Multicast flow and congestion control over combined wired/wireless networks.

摘要

Multipoint multimedia communications are becoming more and more important in today's Internet environment as well as in more futuristic mixed wired/wireless network environments. To deliver multicast information over networks which are designed primarily for point-to-point communications poses interesting management and control issues that need to be resolved. Flow and congestion control is one such challenging problem.; In this dissertation, we propose a unified framework for the design and performance analysis of multicast flow control algorithms, as well as some possible source policies, such as listening to the slowest request, random listening, weighted sum, estimation/prediction, etc.; Given the complexity of the problem, we first address the simplest model with two receivers using binary on-off control. Three ad hoc algorithms are proposed and studied. These include Listen to Slowest Request (LSQ), Source Estimation (SE) and Open Loop Control (OLC). Our simulation shows that LSQ, or SE, or OLC, works the best under different situations. A fluid flow technique is used to analyze the performance of the algorithms assuming zero propagation delay. The analytical result helps to provide insight into the simulation result.; We then proceed to examine more complicated control algorithms. We apply the random listening idea to a TCP-like window-based scheme to come up with a multicast algorithm which we prove achieves bounded fairness to TCP traffic under a restricted topology in the current Internet infrastructure. Our simulation shows that the algorithm achieves good performance under various network topologies.; Finally, we study a rate-based multicast flow control scheme with a weighted sum source policy, which is an extension of a previously proposed unicast algorithm. The stability of our multicast algorithm is analyzed using delay-differential equations. The algorithm is analytically proven to be stable, which is a breakthrough in performance analysis of rate-based multicast algorithms.; Our work suggests that different situations may require different source policies to deliver the desired performance. The LSQ policy, which is a common practice in today's multicast algorithms, is not always the best solution. Non-LSQ algorithms are of both theoretical and practical interest to study.