The Case for Streaming Multimedia with TCP

摘要

In this paper, we revisit and challenge the dogma that TCP is an undesirable choice for streaming multimedia, video in particular. For some time, the common view held that neither TCP nor UDP, the Internet's main transport protocols, are adequate for video applications. UDP's service model doesn't provide enough support to the application while TCP's provides too much. Consequently, numerous research works proposed new transport protocols with alternate service-models as more suitable for video. For example, such service models might provide higher reliability than UDP but not the full-reliability of TCP. More recently, study of Internet dynamics has shown that TCP's stature as the predominant protocol persists. Through some combination of accident and design, TCP's congestion avoidance mechanism seems essential to the Internet's scalability and stability. Research on modeling TCP dynamics in order to effectively define the notion of TCP-friendly congestion avoidance is very active. Meanwhile, proposals for video-oriented transport protocols continue to appear, but they now generally include TCP-friendly congestion avoidance. Our concern is over the marginal benefit of changing TCP's service model, given the presence of congestion avoidance. As a position paper, our contribution will not be in the form of final answers, but our hope is to convince the reader of the merit in re-examining the question: do applications need a replacement for TCP in order to do streaming video?