Evaluating the efficiency of clustering on routing and network performance Imen Jemili 1, 2 , Abdelfettah Belghith 1 , and Mohamed Mosbah 2 1 HANA Research Group, National School of Computer Sciences University of Manouba,,Tunisia 2 Laboratoire Bordelais de Recherche en Informatique, Université de BordeauxI - ENSEIRB 351, cours de la Libération, F-33405 Talence cedex Abstract - The intrinsic characteristics of ad hoc networks, such as the frequent connectivity changes and the strict bandwidth and power constraints, impose further challenges, especially for routing tasks. Establishing and maintaining routes are hard tasks in a dynamic environment, in particular for large and dense networks where the control overhead induced by routing packets increases with the number of nodes. To overcome the scalability problem, creating hierarchies among the nodes seems a promising venue and an effective approach to organize the network as the number of nodes increases. In this paper, we investigate the influence of the underlying clustering algorithm on the quality of the routing and illustrate through conducted simulations the gain in network total throughput that can be attained using a clustering scheme that does not rely on periodic exchange of neighborhood information. Keywords: Clustering, connected dominating set, MAC layer, ad hoc networks. 1 Introduction The main attractive feature of ad hoc networks is their ability to provide a fast deployable multi-hop wireless infrastructure for a growing number of applications when a wired network is neither available nor economical to build. However, the intrinsic characteristics of these networks, such as the frequent connectivity changes and the strict bandwidth and power constraints, impose further challenges. Many researches were conducted in order to try to bring solution to the various problems confronting the ad-hoc network community such as routing, quality of service, energy consumption, security, and so on [1]. In fact, in an ad hoc network, mobile hosts communicate over a shared wireless channel. A direct communication session is achieved only when the communicating nodes are close enough. Otherwise, communication between any two hosts will be through multi hop wireless links involving intermediate nodes to relay messages. Thus, all hosts collaborate among themselves to achieve functionality usually provided by the network infrastructure (like routing) and to create a sort of a virtual infrastructure to support efficient communications between nodes and enhance network quality of service. Besides, the absence of a wired backbone or any sort of centralized network management system leads indeed to the problem of keeping track of the topology connectivity and maintaining routes as hosts move or turn off or on. In order to be able to act cooperatively to route traffic, mobile nodes must be aware of what is happening around them. They must also be able to adapt the network to the dynamic state of its links and its mobility patterns. The mobility of nodes and the high rate of link failure put special challenges in routing in ad hoc networks. However, existing routing algorithms devoted to ad hoc networks and based on proactive or reactive schemes suffers from scalability. First, all nodes have egalitarian role. In fact, every node has to participate during discovering phase, forwarding data packets and during the maintenance of paths due to the eventual displacements of nodes involved in the transfer of packets. These additional responsibilities contribute to deplete more quickly the energy resource of nodes. Moreover, most existing well-known routing protocols resort to flooding for route construction. The proactive scheme tries to maintain a routing table up-to-date. It is kept actualized thanks to a periodic exchange of control packets. This communication overhead exponentially increases with the total number of mobile nodes of the network, consuming consequently a large proportion of the bandwidth. In a dynamic environment, the frequency of routing table exchange between neighboring nodes increases to reflect the changes in the network topology over time. Thus, maintaining topology information up to date incurs excessive overhead which impairs network scalability, particularly for large-scale networks. On the other hand, for reactive schemes, no permanent routing information is maintained; a discovering phase is started when a route is needed. The route request flooding incurs considerable route setup delay and large amount of communication overheads as the network size increases. Moreover, the path failure due the eventual node mobility triggers a new discovering phase. Due to these intrinsic