Clustering-Biased Random Algorithm for Load ... - Semantic Scholar

18

Int. J. Intelligent Systems Technologies and Applications, Vol. 12, No. 1, 2013

Clustering-Biased Random Algorithm for Load Balancing (C-BRALB) in Wireless Sensor Networks Barra Touray*, Jie Lau and P. Johnson Faculty of Technology and Maritime Operations, Department of Engineering, University of Liverpool John Moores University, Byrom Street L3 3AF, UK E-mail: [email protected] E-mail: [email protected] E-mail: [email protected] *Corresponding author Abstract: A Wireless Sensor Network (WSN) consists of large collection of minute nodes organised into a cooperative network used for gathering data in diverse environments. The data collected is transmitted by the sensors to the Sink with the help of a routing algorithm. The energy consumption is a key design criterion for WSN routing algorithms. In this paper Clustering-Biased Random Algorithm for Load Balancing (C-BRALB) in WSNs is proposed. The clustering technique used in C-BRALB is adopted from the clustering technique used in Improved Directed Diffusion (IDD) while the routing mechanism is based on energy biased random walk. It is shown in this paper using simulation that C-BRALB is energy efficient, scalable and can load balance the traffic among nodes. Keywords: biased random walk; routing algorithm; clustering; WSN; wireless sensor network. Reference to this paper should be made as follows: Touray, B., Lau, J. and Johnson, P. (2013) ‘Clustering-Biased Random Algorithm for Load Balancing (C-BRALB) in Wireless Sensor Networks’, Int. J. Intelligent Systems Technologies and Applications, Vol. 12, No. 1, pp.18–27. Biographical notes: Barra Touray is currently a commonwealth scholar doing a PhD in the School of Engineering Technology and Maritime Operations at Liverpool John Moores University. He gained a BEng (Honours) in Networks and Telecommunication Engineering from the same University in 2008. He is also Cisco certified in CCIE (r&s) (written), CCNP (r&s), CCIP (sp) and CCNA (r&s), with more than five years experience as a Cisco network Engineer in an ISP. His research interests include energy efficient routing algorithm for wireless sensor networks and Ad hoc networks. Jie Lau has completed the MSc Telecommunication Engineering Degree with a first class at Liverpool John Moores University. He worked on routing protocols for WSN for his final project. His current research interests include Telecommunication Networks and programming. P. Johnson gained a Bachelor of Engineering in Electronics and Communication Engineering from GCT, Coimbatore, India followed by a Master of Engineering in Applied Electronics from Guindy Engineering College, Madras, India, both with distinction. Funded through Commonwealth

Copyright © 2013 Inderscience Enterprises Ltd.

Clustering-Biased Random Algorithm for Load Balancing (C-BRALB)

19

Scholarship she received her PhD from King’s College London in 2000. She has worked as a Research and Development engineer at Nortel Networks UK Ltd for two years during which she had a patent issued for a novel reconfigurable OADM. She is at Liverpool John Moores University and her research interests include energy-efficient techniques and protocols for mobile wireless and sensor networks. She is currently developing an active EU research consortium on this topic. She is an active member of IEEE since 2001.

1

Introduction

A Wireless Sensor Network (WSN) consists of a collection of minute nodes organised into a cooperative network. A sensor node is made of a processing capability (microcontrollers one or more, CPUs or DSCP chips), may have multiple type of memory (program, data and flash memory), an RF transceiver (very often with a single Omnidirectional antenna), a power source (batteries or solar cells or both), and various actuators and sensor. These features of the sensor node are all limited in resources and hence various schemes or algorithms have been devised to maximise their efficiency (Al-Karaki and Kamal, 2004; Touray et al., 2012). Parts of these schemes are routing protocols which can be classified into flat and hierarchical routing algorithms. The proposed routing algorithm C-BRALB fall under the hierarchical routing algorithm class, which differs from the flat routing algorithm in that it gathers and aggregates the data before transmitting to the sink so as to avoid duplicate transmission. In the hierarchical routing algorithm the sensor network is divided into clusters with each cluster electing a Cluster Head (CH) whose responsibility is to gather and aggregate the data in its cluster (Boyinbode et al., 2010; Qin et al., 2012; Poonguzhali, 2012). There are various techniques for generating the clusters and selecting the CHs. In some of the schemes CH rotation is also implemented using various methods mostly based on the energy residual of the CH. It is very common that hierarchical routing algorithms are more energy efficient than their flat based counterpart. One common assumption among the hierarchical routing algorithm is that the CHs can directly communicate to the sink in a single hop. However, in a large network which is typical of WSNs this assumption is not valid as the distance between the sink and a CH may be way greater than the transmission range of the CH (Jung et al., 2012). Even if this was possible it is not very energy efficient to route packet over long distance, as the energy consumed is proportional to the distance (d2 or d4 depending on the environment) between the source and destination (Min, 2003). Therefore in consideration of the limited energy resources of the nodes it is better to transmit the data in a multi-hop fashion thereby shortening the distance between hops. In this paper C-BRALB, a clustering technique based on IDD is proposed as an enhancement to our previously proposed Biased Random Algorithm BRALB (Touray et al., 2012). This is suitable for large WSNs and the simulation results show significant improvements in C-BRALB compared to BRALB without clustering. The reminder of this paper is divided into five sections. In section 2 related works is presented while in section 3 the proposed routing algorithm is discussed. In section 4 and 5 the performance of C-BRALB is analysed and the paper concluded respectively.

20

2

B. Touray et al.

Related work

The advent of wireless sensor networking created much interest in the research for networking. There are various sections of WSN which have seen much research input; notable the routing aspect. This resulted to a number of research proposals some of which are interest to this paper and therefore will be discussed in this section.

2.1 Directed Diffusion Based on Clustering and Inquiry (DBBCI) Directed Diffusion Based on Clustering and Inquiry (DBBCI) is an improved version of Improved Directed Diffusion (IDD) that is based on clustering. DDBCI is divided into four stages: cluster formation protocol state, interest diffusion, Data propagation and Reinforcement (Yu and Zhang, 2010). The major difference between IDD and DDBCI is that in DDBCI when a CH node receives an interest message it diffuses it into the cluster only if the target area of the interest message is in the cluster. Unlike IDD which will always do even if the target area is not in the cluster. In order to avoid diffusion of interest into a non-targeted cluster a CH in DDBCI maintain a table of member information whereby it queries its members of interest received. If no responses are received it meant the target area is not in its cluster and therefore the interest is not unnecessarily diffused into the cluster. After the successful propagation of the Interest to the source node, the target nodes that get the Interest send their data to the base station through paths consisting of CH nodes and border nodes. The base station is likely to receive the data from various paths; however, it reinforces a data transmission path using certain criteria. The first state of the DDBCI is the cluster formation protocol which is made of two states namely initial state and cluster formation state. The cluster formation protocol for DDBCI is inherited from IDD. The two algorithms use the same clustering technique. A node can exist in the following states namely: Ordinary Nodes (ORN), Cluster Head Node (CHN), Cluster Member Node (CMN), Candidate Cluster Head Node (CCHN), and finally Border Node (BN). All nodes start as ORN and they may transition into other states after the cluster formation protocol process (Yu and Zhang, 2010; Cui and Cao, 2007). The cluster formation protocol in DBBCI is the same as in IDD and is based on energy threshold THe, that is propagated by the base station to its neighbours and through the entire network. A node that receives this broadcast message will inspect the message and compare its residual energy to the energy threshold THe, setting. If the THe is greater than the node residual energy the node does not change its state. On the other hand if the node residual energy is greater than THe, the node transits to the CCHN state, and declares its state to the neighbours. In the case of two or more neighbour nodes being in the CCHN at the same time the tie breaker is their residual energy and the neighbour with the highest residual energy will win and transit to the CHN. The winner then broadcasts its CHN state to its neighbours. The CCHN node that lost the tie breaker will then transit to ORN state. The ORN nodes that receive a CHN message will join the cluster and sets its state to CMN. The ORN nodes that receive two or more CHN message at the same time set their node state to BN. In order to maintain the clusters a CHN keeps a member information table to store the information of each CMN in the cluster. When the cluster is fully formed the CMN nodes put their ID information and position into a message called mem_msg and send it to their


21

CHN node. When a CHN node receives a mem_msg it will check it against its member information table. If the record is already in the table the message will be discarded otherwise it is added to the table. After the cluster protocol formation, the next state is called the Interest Diffusion state. The Interest message is diffused into the network by the base station and is only forwarded to CHN s and BNs. The Interest message contains network information like target area, task type, data rate and timestamps. Interest table is kept by each sensor node to record Interest. The record of an Interest in MCN node and BN node includes its CHN node, data rate timestamp while the record of an Interest for CHN node contains neighbour CHN, data rate and timestamps. DDBC uses the local Interest tables stored in the sensor nodes to set up data propagation gradient. A Cluster Head that receives an interest will check its interest table to see if the Interest has already been received. If the same Interest is already in its Interest table then it will discard the Interest so as to avoid the propagation of redundant Interest. If it has not received a similar Interest the CHN will then update its local Interest table with the new Interest. It will then inquire the member information table to see if there is any item that matches the targeted area of the Interest. If there is no match then the data does not exist in its cluster; therefore the CHN will flood the Interest to its neighbour CHN nodes and BN nodes. On the other hand if there is a match, then the Interest is in the cluster so it will then inspect the member information table and then unicast the message to the particular neighbour that holds the matching data to the Interest (Yu and Zhang, 2010). The next state is the Data Propagation State. If the Interest propagated has a data match from a CMN node, BN node or a CHN node then there are two sets of propagation methods depending on the node type. The BN node or CM node with a match of Interest collected and data sensed, will sent the data to the neighbour CH node first and then to neighbour CH nodes gradients directly. In the case that the data source node is a CHN then the data matching the Interest will be sent to neighbour CHN nodes in gradients directly. With this scenario a CH node is likely to receive or send sensing data from many neighbour CH nodes; therefore the base station may receive the same sensing data from different paths across the network. So, in order to avoid the unnecessary duplication of sensed data transmission to the base station a middle CHN node that receives a sensing data from other CHN searches its Interest table for match Interest entry in its cache. If there is no match then the message is discarded. If there is a match the data is also discarded as the match indicates that the sensing data has already been forwarded. The CHN checks information of the neighbour cluster head and based on the neighbour head’s transmission rate the node may send all the received data to the appropriate neighbours or send the data to the neighbour in proportion (Yu and Zhang, 2010). The next state is the Reinforcement state. During the initial data propagation state several transmission paths from the source node to the base station are set up. In order to establish the probe gradient the source node, BNs and CHN s transmit the sensing data at a lower rate. After this initial low rate data transmission the base station sets up a reinforcement path so as to get the best path from the source node to the base station. Once the best path is selected the next incoming data will be transmitted along this path at a greater rate (Yu and Zhang, 2010).

22

B. Touray et al.

2.2 The LEACH routing protocol Low Energy Adaptive Clustering Hierarchy (LEACH) is a clustering-based protocol that randomly rotates local cluster heads so as to load balance the energy requirement among the sensors in the network. Its data aggregation technique reduces the amount of information to be sent to the base station thereby reducing the energy usage of the network as computation is much cheaper than communication (Heinzelman et al., 2000). With LEACH many clusters are created where each cluster has a cluster head whose main job is to collect and aggregate data from CMNs and transmit the data to the base station directly. With a large network there will be many clusters and therefore some of the cluster heads far away from the base station will not be able to transmit directly to the base station. This is one of the limitations of LEACH which is eliminated by the proposed protocol C-BRALB.

2.3 The HEED routing protocol Hybrid, Energy-Efficient, Distributed Clustering Approach for Ad Hoc Sensor Networks (HEED) (Younis and Fahmy, 2004) is an iterative clustering protocol that is primarily based on the remaining energy of each node which is fairly estimated. Intracluster “communication cost” is considered as a secondary clustering parameter whereby cost can be a function of neighbour proximity or cluster density. In HEED the primary clustering parameter is used to probabilistically choose an initial number of cluster heads whereas the secondary parameter is used to break a tie among cluster heads in the case that a node falls within the “range” of more than one cluster head. The cluster range or radius depends on the node configurable transmission power level used for intracluster announcements and during clustering. These cluster power level needs to at least cover two or more cluster diameters for the resulting interclustering overlay to be connected. Intracluster communication cost which is the secondary clustering parameter depends on cluster properties (e.g., size) and the type of intracluster communication allowed. In the case that all the nodes use the same power level for intracluster communication then the cost can be set to be directly proportional to node degree with the objective of load balancing the load among cluster heads, or inversely proportional to the node degree with the objective of creating dense clusters. In the case that different power level is used for intracluster communication then the cost for a node to select a cluster head depends on the expected intracluster communication energy consumption rather than the nearest cluster head. HEED is a very good protocol but it comes with loads of overhead as a result of its sophisticated clustering technique which requires loads of processing power for the sensor nodes.

3

The proposed routing algorithm

The proposed scheme is a modified version of BRALB routing protocol along with the clustering scheme. In C-BRALB when a node has a data to transmit it sends it to its cluster head. It is the duty of the cluster head to aggregate traffic and transmits it to the base station through neighbouring cluster head. A cluster keeps a counter for each neighbour where it records the messages it sent or received from them. When a Cluster


23

Head (CH) wants to forward data it inspects its neighbour cluster counters and then forwards the data to the CH whose counter has the least message. The messages sent or received by a cluster can be used to estimate the energy value of a cluster. Therefore the CH can always route to the cluster with more energy by inspecting its CH counters and hence load balance the traffic across the network. The clustering scheme proposed is based on the IDD clustering technique, the same technique used by DDBCI, however, in C-BRALB some modifications are made to make the clustering technique more efficient. The major differences are the omission of Interest diffusion phase and the introduction of a total different Data Propagation phase. C-BRALB is composed of two phases which are the Cluster formation and Data Propagation. The first phase of C-BRALB is the cluster formation protocol made of two states namely initial state and cluster formation state. A node can exist in the following states namely: Ordinary Nodes (ORN), Cluster Head Node (CHN), Cluster Member Node (CMN), Candidate Cluster Head Node (CCHN), and finally Border Node (BN). All nodes start as ORN and they may transition into other states after the cluster formation protocol process. The cluster formation protocol in C-BRALB is the same as in IDD with added information and message counters. In order to maintain the clusters a CHN keeps a member information table that stores the information of each CMN in the cluster. When the cluster is fully formed the CMN nodes put their ID information, energy level information and position into a message called mem_msg and send it to their CHN node. A Cluster Head keeps a member node table and a message counter for each neighbour cluster. It uses its member node table to record information about nodes within its cluster while the neighbour cluster counter is used to record messages sent or received from neighbour clusters. When a CHN node receives a mem_msg it will check it against its member information table. If the record is already in the table the message will still be inspected for energy level update otherwise it is added to the table together with its new parameters. When a CHN’s energy value falls below the energy threshold THc, it then broadcasts its THc value to all the nodes in its cluster. A node that receives this broadcast message will respond to the CH with a mem_msg updating its energy level to the CH. The Cluster Head will then inspect all the mem_msg packets and will select the node with the highest energy as the next CH. The previous CH will then update the new CH with the cluster member table and the neighbour cluster counters. This process is repeated every time a CHN’s energy falls below the THc so as to rotate the CH duty among the nodes within a cluster. This helps to load balance the transmission of traffic among the nodes within a cluster. The next phase is the Data Propagation. When a node has a data to send it will transmit the data to the CH. When the CH receive the message it will send it to one of its neighbour CHs at random at the initial stage. After forwarding the message to a neighbour cluster it will then update the message counter for that neighbour cluster. If the same CH have a data to forward now it will then inspect its neighbour cluster counters and will forward to the one with the lowest message counter. In this way each cluster will load balance the data transmission to its neighbour clusters and hence maximise the network life time. This process is repeated every time a CH have a data to transmit.

24

4

B. Touray et al.

Performance evaluations

NetLogo simulator is used in order to evaluate and compare the proposed C-BRALB routing algorithm with BRALB. NetLogo is well suited for modelling complex systems developing over time. It has an inbuilt library with sample models as guide. There is even a sample model for Networks showing the implementation of the directed diffusion algorithm and preferential attachment thus it is suitable for modelling routing protocols. The environment used to test the performance of these two algorithms was modelled using Netlogo’s graphic design tool in order to simulate a network. By using Netlogo, network parameters were varied in order to study their effect on the overall performance of each algorithm and do the comparison. The simulator facilitates to deploy the number of resource-constrained nodes and their required topology connectivity. The simulation was run on an m-dimensional node network with the number of nodes equal to m × m each having four neighbours excepting the boundary nodes for the BRALB algorithm. For the C-BRALB the number of nodes and system parameters remain the same except that the nodes are randomly deployed within the same area. In the simulation test bed as depicted in Figure 1, the total number of nodes is considered to be m × m with node connectivity of 4. Figure 2, considered the same scenario as in Figure 1 except that the nodes are randomly deployed and clustered using the C-BRALB algorithm. Energy consumption and end to end delay are used to evaluate the performance of the two algorithms. The algorithms were implemented in the simulator for comparison. Tests were run for 100 time units, which were considered as ticks. For every performance metric 20 samples were taken and the mean calculated. The performance metrics and their comparisons for each algorithm are described below: Figure 1

Simulation snapshot for BRALB (see online version for colours)

Figure 2

Simulation snapshot for C-BRALB (see online version for colours)


25

4.1 Delay The delay is the average time for a message to be routed from the source node to the destination node. The delay is expressed as the number of hops in each route as opposed to the actual time it takes to traverse the route since the most significant delay is the delay experience in each node along the route. The delay is expressed in time by taking each hop as a delay of 1ms. In Figure 3 BRALB has more delay than C-BRALB. The small delay experienced by C-BRALB is due to the clustering which drastically shortens the minimum path length between a source and a destination. With these results C-BRALB is therefore even suitable for delays sensitive application thus overcoming one of the limitations of BRALB. Figure 3

End to end delay (see online version for colours)

4.2 Energy consumption The energy consumption here in Figure 4 is the total amount of energy used in the network for routing purposes and maintenance. The total amount of energy is equivalent to the total number of hops the messages traversed as each hop use one unit of energy. The energy used by C-BRALB is significantly less than that of BRALB. Implementation of clustering scheme has also contributed to this reduction in energy consumption. Figure 4

Total spent energy (see online version for colours)

26

B. Touray et al.

4.3 Packet Delivery Ratio/Fraction (PDF) Packet Delivery Ratio/Fraction (PDF) is the ratio between the number of packets delivered to the receiver and the number of packet sent by the source. From the results obtained in Figure 5 the pdf for C-BRALB is better than that of BRALB and it is almost 100%. This significant improvement again is due to the clustering schemed used in C-BRALB. Figure 5

5

Packet delivery ratio (pdf) for 100 messages (see online version for colours)

Conclusion

In this paper a new hierarchical routing algorithm namely C-BRALB is proposed and evaluated using simulations. It divides the sensor networks into many clusters and CHs are elected in each cluster. There is also a CH rotation within each cluster in order to make the algorithm more energy efficient. The Cluster Head is responsible of collecting and aggregating the sensed data within the cluster and then transmit it to the base station through a neighbour CH. From the above results, C-BRALB is shown to have improved the energy consumption by 1000 s of units compared to the biased random algorithm. This is a very significant saving especially, as this algorithm distributes the load among the nodes unlike the traditional algorithm and thus avoids network partitioning. This in turn extends the network lifetime. This energy conservation is also seen to be scalable with increasing network size. C-BRALB is able to shorten the network diameter by several factors compared to BRALB and hence increased the energy efficiency of the network. Another major advantage achieved by this proposed algorithm is the considerably reduced network latency. The end-to-end delay for a message is reduced to a few ms as opposed to the former value ranging up to 1000 ms. This is a huge improvement on the previously demonstrated biased random algorithm and is on par with the values reported for traditional algorithms algorithm. This also makes C-BRALB to be applicable to delay sensitive applications. The larger the network size the more energy efficient is C-BRALB as compared to BRALB. C-BRALB is proposed so as to address the issues encountered by BRALB with increasing network size and node density. With BRALB the energy efficiency of the algorithm decreases with increasing network size. With C-BRALB this problem is addressed and the algorithm is scalable with network size. In the next phase of research,


27

performance of the C-BRALB algorithm will be evaluated using extensive simulations and will be compared with the LEACH routing algorithm.

References Al-Karaki, J.N. and Kamal, A.E. (2004) ‘Routing technologies in wireless sensor networks: a survey’, IEEE Wireless Communications, Vol. 11, No. 6, pp.6–28. Boyinbode, O., Hanh, Le., Mbogho, A., Takizawa, M. and Poliah, R. (2010) ‘A survey on clustering algorithms for wireless sensor networks’, Network-Based Information Systems (NBiS) 2010 13th International Conference, September, pp.358–364. Cui, Y. and Cao, J. (2007) ‘An improved directed diffusion for wireless sensor networks’, Wireless Communications, Networking and Mobile Computing WiCom 2007 International Conference, September, pp.2380–2383. Heinzelman, W.R., Chandrakasan, A. and Balakrishnan, H. (2000) ‘Energy-efficient communication protocol for wireless microsensor networks’, System Sciences, 2000. Proceedings of the 33rd Annual Hawaii International Conference, January, Vol. 2, p.10. Jung, S-M., Kim, N-U. and Chung, T-M. (2012) ‘The clusterhead chaining scheme considering scalability of the wireless sensor networks’, Information Networking (ICOIN), 2012 International Conference, February, pp.497–500. Min, R.K. (2003) Energy and Quality Scalable Wireless Communication, PhD Thesis, Massachusetts Institute of Technology, USA. Poonguzhali, P.K. (2012) ‘Energy efficient realization of Clustering Patch routing protocol in wireless sensors network’, Computer Communication and Informatics (ICCCI) 2012 International Conference, January, pp.1–6. Qin, X., Zhang, H. and Zhang, Y. (2012) ‘Research on wireless sensor networks clustering routing algorithm based on energy balance’, Measurement, Information and Control (MIC) 2012 International Conference, May, pp.387–391. Touray, B., Shim, J. and Johnson, P. (2012) ‘Biased random algorithm for load balancing in wireless sensor networks (BRALB)’, 15th International Power Electronics and Motion Control Conference, EPE-PEMC 2012 ECCE Europe, LS4e.1-1, LS4e.1-5, September 2012. Younis, O. and Fahmy, S. (2004) ‘HEED: a hybrid, energy-efficient, distributed clustering approach for ad hoc sensor networks’, IEEE Transactions on Mobile Computing, Vol. 3, No. 4, October–December, pp.366–379. Yu, J. and Zhang, H. (2010) ‘Directed diffusion based on clustering and inquiry for wireless sensor networks’, Computer Science and Information Technology (ICCSIT), 2010 3rd IEEE International Conference, July, Vol. 2, pp.291–294.