Hi, I've put a linux patch at www.hamilton.ie/net that might be of interest to those working on tcp for high-speed networks. One of the issues that has become pretty clear from initial tests of new tcp congestion control algorithms in high-speed regimes is that the standard linux tcp stack often performs poorly due to implementation problems. This was one of the main themes at the recent PFLDnet workshop at Argonne. Sack processing is very inefficient, leading to frequent timeouts, the burst moderation implementation often does more harm than good and packet reordering etc (which seems to be quite common on real networks) degrades throughput. I've tried to document some of these problems in the report at http://www.hamilton.ie/net/LinuxHighSpeed.pdf. The problems are sufficiently serious that they frequently obstruct comparisons and evaluation of proposed tcp congestion control algorithms. The patch at http://www.hamilton.ie/net/2.4.23.HSv4.patch.gz tries to address these problems, and seems to work fairly well in the tests I've carried out on a variety of machines and network paths. Its still very much experimental code, however, and comments would be appreciated. The code is a patch against linux 2.4.23 and includes web100 (with implementations of scalable cp and high-speed tcp) and also an implementation of htcp. Doug Hamilton Institute www.hamilton.ie - : send the line "unsubscribe linux-net" in the body of a message to majordomo@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html
