Abstract
While the current definition of TCP friendliness has enabled a wide variety of traffic control protocols other than TCP, it still considerably restricts the design space of TCP-friendly traffic control protocols. For example, some multimedia streaming applications prefer a smooth sending rate on a time scale of minutes, however, a UDP flow maintaining a smooth sending rate on such a long time scale is naturally not TCP friendly by the current definition. In this paper, we first give a new class of TCP friendliness definitions, called stochastic TCP friendliness, which greatly expands the design space of TCP-friendly traffic control protocols, while still effectively maintaining the desired stability and fairness of the Internet. In particular, we propose stochastic TCP friendliness in usual stochastic order as a more appropriate design guideline for traffic control protocols, which intuitively ensures that a UDP flow is friendly to all competing TCP flows with any increasing utility function. Second, we develop a congestion control protocol, called TCP-friendly CBR-like rate control, which achieves a smooth sending rate on a time scale of minutes, and at the same time is stochastically TCP friendly in usual stochastic order in most network environments.