There are many other characteristics that could be different. The cluster scheme can be applied in many different manners in order to achieve diverse benefits. This section compares our proposal with others in existence in order to show the benefits of our protocol. Only those architectures whose protocol has been available and accessible are included.
7.1. Architecture Comparison
Krishna et al.
presented in [35
] a methodology for routing and topology information maintenance in mobile wireless network based on the existence of clusters in random graphs. They divided the graph into a number of overlapping clusters. There are no cluster heads in the proposal. It is an on-demand source routing. The performance of the routing protocol proposed by them is determined by the average cluster size. The effectiveness of their approach lies in the fact that any routing protocol can be directly applied to the network, replacing the nodes by clusters. They designed nine messages, but their approach has to be implemented over another routing protocol. They proposed a standard distance vector routing protocol to apply their proposal. Taking into account that AODV has seven messages. Their implementation needs 16 messages.
CBRP (cluster-based Routing protocol) was proposed in [6
]. The protocol divides the nodes of the ad hoc
network into a number of overlapping or disjoint 2-hop-diameter clusters in a distributed manner. CBRP uses IP Protocol for routing purposes and interoperability with fixed networks. It uses six messages plus the ARP messages, so it needs eight messages to work properly. As a source routing protocol, there is an overhead of bytes per packet.
], the authors proposed a Cluster-Based Security Architecture for ad hoc
networks. They proposed a division of the network into clusters, with one special head node each, for a distributed public key infrastructure. These cluster head nodes execute administrative functions and hold shares of a network key used for certification.
KCLS protocol was proposed in [37
]. The paper describes a location-service protocol based on the clustering architecture, which is able to balance the tradeoff between the communication overheads and the accuracy of location information. It has the capability of cluster-level self-route recovery against interlink failures. Taking into account that KCLS is based on the KCMBC protocol and on a Link State protocol, it needs 13 messages plus the protocol needed to acquire their position using GPS.
] the authors proposed an adaptive clustering scheme for spatial reuse of the bandwidth (relying on a code division access scheme) for multimedia support in Mobile Wireless Networks. They use only one code within each cluster. The clusters are independently controlled and are dynamically reconFigured as nodes move. Bandwidth can be shared or reserved in a controlled fashion in each cluster.
LEACH (Low-Energy Adaptive Clustering Hierarchy) protocol was proposed in [34
]. It is a clustering-based protocol that utilizes randomized rotation of local cluster-heads minimizing the global energy usage by evenly distributing the energy load among the sensors in the network. LEACH uses localized coordination and incorporates data fusion into the routing protocol to reduce the amount of information that must be transmitted to the base station. It is completely distributed, requiring no control information from the base station, and the nodes do not require knowledge of the global network in order to operate.
] presented the “Base Station Controlled Dynamic Clustering Protocol” (BCDCP). Their proposal utilizes a high-energy base station to set up clusters and routing paths, perform randomized rotation of cluster heads to avoid cluster head overload, and carry out other energy-intensive tasks. It distributes the energy dissipation evenly among all sensor nodes to improve network lifetime and average energy savings.
], authors proposed CLACR (Core Location-Aided Cluster-Based Routing Protocol for Mobile Ad Hoc
Networks). CLACR splits the network into square clusters. Cluster heads compute the desired route using Dijkstra algorithm, which reduces the number of nodes participating in routing, the routing traffic and the route setup time.
CBLARHM (Cluster Based Location-Aware Routing Protocol for Large Scale Heterogeneous MANET) was proposed in [40
]. The system uses the geographical location information of mobile nodes, provided by global positioning systems (GPS), to confine the route searching space, for the specified destination node.
WCA (A Weighted Clustering Algorithm for Mobile Ad Hoc
Networks) was presented in [41
]. They proposed an on-demand, weight-based distributed clustering algorithm that takes into consideration the ideal degree, transmission power, mobility, and battery power of mobile nodes. The clustering algorithm tries to distribute the load as much as possible aimed to reduce the computation and communication costs. The algorithm is executed only when there is a demand, i.e.
, when a node is no longer able to attach itself to any of the existing cluster-heads.
] CLTC was presented, a Cluster-Based Topology Control Framework for Ad Hoc
Networks. CLTC uses a centralized algorithm within a cluster and between adjacent clusters to achieve strong connectivity. It utilizes a hybrid approach to control the topology using transmission power adjustment and yet achieves the scalability and adaptability of a distributed approach with localized information exchange between adjacent clusters. CLTC framework guarantees global k-connectivity as long as the original topology is k-connected.
compares the described protocols. The number of messages of some protocols is provided by the explanations of their authors in the referenced paper. Some of them do not take into account messages such as the new nodes messages or messages to provide fault tolerance, or are based on other algorithms not described in the paper (although in some cases we have found these and taken them into account). Many other proposals have been found in the literature, but they have not been included because of the lack of sufficient information in the original publication to fill in the rows of Table 3
, or because the authors just described the algorithm, not the protocol, or because they are slight extensions of the proposals shown in the Table. Several features in our proposal may be highlighted. First, it is the only one that is able to use several routing protocols in the same network, and it does not depend on a specific routing protocol so it could be adapted to the environment issues. It does not have too many messages compared with the other ones (despite the simplicity of the CBRP, some procedures are not explained and it does not provide fault tolerance). Second, it is the only one where a new node selects the cluster not only by the proximity or radio signal strength but also takes into account the available capacity of the neighbors (which depends on the available energy). Third, it is the only one that has been proposed to create parallel networks. Our proposal has been designed to provide fault tolerance and our design is described in detail giving all the messages needed to run properly.
There are several differences in the metric used to elect the Cluster Head. Some of them use the lowest NodeID or their position, while others use just a random system. Others are not explained or propose a weak system. Between the most complex metrics, we distinguish CBLARHM that uses a node-weight heuristic parameter, based on the ideal number of nodes in a cluster, the battery power, the average link stability and the average dependency probability, to elect the head cluster node, but this is very impractical because it is difficult to determine the value of some of these parameters.
WCA uses a combined weight metric based on the ideal node degree, transmission power, mobility and the battery power of the nodes. It is a good idea and seems very useful, but they propose this metric for a cluster-based general purpose algorithm, and maybe these parameters are good for a specific case, but not for all cases, as some parameters could be missing such as the node's position or the node's load. Our metric does not take mobility into account since the entire cluster could be moving and avoids continually selecting the motionless nodes. On the other hand, it does take into account the more stable node in the cluster and its energy, thus making the system very simple and practical.
7.2. Measurement Comparisons
First of all, we want to emphasize that all works found in the literature provide only measurements taken from simulations, not from real deployments nor from controlled testbeds. Table 4
gives the type of measurements provided by several papers in the literature. Some of them are focused on measuring parameters related with the cluster size and the number of clusters. Due to cluster-based networks are mostly used for energy saving, most of them simulate energy issues.
Taking into account that the more message transmissions, the more energy dissipation, only CBRP [6
] and CLACR [38
] (and may be CLTC [41
], depending on the number of messages transmitted using GPS) will consume less energy than our proposal.
Taking into account the measurements provided in this paper, we will compare our measurements with some measurements provided in other authors in their works. We are not going to implement the same test bench, we will just take their measurements and compare them with our measurements in some particular cases.
Studying the results presented by Bechler et al.
], taking into account the same keepalive interval (30 seconds), we obtained an average value of 17.41 packets per second for a topology of three hops, while they obtained higher values (between 50 and 100 packets per second) in a random topology with 15 nodes. In terms of overhead (packets per second), it gives us an improvement of more than 65.2 % compared the best case of their protocol.
Hollerung presented in [45
] several graphs that show the packet delivery ratio versus the number of nodes in the cluster network. Close to 1 packet delivery ratio (0.99 packet delivery ratio) was shown for 25 nodes, while we can see that it agrees our measurements. Once our network has converged (after the setup phase), we obtain 15.9 packets/s for 16 nodes. It gives us almost the same packet delivery ration. The worst case has been presented in [41
] by Shen et al.
because they measured 2.5 messages per node for 100 nodes.
In Subsection 6.4 we have obtained a mean value of 19.16 milliseconds when there were three hops between the source and the destination. In [30
] we can see that the average delay for three hops was higher than 500 milliseconds in the best case (and lower than 1,500 milliseconds in the worst case).