Origins and Solutions of Proportional Unfairness in IEEE 802.11 Multi-hop Wireless Networks

Proportional unfairness exists in multi-hop wireless networks where IEEE 802.11 is employed as the medium access control protocol. This thesis investigates the origins of unfairness and proposes solutions to improve fairness in such networks. This thesis first studies multi-hop topologies consisting of two disjoint flows and four nodes. All possible topologies depending on whether the different nodes are within range of each other are systematically identified and further classified into three groups. We show that in the first group the two senders are in radio range of each other and contend fairly. However, long-term unfairness arises in the second group where the two senders do not have complete information of the channel and the topology is asymmetric. In the third group where the channel information each flow obtains is incomplete but the topology is symmetric, the system achieves long-term fairness yet incurs short-term unfairness. We develop analytical models to characterize the MAC performance in the second and third group. We then analyze the compounding effect of MAC and congestion control on fairness. We identify an important topology that is necessarily embedded in mesh networks, in which a one-hop TCP flow contends with a two-hop TCP flow for gateway access. We describe how the congestion control and the MAC jointly lead to unfairness between the two flows, and analytically model the system with a six-dimensional Markov chain model. Motivated by the model, we devise and evaluate a simple Contention Window Policy that only requires the one-hop mesh nodes to increase their minimum contention window. Extensive experiments and simulations show that regardless of its simplicity, this window policy has a powerful effect in improving fairness of the network. To improve fairness in mobile ad-hoc networks, we devise a distributed multi-channel medium access protocol, Asynchronous Multichannel Coordination Protocol (AMCP), that not only increases aggregate throughput, but more importantly, addresses the fundamental coordination problems that arise in a multi-hop network. We analytically derive and experimentally validate a lower throughput bound for any flow in an arbitrary topology under the operation of AMCP.