New Algorithm to Control TCP Behavior over Lossy Links

TCP (Transmission Control Protocol) is one of the most used  transport protocol in the Internet. Many widely used applications use TCP for sending data like the File Transfer Protocol (FTP), Telnet and Web-HTTP  connections. TCP was found to be performing poorly over lossy links when transmission (non-congestion) errors existed. This is due to the fact that TCP is unable to distinguish between congestion and transmission errors, and hence reduces its sending rate for all errors, assuming a congestion exists in the connection path. One solution is to discriminate between errors and deal with each error type differently. In this paper we present a new TCP congestion window cut algorithm to be used by any end-to-end TCP error discriminator when transmission errors occur. Instead of cutting the congestion window to half (like standard TCP) we delay the cut decision until TCP receive all duplicate acknowledgment for a given window of data (packets on flight). This will give TCP a clear image about the number of drops from this window. Then congestion window size is reduced only by number of dropped packets. Using this approach, the new algorithm managed to improve TCP performance in non-aggressive way by increasing the average congestion window size. On the other hand, this algorithm cut the congestion window for both congestion and non-congestion errors which help to reduce the effect of error mismatch on the network. Simulation results shows noticeable improvement when the new technique is added to an error discriminator. Also we discusses some limitations of the new algorithm.