Energy Efficient Load Balancing among Heterogeneous Nodes of Wireless
Sensor Network
Chandrakant N
Bangalore, India
[email protected]Abstract
Energy efficient load balancing in a Wireless Sen-sor Network needs to distribute workloads across mul-tiple computing nodes based on its type of functional-ity such as temperature sensing, light sensing etc keep-ing in mind of energy cost. Hence, energy efficient load balancing can be achieved and it helps to optimize re-source usage, maximize throughput, maximize network lifetime, minimize response time, and avoid overload by distributing work among similar type of sensor nodes with energy efficient routes. This will make use of mul-tiple sensor nodes with load balancing instead of a sin-gle sensor nodes which may increase reliability through redundancy. Physical group represents a set of nodes which are physically neighbours to a node, where as logical group represents a set of nodes which grouped based on its functionality. Energy efficient routes can be evaluated in virtual groupings(PG,LG), route will cho-sen based on the cost of energy inspite of the route be-longingness.
Keywords: WSN, Energy Efficient, Load Balance, Het-erogeneous, Logical Group, Physical Group
1. Introduction
A Wireless Sensor Network (WSN) is set of sensor nodes which communicates via wireless links and these cannot have a unique topology. These nodes will coop-eratively pass their data through the WSN to a main lo-cation. Load balancing in WSN involves distribution of all computational and communicational activities over two or more nodes in the network. This load balanc-ing can help us to reduce the execution time of the ac-tivities and to make sure that all the resources present in the system are utilized optimally. Ideally load bal-ancing algorithm selects the node for process execu-tion based on the available informaexecu-tion about all the
resources present in the network. Load balancing al-gorithms can be Static, Dynamic and Adaptive. Static algorithms takes decisions using a priori knowledge of the network, hence the overhead entailed in static al-gorithms is almost zero. In the case of dynamic algo-rithms, decisions are based on system state information (the loads at nodes), hence they incur overhead in the collection, storage and analysis of network state. Adap-tive Algorithms is kind of dynamic algorithms which adapt their activities by dynamically changing the pa-rameters of the algorithm to suit the changing network state.
WSNs require appropriate algorithm that make judi-cious use of the finite energy resources of the heteroge-neous sensor nodes, hence we need justify the compu-tation cost before executing load on it !
2. Literature Survey
Load balancing in heterogeneous nodes of WSN can be evaluated through grouping nodes, this is a novel idea but we used few paper as input to this concept.
Paper [7] proposed a load-balanced clustering algo-rithm [13] for Wireless Sensor Networks on the basis of their distance and density distribution. In the cluster, nodes can join the cluster head by considering the clus-ter size and communication radius. Further, load bal-ancing with energy efficiency[9][1] comprises of two parts [11]. First part being determining the number of cluster heads based on the nodes’ distribution and communication radiuses and second being to select the cluster heads according to the residual energy, mobility, number of single-node cluster and distances to cluster heads from their member nodes and to the server from cluster heads.
Paper [6] proposed a architecture where new appli-cations can be rapidly developed through flexible ser-vice composition. This architecture helps to congestion control and load-balancing which adaptively adjust the
work load over multipaths. In this algorithm, an evalua-tion metric and path vacant ratio is used to evaluate and find a set of link-disjoint paths from all available paths. On top of this algorithm, a threshold sharing algorithm is applied to split the packets into multiple segments that will be delivered via multipaths to the destination depending on the path vacant ratio.
In paper [10], author have investigated the load balancing effect of stochastic routing in undirected and directed WSNs with randomly positioned nodes. Concluded that stochastic routing does not necessar-ily achieve energy efficient load balancing in undi-rected networks. They have analyzed the performance of the distributed and decentralized stochastic routing algorithm, namely expander routing method, expander routing method performed significantly better in terms of packet transmission delay while achieving energy-efficient load balancing in directed networks.
In Paper [12], proposes Secure Load Balancing (SLB) protocol that employs pseudo-sinks that are a small number of special, tamper-proof sensor nodes with more computational, storage, and energy re-sources. This algorithm mitigates accuracy problem by securely relaying data from congested clusters to nearby free clusters or pseudo-sinks.
In Paper [5], authors have studied the potential en-ergy conservation parameters and achieving maximum network lifetime by balancing the traffic throughout in the WSN. They have shown that distributing the traffic generated by each sensor node through multiple paths instead of using a single path allows significant energy savings, hence they proposed a new analytical model for load-balanced systems.
Paper [3] talks about energy efficient routing in WSN which is not straightforward inspite of having Directed Diffusion data-centric routing protocol. Data is for-warded through all sink nodes imposing the overhead of sending useless data, hence author has proposed a Multi-Sink Directed Diffusion (MSDD) to address this problem by forwarding data toward the nearest sink. This protocol implements a load-balancing by selecting the next nearest sink after the energy level of nodes in the original path falls below a certain threshold value.
3. Proposed Techniques and
Implementa-tion
This paper is using techniques of paper [2] with load balancing algorithms. Initially we are considering load balancing factor and generating simulated results.
The network is logically(Logical Group-LG) and physically(Physical Group-PG) separated based on the functionality and its physical existence respectively. In
Figure 1. WSN Group Formation
this network, each node is having a LG (Logical Group) Id which is different from cluster group, where LG Id is unique and logically grouped based on the identical functionality of sensor nodes, but a node can have more than one group’s Ids to indicate that it can participate in more than one feature extraction e.g., earthquake de-tection and landslide observation. Whenever a sensor node sends a packet with LG Id to its neighbours, if any neighbours are logically and physically available within the coverage area then such nodes can receive this packet or else any immediate neighbour can pass this packet to its immediate neighbour and so on, until it reaches to appropriate LG nodes. Here Group Id is nothing but Logical Group ID which is representing the set of nodes of similar functionally, hence communica-tion between different groups is easy based on the group Ids. In Fig. 1, node 1 and 5 are in PG(Physical Group) and LG to the node X respectively, however node 2 is included in both groups(LG and PG). In each LG, ev-ery node can receive the packets which is mentioned for entire group, but nodes of PG are neighbours hence interested node can receive these packets.
Algorithm 1 Load Balancing in WSN
Require: Initialize N Nodes with L, P G, LG, etc
Require: L <= Number of Work Load
Pro-cesses(l1,l2,l3...ln=L)
1: while l <= L do
2: while i <= N do
3: if Type of node i belongs to a LG == Process Type of l then
4: Allocate this process l to Node i.
5: else
6: Allocate load to free node which can
be-longs to PG.
7: end if
8: end while
9: end while
process assigning to nodes in the network. The average amount of energy consumed by node u per unit of time due to the different transmissions within the WSN is denoted by E(u) [4],
E(u)=Eidle(u) + X v∈V X p∈P (v) w(p) ∗ A(v) ∗ E(u, p) (1)
Here, Eidle(u) is the average amount of energy con-sumed by node u per unit of time during its idle state. The lifetime of sensor node u is calculated by,
T(u)=Einit/E(u) (2)
Here, Einit is the initial amount of energy provided to each sensor node.
Generally, the load balance for a given graph G rep-resenting the network with n nodes where each node contains work load wi, the goal is to distribute load across the edges so that finally the weight of each node is (approximately) equal to,
¯ wi= n X j=1 wj/n (3)
Let f be the fraction of the total network area covered by a mobile node [7] , then
f =πR 2
A (4)
The average number of neighbours ¯n of the network can be obtained by using the following equation,
¯
n = (N –1)kf (5)
where k is a constant, referred to as a connectivity pa-rameter.
The relationship between the local density, the cov-ered set and the forwarding probability has been sum-marized through equation (6). Assuming that, g be the number of adjacent neighbours of node n1 and gbbe the number of nodes of n1 that are covered by the broad-cast and the forwarding probability at the node n1 is as follows, Pn1 = g−gb ¯ g ; if g ≤ ¯g (6) g−gb g ; if g > ¯g
Adding all the nodes of physical or logical groups are equivalent number of nodes in the network. Say, K,L be the total number of groups of PG and LG respectively and R, S be the size of each group of PG and LG respec-tively which is specified in the below equation (7) and (8).
Figure 2. Number of processes in 100 nodes v/s time in load-balancing .
N= K X i=0 Ri (7) N= L X i=0 Si (8)
Group Relations can be defined as follows, let there is a set of 2 groups like M and W and wanted to express which node of M is communicating with which node of W. Here, one way to do that is by listing the set of pairs (m,w) and recognizing the nodes. The accessing relation can be represented by a subset of the Cartesian Product M × W. In general, a relation R from a set A to a set B will be understood as a subset of the Cartesian Product A × B, i.e., R ⊆ A × B. If an element a ∈ A is related to an element b ∈ B, we often write aRb instead of (a, b) ∈ R. The set
{a ∈ A | aRb for some b ∈ B} is called the domain of R. The set
{b ∈ B | aRb for some a ∈ A} is called the range of R.
The load balancing [8] in the given a graph G(summation of PG and/or LG) contains N nodes where each node contains work load wi, here work load is distributed across the edges/nodes so that finally the weight of each node is (approximately) equal to ¯wi, i.e.,
¯ wi= n X j=1 wj/n (9)
The simulated results for load balancing are shown in Figure 2 and 3, we have considered PG and PG with
Figure 3. Number of processes in 500 nodes v/s time in load-balancing .
Figure 4. Total Energy of 50 Nodes v/s Number of Processes in the Network.
Figure 5. Total Energy of 200 Nodes v/s Number of Processes in the Network.
LG scenarios. Load balancing via PG is a normal al-gorithm to distribute the load among sensor nodes, it can be traditional way of assigning processes to sen-sor nodes. Applying algorithm of PG with LG concepts takes less time compared with PG concepts. Figure [8] and [7] shows simulation results of 100 and 500 nodes respectively. Hence, the proposed algorithm can be use-ful in the case of heterogeneous nodes.
With load balancing algorithm, further we have ap-plied energy efficient strategy so that network life time can be improved/increased along with load balancing. A brief algorithm is given in Algorithm 2, which de-scribes the process of assigning work load to nodes with energy efficient.
In this network, L is Number of Work Load Pro-cesses called l1,l2,l3...ln and N is the Number of Nodes in the network called n1,n2,n3...nm and E is Energy Level of each node in N called e1,e2,e3...en, here e1 corresponds to n1, similarly e2 corresponds n2 and so on.
The simulated results for energy efficient load bal-ancing are shown in Figure 4 and 5 for 50 and 200 nodes respectively, both graphs of simulation results are iden-tical and shows that we can increase the network life time while allocating loads across nodes.
4. Conclusions
Energy efficient load balancing is very essential in Wireless Sensor Network which helps to optimize re-source usage, maximize throughput, maximize network lifetime, minimize response time, and avoid overload
Algorithm 2 Energy Efficient Load Balancing in WSN Require: Initialize N Nodes with L, P G, LG, etc
Require: L <= Number of Work Load
Pro-cesses(l1,l2,l3...ln=L)
Require: N <= Number of Nodes in the
net-work(n1,n2,n3...nm=N)
Require: E <= Energy Level of each node in N (e1,e2,e3...en=E), here e1 for n1, e2 for n2....en for nm
1: while l <= L do
2: while i <= N do
3: if Type of node i belongs to a LG == Process Type of l then
4: while Until find a node from i which
con-sumes min. energy e do
5: Allocate this process l to Node i.
6: end while
7: else
8: while Until find a node from i which con-sumes min. energy e do
9: Allocate this process l to Node i.
10: Allocate load to free node which can
be-longs to PG.
11: end while
12: end if
13: end while
14: end while
by distributing work among similar type of sensor nodes with energy efficient routes. This paper proposes energy efficient load balancing among sensor nodes based on the logical and/or physical grouping of Wireless Sensor Nodes. The simulation result is encouraging hence it is worth of using proposed algorithm for energy efficient load balancing heterogeneous WSN.
References
[1] F. Bouabdallah, N. Bouabdallah, and R. Boutaba. Load-balanced routing scheme for energy-efficient wireless sensor networks. In Global Telecommunications Con-ference, 2008. IEEE GLOBECOM 2008. IEEE, pages 1–6, 2008.
[2] N. Chandrakant, P. D. Shenoy, K. R. Venugopal, and L. M. Patnaik. Restricting the admission of selfish or malicious nodes into the network by using efficient se-curity services in middleware for manets. In Proceed-ings of the 2011 International Conference on Commu-nication, Computing & Security, ICCCS ’11, pages 489–492, New York, NY, USA, 2011. ACM.
[3] A. Eghbali, N. Javan, A. Dareshoorzadeh, and M. De-hghan. An energy efficient load-balanced multi-sink routing protocol for wireless sensor networks. In
Telecommunications, 2009. ConTEL 2009. 10th Inter-national Conference on, pages 229–234, 2009. [4] N. B. Fatma Bouabdallah and R. Boutaba. On balancing
energy consumption in wireless sensor networks. pages 1–16, march 2008.
[5] Fatma Othman, Nizar Bouabdallah and Raouf Boutaba. Load-balanced routing scheme for energy-efficient wireless sensor networks.
[6] S. Li, S. Zhao, X. Wang, K. Zhang, and L. Li. Adaptive and secure load-balancing routing protocol for service-oriented wireless sensor networks, 2013.
[7] Y. Liao, H. Qi, and W. Li. Load-balanced clustering algorithm with distributed self-organization for wireless sensor networks. Sensors Journal, IEEE, 13(5):1498– 1506, 2013.
[8] Robert Elssser and Burkhard Monien and Stefan Schamberger. Load balancing in dynamic networks. [9] A. Tarachand, V. Kumar, A. Raj, A. Kumar, and P. Jana.
An energy efficient load balancing algorithm for cluster-based wireless sensor networks. In India Conference (INDICON), 2012 Annual IEEE, pages 1250–1254, 2012.
[10] U. Wijetunge, A. Pollok, and S. Perreau. Load balanc-ing effect of stochastic routbalanc-ing in wireless sensor net-works. In Telecommunication Networks and Applica-tions Conference (ATNAC), 2012 Australasian, pages 1–6, 2012.
[11] F. Xia, X. Zhao, H. Liu, J. Li, and X. Kong. An energy-efficient and load-balanced dynamic clustering proto-col for ad-hoc sensor networks. In Cyber Technology in Automation, Control, and Intelligent Systems (CY-BER), 2012 IEEE International Conference on, pages 215–220, 2012.
[12] S. zdemir. Secure load balancing for wireless sensor networks via inter cluster relaying. In Kithab Proceed-ings, pages 249–253, 2007.
[13] R. Zhang, Z. Jia, and L. Wang. A maximum-votes and load-balance clustering algorithm for wireless sen-sor networks. In Wireless Communications, Networking and Mobile Computing, 2008. WiCOM ’08. 4th Interna-tional Conference on, pages 1–4, 2008.
