From: Kevin Wampler (wampler@cs.washington.edu)
Date: Tue Oct 19 2004 - 22:02:40 PDT
Motivated by indications that the congestion avoidance and control
mechanisms in TCP will cease to function effectively as both bandwidth and
latency increases, the paper "Congestion Control for High Bandwidth-Delay
Product Networks" presents a new protocol, XCP, to address these issues.
XCP has numerous advantages over TCP. It more aggressively adjusts to
take advantage of spare bandwidth in the network (using a MIMD approach),
and does so while providing greater stability even under conditions where
TCP does not perform well. XCP achieves this using a system where the
fairness and efficiency controllers in routers are decoupled. This noy
only gives greater modularity to the system, but allows the flow of
traffic to be controlled more precisely. The results of this, as shown in
the paper, are quite impressive.
If the results in the paper are indicative of XCP's real world
performance, then it seems that it would outperform TCP in almost all
respects. The primary problem then would be that XCP requires routers to
support new functionality. Furthermore, for XCP to be used all routers
along a path would have to support XCP. This may prevent XCP from giving
substantial benefits in a partial deployment, since any non-XCP router
along a path is sufficient to force the two endpoints to use a
conventional protocol. Furthermore, placing XCP functionality in the
routers does not conform to the end-to-end argument, although I personally
think that in the case of network congestion this is a worthwhile time to
put functionality in the middle.
This archive was generated by hypermail 2.1.6 : Tue Oct 19 2004 - 22:02:41 PDT