Receiver-assisted recovery mechanism for TCP after long channel outage

Abstract

We propose a rapid-recovery algorithm for use at the TCP layer for a bursty packet loss channel in a server-client model. Bursty packet loss caused by a channel outage will be a significant problem in the communication systems now being developed because TCP applications, especially after experiencing bursty packet loss, will take a much longer time to recover. Our algorithm, which needs to be installed only in a client, includes transmission of pseudo-acknowledgement packets and adjustment of the advertised receive window (rwin) size as soon as the loss ends. The algorithm works with a standard TCP-Reno version with a SACK option used in both the server and the client. While the standard TCP takes at least a few minutes to recover after a 0.5-s outage - because of the native slow-start algorithm on a wireless link with bursty 1-Mbps traffic - a TCP with the proposed algorithm can resume transmission immediately after the client learns the channel has recovered.