F-RTO

Abstract

Spurious TCP retransmission timeouts (RTOs) have been reported to be a problem on network paths involving links that are prone to sudden delays due to various reasons. Especially many wireless network technologies contain such links. Spurious retransmission timeouts often cause unnecessary retransmission of several segments, which is harmful for TCP performance. Recent proposals for avoiding unnecessary retransmissions after a spurious RTO require use of TCP options which must be implemented and enabled at both ends of teh connection. We introduce a new TCP sender algorithm for recovery after a retransmission timeout and show that unnecessary retransmissions after a spurious retransmission timeout, improving the TCP performance considerably. The algorithm is friendly towards other TCP connections, because it follows the congestion control principles and injects packets to the network at same rate as a conventional TCP sender. We implemented the algorithm and compared its performance to conventional TCP and Eifel TCP when RTOs occurred either due to sudden delays or due to packet losses. The results show that our algorithm either improves performance or gives similar througput as the other TCP variants evaluated in different test cases.