From: Alan L. Liu (aliu@cs.washington.edu)
Date: Sun Oct 17 2004 - 22:35:11 PDT
The purpose of this paper was to present slow-start and congestion
avoidance to alleviate severe network congestion caused by TCP.
I can only imagine that back when the paper was written, congestion was
a known problem that people had encountered. However, the strength of
the paper was in describing the situation, using flow, in a way that
made analyzing congestion more concrete. After doing this, it makes
sense that congestion is a problem caused by introduced packets beyond
the limits of the pipes. The notion of a "conservation of packets" makes
intuitive sense, but again, before this paper, who knows what
assumptions people had about the causes of congestion and working
methods for congestion control.
One assumption that the paper makes is that damaged packets are much
rarer than dropped packets, with no empirical evidence to back it up.
Another strange problem that appears occasionally in the paper are weak
or non-existent justifications for some of the claims, such as what the
appropriate increase to the window size should be after a timeout.
One way the paper could have been strengthened is to include more data
on the techniques introduced. One small trace of a few hosts does not
even cover all the cases that must have existed at the time, let alone
the possibilities of future loads on the Internet. There should have
been consideration for what cases might expose weaknesses in the
algorithms. Instead, the paper makes claims such as how slow-start is
"quick enough" to have a "neglible effect on performance," as though
slow-start isn't an issue when it seems clear that short TCP transfers
suffer greatly.
One acknowledgement of a deficiency in the work is that the algorithms
presented only cover the end-hosts, and if certain hosts don't play
fair, there's nothing that can be done. So a line of future work would
be in how to support fair sharing using gateways.
This archive was generated by hypermail 2.1.6 : Sun Oct 17 2004 - 22:35:13 PDT