Implementation and performance evaluation of explicit congestion control algorithms

As the Internet adapts to meet consumer demands for high performance in ubiquitous networking environments, congestion must be addressed using novel approaches to ensure system stability over increasingly unreliable and dynamic media. Our work focuses on implementing and studying the performance of two recent explicit congestion control protocols: XCP, eXplicit Control Protocol, and RCP, Rate Control Protocol. This dissertation contains implementation details and subsequent test results of the XCP-b algorithm, a variant of XCP optimized for variablecapacity media, such as IEEE802.11. We show that, by inferring network capacity from queue dynamics, XCP-b proves more efficient than XCP when the link capacity is not explicitly known. Based on experimental feedback we present two alternative algorithms which further refine previous proposals. Furthermore, this dissertation documents the implementation of RCP, for which no formal specification yet exists, using the existing XCP protocol as a framework, as well as comparing both protocols by emphasizing their benefits. We explore the core algorithm in an RCP system and the implications aggressively distributing bandwidth has on overall system stability. By quantifying conservative bounds within which such a system can withstand surges in network traffic, we demonstrate how to accurately predict system response to common arrival distributions.