Improving TCP Performance: Identifying Corruption Based Packet Loss

Most TCP variants suffer a decrease in performance when used over networks that exhibit a high bit error rate, such as those that utilise wireless or satellite links. This is primarily due to TCP's inability to distinguish between congestion and corruption based packet loss. Instead, Jacobson based TCP variants assume that all loss is due to congestion. When packet loss occurs due to corruption the transmission rate is reduced in order to avoid the falsely identified congestion, rather than simply retransmitting the corrupted packets. In this paper we detail research undertaken to model network state, packet state and packet loss, allowing us to identify loss that is due to corruption. Furthermore, we propose an algorithm to identify the cause of packet loss on a probabilistic basis.