Congestion state-based dynamic FEC algorithm for media friendly transport layer

Abstract

TCP-friendliness has been adopted as the most important property for the design of new media-specific transport layers in the Internet. The TCP protocol is mainly concerned with achieving as much throughput as possible while preventing long-term congestion. Various TCP protocol designs do this by inducing brief episodes of network congestion, measuring it, then reducing the offered load quickly to remove the congestion. Media flows, on the other hand, are very sensitive to even brief episodes of congestion. The question therefore arises: how can we protect media flows against TCP-induced network congestion? In this paper, we focus on combining the TCP Friendly Rate Control (TFRC) protocol with Forward Error Correction (FEC) to achieve such protection. We observe that FEC methods that solely rely on loss statistics generate significant overhead in terms of the redundant parity packets transmitted over the network. Accordingly, we investigate the loss and delay characteristics in several TCP-induced congestion scenarios in order to identify potential periods of increased congestion and apply FEC protection during those periods judiciously. We find out that indeed efficient models can be developed and incorporated into a dynamic FEC framework which can achieve substantially better overhead vs. reliability tradeoff (e.g., up to 60% improvement at high reliability region) than an FEC approach that uses fixed coding rate to satisfy a given reliability.