Starvation in Ad-hoc and Mesh Networks: Modeling, Solutions, and Measurements

In this thesis, starvation is defined as the phenomena where a flow receives throughput persistently and severely lower than its reference rate. Throughout the thesis, I investigate the cause of starvation, explore its solution space, and present mechanisms to counter starvation in both ad-hoc and mesh networks where CSMA protocols are deployed as the medium access mechanism. I first show that the root cause for starvation is incomplete and asymmetric channel state perceived by different stations even in very simple multi-hop scenario consisting of only two one-hop flows. Other factors that can contribute to starvation include closed-loop congestion control, multi-channel coordination, and imperfect clock synchronization. I analyze the effect of these factors on starvation by simulations, analysis and real world experiments. Motived by the analysis, I devise a protocol that utilizes multiple channels to counter starvation in ad-hoc networks, and propose a contention window policy to counter starvation in mesh networks.