Performance Optimization of Wireless Mesh Networks

Abstract

Wireless mesh network (WMN) is a communication network consisting of radio nodes organized in a mesh topology. The components of a mesh network are mesh clients, mesh routers for forwarding packets to mesh gateways that connect to the internet. WMNs can be integrated with wired and wireless networks. The wireless networks can be infrastructure based networks such as 802.11 and cellular networks or mobile ad hoc networks. WMNs thus provide a low-cost platform for offering ubiquitous broadband access ranging from local area, metropolitan area and wide area networks. WMNs have interesting properties like self organization and self healing and offer higher reliability. However, there are significant challenges that influence the architecture, design, and deployment of WMNs. Clever algorithms and protocols are needed to realize the true potential of WMNs to address performance, mobility and security issues. WMNs have applications ranging from civilian wireless internet applications to tactical and emergency response applications

In this thesis, we focus on fairness and scalability issues in WMNs and propose solutions to optimize network performance. Two ideas are proposed as a solution to the above problems. First, a differentiated client algorithm is proposed as a potential solution to the scalability problem. In this algorithm the clients are differentiated based on the amount of traffic they generate. It is shown through NS-2 simulations that there is significant improvement in scalability especially when the network is congested. Second, a gateway classification algorithm (GCA) is proposed to solve the throughput unfairness issue. In general, WMNs are multi-hop networks. The unfairness problem due to the fact that clients (and routers) that are nearer (smaller number of hops from) the mesh gateways may have undue advantage over those that are farther. In this algorithm, the gateways are classified based on their data rate handling capabilities. Extensive NS2 simulations demonstrate the effectiveness of the above two algorithms.