*3.9.6. Cluster overhead*

Clustering requires explicit clustering-related information exchanged between node pairs. Clusters cannot be formed or maintained by non-clustering-related messages, such as routing information or data packets. The main challenge in clustering is the communication overhead introduced to formation and maintenance of a stable cluster, and elects its stable CH. Most of the recently proposed protocols discuss mainly on how CHs are selected. The control overhead for the creation and reconfigurations of clusters have not been considered completely. There have been not many papers that analyze analytically the control overhead incurred in hierarchical routing. Furthermore, the overhead is bound by a constant per vehicle per time step, avoiding expensive re-clustering chain reactions; hence, this overhead increases with the number of nodes. Since a CH acts as a coordinator in a cluster, if it is absent for any reason, the clustering architecture has to be reconfigured; this will significantly increase the message overhead.

Communication complexity represents the total amount of clustering-related message exchanged for the cluster formation. For clustering schemes with ripple effect, the communi‐ cation complexity for the re-clustering in the cluster maintenance phase may be the same as that in the cluster formation phase. But for those with no ripple effect, the communication complexity of re-clustering should be much lower. From analysis of different clustering protocols, we believe that a more efficient way to form a stable network structure, with reduced overhead, are that a vehicle should be associated to a cluster and not to a CH. Indeed, replacing CH is considered only as an incremental update and does not require a whole reconfiguration of the cluster structure; this will definitively increase the lifetime of the clustering architecture. The resulting clusters are stable and exhibit long average CM duration, long average CH duration, and low average rate of CH changes. The cluster creation and maintenance overhead should be calculated to be compared with non-clustering algorithms in terms of the reliability, fairness, and scalability of the algorithms. By optimizing cluster stability, cluster reconfigura‐ tion, number of clusters and cluster size can reduce the overhead caused in clustering.
