Clustering algorithms for ad hoc wireless networks

Abstract

An ad hoc network is a multihop wireless communication network supporting mobile users without any existing infrastructure. To become commercially successful, the technology must allow networks to support many users. A complication is that addressing and routing in ad hoc networks does not scale up as easily as in the Internet. By introducing hierarchical addresses to ad hoc networks, we can effectively address this complication. Clustering provides a method to build and maintain hierarchical addresses in ad hoc networks. Here, we survey several clustering algorithms, concentrating on those that are based on graph domination. In addition, we describe results that show that building clustered hierarchies is affordable and that clustering algorithms can also be used to build virtual backbones to enhance network quality of service. 1. Introduction. In a speculative paper, Kleinrock [32] described ad hoc net-working technology as a blend of nomadicity, embeddedness, and ubiquity. In a network of the future, users and computing devices will be able to connect to such a network conveniently and even transparently. Computing and communication ca-pabilities will not only be restricted to standard electronic devices, but every gadget can afford to embed a considerable amount of intelligence. On a global basis, devices in the network will be able to rely on other devices to relay packets for them if neces-sary. The entire world will be heterogeneously networked by a vast " invisible global infrastructure " . The idea of ad hoc networking has been around for over 30 years. As early as 1972, DARPA started the pioneering PRNet (Packet Radio Network) project [31]. Subsequently, various projects sponsored by the military, such as SURAN (Surviv-able Radio Networks), TI (Tactical Internet), and GloMo (Global Mobile Information Systems), were launched to implement the ad hoc networking paradigm [20]. In the meantime, many enabling technologies, such as wireless signal processing and encoding, distributed computing, VLSI circuit design and manufacturing, cryptogra-phy, positioning services, et al. have been invented and developed that can address various problems confronting the ad hoc network community. Given the successful commercial use of the Internet, one cannot help asking why there are no cost-effective off-the-shelf commercial ad hoc networking systems. Among the many challenges for ad hoc network designers and users, scalability is a critical issue. In particular, when a flat-topology network contains a large number of nodes, control overhead, such as routing packets, requires a large percentage of the limited wireless bandwidth. A technology can be sustainably viable only if it can find widespread use. In order to allow ad hoc networks to achieve commercial success, we must solve the scalability problem. One promising approach is to build hierarchies among the nodes, such that the network topology can be abstracted. This process is commonly referred to as clustering and the substructures that are collapsed in higher levels are called clusters. In this chapter, we first explain why scalability is a hindrance for ad hoc networks and why the scaling techniques used successfully by the Internet are not directly applicable. We then survey some of the clustering algorithms for building network hierarchies. Finally, we consider the costs associated with using clusters in hierarchical