TCP bulk repeat

TCP is the most widely used transport layer protocol. However, it was designed for wired networks and suffers performance degradation in error-prone wireless networks. When multiple error losses occur in the same window, existing NewReno-based TCP variants need one RTT to retransmit each lost packet, an inefficient recover strategy. Exponential retransmit backoff and frequent window reduction also lead to TCP performance degradation. In this paper we propose three end-to-end, sender-side-only changes, collectively called as Bulk Repeat (BR), to improve TCP performance in the presence of heavy errors. An analytic model is developed and validated with the Ns-2 simulator. We implement TCP BR changes on top of TCP Westwood (TCPW) in Ns-2 and study its throughput performance as well as fairness and friendliness via simulations. Results show that TCP BR can improve TCP performance up to an order of magnitude in error-prone wireless networks.