ITR: Design and Analysis of End-to-End Feedback Congestion Control Schemes

最近，改进互联网拥塞控制技术的需求日益增长。非实时数据流量目前是在“尽力而为”的基础上传输的，为了满足关键的科学、企业和电子商务应用，它将越来越具有更严格的延迟和吞吐量要求。为了满足这种增强的服务质量要求，需要理解包含在传输控制协议(TCP)中的现有反馈拥塞控制方案。为此，PIS打算：1)建立一个分析模型，该模型可以包含IETF正在考虑的TCPReno和TCPEnhance的主要功能。这里提供的PI模型很有趣，因为它可以容纳多个TCP流，可能还可以容纳多个网络节点。2)将TCP置于控制理论框架中，以便很好地理解其稳定性和暂态行为。这里提出的非线性控制技术的使用是新颖的，并且应该提供一组新的数学工具来研究这个问题。3)这两个活动的综合将反过来提出方法，以确保任何建议的TCP后继者是稳定的，以及提高吞吐量和公平性。一个更有限但可能会产生更直接影响的目标是更好地了解TCP Reno的稳定性和性能。所有这些工作都将在正在进行的补充拥塞控制工作中完成，例如使用多协议标签交换(MPLS)协议的流量工程、流量整形和策略、服务调度和缓冲区管理。除了对技术现状的调查外，PI还提供了一些初步结果和未来研究的想法。PI演示了单节点情况下的TCP Reno的分析模型，以及即使是这样一个简单的模型也能提供的一些见解。PI还提供了一种方法，将最近在流量控制中的控制理论工作扩展到更现实和更一般的环境，其中边界效应和未知的、时变的传播延迟出现在网络中。李亚普诺夫理论、泛函(延迟)微分方程理论和现代非线性控制中的构造性设计方法将在新的互联网反馈拥塞控制方案的合成中发挥关键作用。这项研究的更广泛影响包括发展控制理论和网络研究社区之间更紧密的互动，研究生和本科生水平的课程改进，通过持续的行业互动应用这项工作的一些实际意义，以及对IETF小组的投入。

The need to improve congestion control techniques for the Internet has grown recently. Non real-time data traffic, which is currently transported on a "best effort" basis, will increasingly have more stringent delay and throughput requirements to meet critical scientific, corporate and e-commerce applications. In order to meet this enhanced quality of service requirement, the existing feedback congestion control scheme incorporated in the Transmission Control Protocol (TCP) needs to be understood. For this purpose, the PIs intend to:1) Build an analytical model that can incorporate key features of TCP Reno and TCP enhancements under consideration by the IETF. The model the PIs present here is interesting in that it can accommodate multiple TCP flows and possibly multiple network nodes.2) Place TCP in a control-theoretic framework so that its stability and transient behavior is well understood. The use of non-linear control techniques proposed here is novel, and should make available a new set of mathematical tools to study this problem.3) The synthesis of these two activities will in turn suggest methods that ensure that any proposed TCP successor is stable, as well as improving throughput and fairness. A more limited goal but with perhaps more immediate impact is a better understanding of the stability and performance of TCP Reno.All of this work will be performed within the complementary congestion control efforts underway such as traffic engineering using the Multiprotocol Label Switching (MPLS) protocol, traffic shaping and policing, service scheduling and buffer management.In addition to a survey of the state of the art, the PIs present some preliminary results and ideas for future research. The PIs demonstrate an analytical model of TCP Reno for the single node case, and some of the insights that even such a simple model provides. The PIs also present a way of extending recent control theoretic work in flow control to a more realistic and general context, where boundary effects and unknown, time-varying propagation delays appear in networks. Lyapunov theory, the theory of functional (retarded) differential equations and constructive design methods in modern nonlinear control will play a key role in the synthesis of new feedback congestion control schemes for the Internet.The broader impacts of this research include the development of closer interaction between the control theory and networks research communities, curriculum enhancement at the graduate and undergraduate levels, applying some of the practical implications of this work through ongoing industry interactions, and inputs to IETF groups.