From: Ioannis Giotis (giotis@cs.washington.edu)
Date: Wed Oct 20 2004 - 01:29:19 PDT
Congestion Control for High Bandwidth-Delay Product Networks
This paper presents a new transport protocol, that aims to deal with
problems occurring from the latest internet trends, high bandwidth large
delayed links in particular.
XCP's main difference from TCP is that it injects congestion state
information in the packets, essentially making the hosts respond in advance
to congestion problems. This allows simpler router implementations and more
importanly better congestion control when dealing with high delay links. The
authors present their case both analytically and with simulations. The
authors present a strong case and proove that XCP is better than TCP when
dealing with these networks.
The question however is if those networks reflect future internet. It seems
that both their analytical and simulation results assume some kind of
uniform distribution on the data flows. While probably closer to the truth
than the simulations done by TCP queueing control systems, scenarios which
would make XCP perform bad still exist, and one cannot be sure if such
scenarios will not actually be met.
Overall, is XCP better than congestion management at the routers ? Who
knows, I believe that only extensive analysis of real internet traffic would
show that. But even that is obsolete as internet is changing so rapidly...
This archive was generated by hypermail 2.1.6 : Wed Oct 20 2004 - 01:29:18 PDT