Stochastic TCP Friendliness: Expanding the Design Space of TCP-Friendly Traffic Control Protocols

The Internet is increasingly being used to support UDP applications, especially UDP-based multimedia streaming applications. Ineffective traffic control for UDP traffic could lead to TCP starvation and congestion collapse in the Internet. In the last few years TCP friendliness has been proposed as a fundamental design principle for the control of UDP traffic in the Internet. The current notion of TCP friendliness is well suitable for applications that can quickly adjust their sending rate to react to network congestion. However, multimedia streaming applications usually have limited rate adaptability. Therefore, the current notion of TCP friendliness considerably restricts the design space of traffic control protocols for multimedia streaming and impairs the performance of multimedia streaming applications. This dissertation develops stochastic TCP friendliness as a new fundamental design principle for traffic control in the Internet to expand the design space of TCP-Friendly traffic control protocols. In the first part of this dissertation, we propose Stochastic TCP Friendliness based on the stochastic ordering theory. Stochastic TCP friendliness considers the statistical impact of UDP traffic on all competing TCP flows instead of the specific impact on each TCP flow. By doing so, it not only effectively maintains the desired stability and fairness of the Internet, but also expands the design space of traffic control protocols. Then we apply this framework to design an efficient traffic control protocol, named TCP-Friendly CBR-like Rate Control (TFCBR), for multimedia streaming applications that prefer a smooth sending rate on a multi-second timescale. TFCBR considerably improves the performance of both TCP applications and multimedia streaming applications when compared with existing TCP-friendly protocols in the sense that it achieves a smooth sending rate on a time scale of several seconds, and at the same time is stochastically TCP-friendly in all network environments. In the second part of the dissertation, we consider the fundamental performance tradeoff between throughput and smoothness timescales for TCP friendliness. Specifically, we derive the TCP-friendly sending rate requirements for three different smoothness timescales (i.e., milliseconds, seconds, and minutes) with flow-level queueing models and verify the analytical results with extensive packet-level simulations. Based on the analytical results, we propose a new stochastic TCP-friendly congestion control protocol, called Long-Time-Scale TCP-Friendly CBR-Like Rate Control (L-TFCBR), for applications that prefer a smooth sending rate on a multi-minute timescale. In the third part, we focus on TCP-friendly admission control in the Internet, which is considered as a better control scheme for streaming applications than congestion control. We develop an admission control protocol, called Stochastically TCP-Friendly Admission Control (STFAC), for multimedia streaming applications. STFAC can provide multimedia streaming applications with a constant-bit rate, and at the same time, provide all competing TCP flows with a higher average sending rate than current TCP-friendly congestion control protocols.