TCP Westwood: congestion control with faster recovery

In this paper we propose a new version of the TCP protocol, called TCP Westwood, which enhances the performance of TCP window congestion control by using as feedback the end-to-end measurement of the bandwidth available along a TCP connection. The available bandwidth is estimated at the TCP source by measuring and low-pass filtering the returning rate of acknowledgments. The estimated bandwidth is then used to properly set the congestion window and the slow start threshold after a congestion episode, that is after a timeout or 3 duplicate acknowledgments. The rationale of this strategy is simple: TCP Westwood sets a slow start threshold and a congestion window which are consistent with the network capacity measured at the time congestion is experienced. In particular, TCP Westwood introduces a mechanism of faster recovery to avoid overly conservative reduction of the congestion window after a congestion episode by taking into account the end-to-end estimation of available bandwidth. The advantage of the proposed mechanism is that the TCP sender recovers faster after losses especially over connections with large round trip times, or running over wireless links where sporadic losses are due to unreliable links rather than congestion. The proposed modifications follow the end-to-end design principle of TCP. They require only slight modifications at the sender side and are backward-compatible. Simulation results show a considerable throughput increment in comparison with TCP Reno and TCP SACK over wired networks and even more over wireless networks.