Enhancements of congestion controls for unicast and multicast over wireless networks

Transmission Control Protocol (TCP) is a connection-oriented, reliable and in-order transport layer protocol that serves most of the data transfers in the current Internet. However, running TCP over wireless networks is challenged by the problem of high bit error rate. The current legacy TCP, namely TCP Reno, is originally designed for the wired networks where packet loss is an indication of network congestion. TCP Reno slows down the sending rate in response to the packet loss in order to alleviate network congestion. But in the wireless networks, packet loss is often induced by high bit error rate (referred to as random loss) rather than congestion. Misinterpretation of random losses as congestion causes TCP Reno to reduce its sending rate unnecessarily, resulting in significant performance degradation. Such problem also exists in TCP-friendly multicast congestion control protocols, such as TCP-Friendly Multicast Congestion Control (TFMCC). TFMCC is a single-rate multicast protocol that makes use of TCP Reno's throughput equation to adjust its sending rate, and provide TCP-like long-term throughput for achieving TCP-friendliness. Moreover, TFMCC's sending rate is shaped to be smooth and responsive to the changes in network conditions, so that TFMCC is suitable for serving streaming multimedia applications such as IP TV. However, due to the nature of TCP Reno in lossy environment, TFMCC, equipped with the Reno equation, will inevitably suffer performance degradation from wireless links. This thesis studies the fundamental issue - congestion control - which is embedded in TCP and multicast.DOCTOR OF PHILOSOPHY (SCE