TCP friendliness has been proposed and used as a fundamental design principle for traffic control in the best-effort Internet in the last few years. The current notion of TCP friendliness is well suitable for applications that can adapt their rates to network congestion. However, a significant fraction of Internet traffic is multimedia streaming traffic, which inherently has limited rate adaptability. As a result, the current notion of TCP friendliness considerably restricts the design space of traffic control protocols for multimedia streaming.
The goal of this project is to develop stochastic TCP-friendliness as a new fundamental design principle for traffic control in the best-effort Internet with the following two major research components: 1) developing a stochastic TCP-friendliness framework to qualitatively measure the statistical impact of a traffic control protocol on TCP performance, and to relatively compare the statistical impacts of two different protocols and 2) designing and evaluating a new class of stochastically TCP-friendly traffic control protocols for multimedia streaming, with the aim to considerably improve both TCP and UDP performance when compared with existing TCP-friendly protocols.
Broader Impacts: Stochastic TCP-friendliness as a new fundamental design principle will enable other researchers to develop many other new protocols better tailored to their specific needs. It is an important step on the path to move the Internet from a world dominated by TCP users to a world shared among diverse users. This project also includes a strong education component to attract more undergraduate students, and to engage both undergraduate and graduate students into research.
