The network layer is a huge research focus area for WSN load balancing because the routes that are chosen have a huge impact on the total energy expended in a WSN over time. Inefficient route determination causes the network to prematurely die out. As long as transmit/receive operations and access to the channel are performed efficiently (those process are well tried and tested) the network layer offers fruits for load balancing opportunities in any custom WSN. Network layer WSN routing protocol surveys have been performed in [5]–[9], [26]–[28]. The literature generally includes two areas of
network layer load balancing: the traditional minimum cost path for information to pass through many nodes to the gateway and group clustering. In Figure 1, an example of a minimum cost route in which the route from node C to the gateway was determined to be C Q N D gateway is provided.
The traditional minimum cost path involves each node determining the next hop to send the information according to an algorithm that incorporates some type of cost parameter. The cost parameter varies significantly in the literature and can be any cost metric of interest to the application being designed. Traditional minimum cost routing can degrade a WSN. In [29], the authors showed that minimum cost routing paths tend to overuse certain nodes. These nodes are then overloaded with traffic from the rest of the network and, thus, experience faster energy depletion compared to other nodes. To address this problem, the authors include a cost metric that accounts for network load where the greater the load, the greater the cost metric. These costs are then used to route around high traffic nodes, which provides an increase in network lifetime and fault tolerance [29].
We implement an energy-efficient version of Dijkstra’s algorithm [30] in our simulations to analyze the performance of minimum cost routing in WSNs. Minimum cost routing is generally based on Dijkstra’s shortest path algorithm [30]. We describe the specifics of this routing algorithm in Chapter VI.
An alternative to traditional minimum cost routing is clustering. In clustering, a cluster head (CH) is chosen from a group of nodes to serve as an intermediate relay between a group of nodes and the gateway(s). In other words, nodes with data to send forward the data to their respective CH, at which point the CH then forwards the data to the gateway. This scenario is presented in Figure 7, which was adapted from Figure 3. Cluster heads P, M and O are chosen from a clustering algorithm at the network layer followed by an association of child nodes to associate with the CH to form three smaller WSNs.
An illustration of a multi-gateway network is shown in Figure 7, which adds further complexity at the network layer. Whether the network layer is using a traditional
minimum cost routing algorithm or a clustering algorithm, the algorithm must be designed to use the gateway that provides the least energy cost to a WSN. In the case of clustering, this may mean that a CH chooses the closer gateway to minimize required transmission power. Another instance may require the CH to choose the gateway that is further away to avoid increased interference at the closer gateway.
Figure 7. Clustering in a multi-gateway WSN with cluster heads: P, M, and O.
The foundation of a clustering algorithm is the method for choosing a CH and the method for nodes in the WSN to associate with their CH. One popular clustering mechanism is LEACH, originally proposed by [1]. This algorithm was originally presented in 2000 as a proposed solution for CH election criteria. Specifically, the high energy CH role is periodically rotated throughout the network so that energy consumption at each node is balanced throughout the network lifetime. There has been a significant amount of research on LEACH and variants of the algorithm that have been proposed. At the time of writing of this thesis, the IEEE Explore database indicated 1604 citations for [1], while GOOGLE citation tracking claimed it was cited 9600 times. The author republished LEACH in 2002, addressing several improvements of the algorithm [2]. We utilize LEACH in our research and explain the algorithm in detail later.
Another clustering mechanism is that of zone based clustering, in which nodes are assigned to a cluster according to a predefined spatial arrangement [31]. Cluster heads are then chosen based on some prioritization of the nodes. This prioritization was performed randomly in [31]. We employ this technique in our simulations. In addition, we implement our own zoning technique and introduce a novel energy-efficient process to elect CHs with the desire to minimize WSN energy depletion.
Network control message schemes are implemented at the network layer and provide an autonomous control mechanism for the network to manage communications and errors. The control message scheme that dominates the internet is called internet control management protocol (ICMP) in which events trigger the generation of control messages to be exchanged between network devices [32]. Version 4 of ICMP (ICMPv4) is based on the internet protocol version 4 (IPV4). The applicability of ICMP to WSN load balancing is that as control messages are generated, they must be processed by WSN nodes, which may cause unnecessary energy use. Thus, control message implementations for WSNs must specifically minimize the amount of messages generated to lower node overhead. For example, [33] introduced a multi-path routing algorithm with a reduced control message structure for WSNs that ensures messages are only sent to specific nodes instead of to everyone (flooding). In terms of clustering algorithms, control messaging is used for CH selection, CH announcement and during the phase that nodes are associating with their CH. We do not simulate control messages in this thesis. However, we adapt best practices from the literature to describe our proposed messaging scheme and offer further simulations as future work.