From: Daniel Lowd (lowd@cs.washington.edu)
Date: Sun Oct 17 2004 - 22:56:10 PDT
This paper discussed a variety of techniques that the authors had found
useful for dealing with congestion on the internet. These techniques are
simple, effective, and easily deployed since they depend only on the
endpoints. I thought this paper was a good demonstration of how the
end-to-end argument can be applied to congestion control, though the
author also discusses using some of these techniques in gateways.
This paper was quite readable, and the intuitions were easy to follow.
From time to time, there would be an allusion to queuing theory or other
mathematical analysis. The biggest weakness was the lack of experimental
evaluation. The graphs included were informative, but seemed anecdotal
more than analytical. There was also no attempt to compare methods
against anything more than the simplest baseline. I would have been
interested to see results of various simulations, along with lesion
studies to show which modifications were helpful most often.
The casual style that contributed to its readability also felt random or
unprofessional at times. It wandered between mathematical specifics,
coarse generalizations, specific arguments, and informal asides (footnote
12: "/did/ lead."). Page 322 uses the word "insure" when the author means
"ensure." Figures were often larger than necessary, with paragraphs and
paragraphs of caption, and even longer footnotes (e.g., on page 321, most
of the words are in footnotes!).
Even so, I think this paper is useful background for its effective, simple
techniques that work in practice. It is useful for system designers and
implementers in a way that most academic papers are not. For a field that
depends on practice as much as theory, this is a great advantage.
-- Daniel
This archive was generated by hypermail 2.1.6 : Sun Oct 17 2004 - 22:56:11 PDT