F-TCP: A Delay-Based Protocol with Fair Co-existence

Various studies have shown that regular TCP is inefficient in high-speed networks. This paper proposes F-TCP, a delay-based TCP variant, which is able to operate efficiently in high-speed networks. The slowstart phase of F-TCP continues until a threshold determined from probing the available bandwidth. When competing with loss-based flows F-TCP reduces it's window to a value derived from the available bandwidth. More specifically, an adaptive bandwidth share estimation with a delay-sensitive instability measure is employed to guide window backoff when congestion is detected by F-TCP. Using ns-2 simulations we show that F-TCP has good throughput efficiency, intra-protocol fairness and TCP friendliness properties. Our results also illustrate fair co-existence between a delay-based protocol (F-TCP) and a loss-based protocol (regular TCP) thus F-TCP maintains it's fair share of the link. In addition we show that F-TCP avoids self-induced packet losses by using delay as the congestion signal hence zero packet loss is experienced in all the simulations where all the flows are F-TCP. In simulations with regular TCP, packet losses are inevitable since it is loss-based. We therefore conclude that when treated carefully, delay (or RTT) can indeed be a useful congestion signal indicator. Bandwidth estimation is a good indication of available bandwidth for delay-based flows if made aware of the early congestion back off experienced by these flows.