Adaptive large-scale wireless networks: measurements, protocol designs, and simulation studies

The popularity of wireless networks as a flexible and convenient last-mile solution for Internet connectivity is growing at a rapid pace. Users now depend on the availability of a wireless network at university and corporate campuses, city downtowns, and residential areas to complete their networking tasks because these networks provide mobility and flexibility for Internet access. So much is the dependence on wireless networks, users now also expect sufficient bandwidth and capacity for several bandwidth-intensive and delay-sensitive applications such as audio and multimedia. In response to these high user expectations and demands, wireless network companies such as Trapeze, Aruba, Cisco, Nortel, Tropos, Strix, and Symbol are rapidly deploying wireless networks that consist of hundreds to thousands of access points (APs) and routers. Moreover, redundant sets of APs and routers are deployed to provide users with extra bandwidth and capacity, and fault-tolerance. These APs and routers are provisioned for use by hundreds to thousands of users. Unfortunately, very little is known about the usage and behavior of large-scale networks under different user conditions. As a result, on one hand, when networks experience a flash-crowd and/or a high density of users, severe congestion occurs, leading to a huge performance degradation for users. Inordinate demands posed by users can even cause network meltdowns that prevents connectivity to all the users in the network. On the other hand, minimal usage of large-scale networks leads to significant energy wastage because a large number of network resources remain idle. As networks grow in size and number, energy wastage will become a serious concern. We argue that the primary cause of congestion, network meltdowns, and energy wastage is that wireless networks are not yet smart enough to modify their operation under different user conditions. We believe that if networks can adapt and protocols and management strategies can be retrofitted accordingly, these three challenges can be overcome. The objective of this PhD dissertation is to evolve large-scale wireless networks by, first, fully understanding their usage and behavior, and then, providing elegant and effective deployable solutions that allow networks to adapt to changing conditions and ensure good end-user performance. We propose the design of adaptive large-scale wireless networks that modify their operation based on the volume and location of user conditions estimated by them. The protocols and solutions we present bring about a significant paradigm shift from the traditional always on and always available networks. We use custom simulators and also propose realistic simulation models for popular simulators that facilitate the thorough evaluation of wireless networks before they are deployed. We also design and implement a novel health diagnosis framework that allows operators to identify faults in large-scale networks, once they are deployed. To demonstrate the feasibility of deploying our protocols and management solutions, we build and utilize network testbeds. This dissertation facilitates an "evolution cycle" for large-scale adaptive wireless networks. We believe that the solutions we present form a platform for future research initiatives that will help realize the successful operation of large-scale wireless networks of the future.