and Min-Hop) needed in case of NNMD and MCDR algorithms are low. Therefore, MCDR is most suitable for applications like disaster management where longer lifetime of WSN and fast data delivery are two desirable criteria. Fig. 12. Average Energy Consumption for packet transmission

Fig. 14.

Round until First Node Die while using the same source nodes

Fig. 13. Round until First Node Die

Fig. 15 depicts network lifetime in terms of last node to die. The proposed routing algorithms outperform the Flooding routing algorithm. MCDR has the longest network lifetime which is close to 8000 rounds. While NNMD and NNMRE algorithms have almost same network lifetime. VII. EVENT OCCURRENCE AT PARTITION BOUNDARIES Because of the random deployment of the sensor nodes and sometime for dense deployment of the nodes, it may happen that an event is sensed by multiple nodes. Since multiple sensing nodes may be placed in different partitions, the sensed data will be forwarded to multiple sinks through different paths. Simulation results for the four algorithms in such situation are shown in the Fig.16, Fig.17, Fig.18 and Fig.19 for NNMD, NNMRE, MCDR and restricted flooding respectively.

Fig. 15. Alive sensor nodes vs. Rounds

An event occurrence is marked with '*'. It is clear from the figures that a packet may need unequal number of hops to reach the sinks. In the above situation, performance of the algorithms is evaluated by finding the number of hops needed for a packet to reach the sink. Two variations of hop counts are considered. a) Min-Hop: Minimum number of hops needed for an event to reach the sink, and b) Max-Hop: Maximum number of hops needed for an event to reach the sink. Fig. 20 depicts the Min-Hop and Max-Hop for an event to reach the sink. Number of events is shown on the horizontal axis, while number of hops (Min-Hop

Fig. 16. Multiple paths from multiple source nodes in NNMD

2013 3rd IEEE International Advance Computing Conference (IACC)

455

Fig. 17. Multiple paths from multiple source nodes in NNMRE

Fig. 20. Number of hops vs. Events

REFERENCES [1]

Fig. 18. Multiple paths from multiple source nodes in MCDR

VIII. CONCLUSION Three algorithms for data forwarding in a partitioned wireless sensor network are proposed here. It has been demonstrated that the proposed algorithms show better performances over flooding routing protocol and using multiple criteria to choose the next node is the best choice in terms of increase in the life time of the network. Future work will focus on including additional criteria, fine-tuning of the weight values etc. The route hole problem discussed in this paper will also be handled in future.

Fig. 19. Multiple paths from multiple source nodes in Flooding

456

B. Elbhiri, S.EI. Fkihi, R. Saadane, D. Aboutajdine, Clustering in Wireless Sensor Networks based on Near Optimal Bi-partitions, in: Procs. of the Next Generation Internet (NGI), 2010. [2] B. Malakooti, I. Thomas, A Distributed Composite Multiple Criteria Routing Using Distance Vector, in: Procs. of the IEEE International Conference on Networking, Sensing and Control, Ft. Lauderdale, FL, 2006. [3] D. Das, Z. Rehena, S. Roy, N. Mukherjee, Multiple-Sink Placement Strategies in Wireless Sensor Network, in: Procs. of the Fifth International conference on Communication Systems and Networks (COMSNETS 2013), Bangalore, India, 2013. [4] E.I. Oyman, C. Ersoy, Multiple Sink Network Design Problem in Large Scale Wireless Sensor Networks, in: Procs. of the International Conference on Communications (ICC), Paris, France, 2004. [5] I. Slama, B. Jouaber, D. Zeghlache, Energy Efficient Scheme for Large Scale Wireless Sensor Networks with Multiple Sinks, in: Procs. of the IEEE Wireless Communications and Networking Conference, WCNC, Las Vegas, USA, 2008. [6] K.N. Chaaran, M. Younus, M.Y. Javed, NSN based Multi-Sink Minimum Delay Energy Efficient Routing in Wireless Sensor Networks, European Journal of Scientific Research, 41, 2010, 399-411. [7] L. Cao, C. Xu, W. Shao, Multiple Sink Dynamic Destination Geographic Routing in Wireless Sensor Networks, in: Procs. of the International Conference on Cyber-Enabled Distributed Computing and Knowledge Discovery (CyberC), Huangshan, 2010. [8] P. Ciciriello, L. Mottola, G.P. Picco, Efficient Routing from Multiple Sources to Multiple Sinks in Wireless Sensor Networks, in: Procs. of the 4th European Conference on Wireless Sensor Networks (EWSN), 2007. [9] Q. Li, J. Beaver, A. Amer, P.K. Chrysanthis, Multi-Criteria Routing in Wireless Sensor-Based Pervasive Environments, Journal of Pervasive Computing and Communication, 1 (2005). [10] S. Tabatabaei, Multiple Criteria Routing Algorithms to Increase Durability Path in Mobile Ad hoc Networks, in: Procs. of the 4th International Conference for Internet Technology and Secured Transactions, ICITST, London, UK, 2009. [11] W.R. Heinzelman, A. Chandrakasan, H. Balakrishnan, EnergyEfficient Communication Protocol for Wireless Micro Sensor Networks, in: Procs. of the 33rd Hawaii International Conference on System Sciences, Hawaii, 2000.

2013 3rd IEEE International Advance Computing Conference (IACC)

[12] X. Wu, G. Chen, S.K. Das, Avoiding Energy Holes in Wireless Sensor Networks with Non uniform Node Distribution, IEEE Transactions on Parallel and Distributed Systems, 19, (2008). [13] Y. Chen, E. Chan, S. Han, Energy Efficient Multipath Routing in Large Scale Sensor Networks with Multiple Sink Nodes, in: Procs. of APPT, 2005, 390-399.

[14] Z. Rehena, S. Roy, N. Mukherjee, Topology Partitioning in wireless Sensor Networks using Multiple sinks, in: Procs. of the 14th International conference on ICCIT, Dhaka, Bangladesh, 2011. [15] Z. Rehena, D. Das, S. Roy, N. Mukherjee, A Comparative Study of Partitioning Algorithms for Wireless Sensor Networks, in: Procs. of the 3rd International conference on Wireless and Mobile Networks, Bangalore, India, 2012.

2013 3rd IEEE International Advance Computing Conference (IACC)

457