Making CSMA/CA optimal in wireless multi-hop networks

Abstract

Over the last decade, significant attentions have been paid to the study of distributed scheduling algorithms in wireless multi-hop networks, with provable performance guarantee in terms of throughput and utility. The main results there can be summarized: first, there exists a tradeoff between complexity and performance, and second, most algorithms rely on heavy message passing for optimal performance guarantee. Despite the long history and great importance of such distributed algorithms, the question of designing a fully distributed contention resolution protocol has remained unresolved. As a promising response, recent theoretical researches have shown that carrier sense multiple access (CSMA) can be designed to achieve throughput- and utility-optimality without any control message exchange, i.e., in a completely distributed way. We refer to such a class of algorithms as {\em optimal CSMA}. However, such advances in the theory domain cannot guarantee the performance in the practice domain, which naturally motivates the need for the validation of theory through real implementations (i.e., experimental evaluation). Thus, we focus on the practicality of the theory-driven algorithm by implementing such a theory-driven algorithm since theory by nature entails many assumptions for its mathematical tractability, which may lead to the fundamental gap between theory and practice. In this dissertation, we first implement the optimal CSMA algorithm with conventional hardware and open device driver, and then extensively evaluate its performance using systematic decoupling approach in terms of topology, channel, and congestion control in the practical setting. By analyzing the experimental results, we investigate the gaps between theory and practice and how to bridge them without hurting guaranteed performance of the original algorithm. Next, we develop the {\em enhanced practical version} of the theory-driven algorithm given the three challenging conditions: (i) over le...