Scalable AQM Routers Supporting Hetergenous Traffic
University Of Massachusetts Amherst, Amherst MA
Investigators
Abstract
In the past decade the world has seen an explosion in Internet activity and with it has come increased expectations for its performance. Internet users now demand faster and higher-quality service for applications ranging from audio/video-on-demand to Internet telephony. At the heart of these information exchanges is the Transmission Control Protocol (TCP) and active queue management (AQM) schemes which work together with the goal of preventing network congestion and improving end-to-end performance. However, it is becoming increasingly evident that the present form of these schemes may not be able to cope with the growing demands on the Internet. It is well documented that congestion-avoidance schemes relying on Drop-Tail routers are prone to high-loss rates while AQM routers deploying random early detection (RED) are difficult to tune. Motivated by this situation, this proposal is concerned with the design of advanced AQM schemes. This projects approach departs from the prevailing techniques by explicitly relying on dynamic network models and feedback control principles. Central to the approach is the recognition that AQM schemes are feedback control systems and that feedback control principles provide essential tools for the analysis and design of AQM strategies. The absence of feedback control principles from the design scene so far is apparently due to a lack of an analytical model of TCP. Fortunately, this roadblock has been recently removed by one of the PIs through the introduction of a fluid-flow model that expresses TCP in a language that allows network control engineers to analyze and design AQM schemes. Indeed, in several recent papers the PIs have accomplished just that by: 1) relating key network parameters to the performance of AQM networks, 2) analyzing RED and suggesting parameter settings for stable queue management, and 3) introducing a new AQM scheme, the PI controller, that compares favorably with RED. The proposed research builds on these recent results and has two objectives. First, to study the interaction of heterogeneous fluid-flows with AQM routers and secondly, to investigate the scalability of PI controllers. The first objective is aimed at the recently developed TCP/AQM movel which assumes only long-lived flows and ignores short-lived flows. This research will develop models of heterogeneous flows and explore their impact on the AQM analysis and design. In the second objective the PI controller is considered which was originally designed and analyzed for a TCP connection encountering only a single bottleneck router. The scalability for the PI controller will be explored where complex network topologies, consisting of many routers each under local PI control, will be considered. Unlike the development of the original PI controller, this research will necessarily use, multivariable feedback control techniques to establish network stability, performance and robustness to network parameter variations.
View original record on NSF Award Search →