From: Charles Reis (creis@cs.washington.edu)
Date: Tue Oct 19 2004 - 23:58:38 PDT
Congestion Control for High Bandwidth-Delay Product Networks
Katabi, Handley, Rohrs, 2002
The paper addresses the problem that TCP becomes unstable and inefficient on networks with very high bandwidth or very high delay, both of which are becoming more common in the Internet (e.g. between fiber and wireless connections). The authors propose a new protocol, XCP, to resolve this issue and find it has many additional benefits as well.
The authors do a very impressive job designing and backing up XCP's mechanisms using control theory, and then showing that it clearly outperforms several TCP variants in simulation. In addition to being very efficient, keeping queues small, and virtually eliminating dropped packets via explicit congestion feedback, XCP also proves to be more flexible by allowing efficiency and fairness to be controlled and adjusted with different mechanisms. This is all accomplished without introducing state in the core of the network and with potentially feasible paths for deployment.
Despite all its benefits, it is still a proposal for a major new protocol, which may end up being difficult to deploy for practical reasons. Additionally, more analysis of the security model (police agents at the edges) might be necessary-- senders might easily manipulate the packet congestion headers, given potential monetary incentives to cheat with price-differentiated service. Nevertheless, XCP sounds quite promising, and appears to be making progress towards deployment with the publication of an Internet Draft earlier this week.
This archive was generated by hypermail 2.1.6 : Tue Oct 19 2004 - 23:58:39 PDT