NeTS-NBD: Illuminating Congestion Control: Analytical Guidance and Practical Implementations

点击翻译按钮获取中文摘要

Effective methods of network congestion control and avoidance are important for the successful co-existence of different applications running on a time-varying network with shared resources such as the current Internet. In particular, congestion control and avoidance methods govern the means by which an application sending packets across a network adjusts its sending bit-rate in order to avoid or react to network congestion indicators in the form of lost packets, increased packet transport delay, or explicit congestion warnings from internal network routers. Clearly, with different applications all sharing the same network resources, congestion control methods that allow for the efficient and fair allocation of network resources without requiring a centralized authority for determining the allocations are needed. This research takes a new approach to the analysis and design of congestion control. Through the use of network calculus notions, which do not impose overly restrictive modeling assumptions on traffic and service endemic in stochastic modeling approaches, this research will provide new analytical guidance and practical methods for congestion control. Through the use of recently developed methods for handling feedback with time-variant delays, optimal congestion controllers for different delay-based congestion avoidance control objectives can be derived. Preliminary results indicate that an approximation of a certain optimal non-causal controller provides better throughput or fairness than any other existing delay-based scheme. Building upon this preliminary work, the investigators will examine the basic problem of delay-based congestion avoidance for both window and variable bit-rate controllers, providing analytical guidance in the form of optimal controllers that can be approximated, leading to practical implementations. In addition, the research will explore the joint use of both delay and packet loss/marking feedback together within a congestion control context. This will ultimately lead to the consideration of the joint design of congestion window controllers and active queue management methods within the network calculus framework. As the network calculus based controllers must be approximated, the issue of control parameter estimation will be considered within a statistical signal processing context. Broader Impact: This research will be performed at the University of Miami, which is recognized as a Hispanic Serving Institution and Postsecondary Minority Institution, meaning that the investigators will be able to involve underrepresented minorities in their efforts to integrate research and education.