(no subject)

This paper tries to answer the question of how effective is the CSMA/
CD protocol as used in 10Mbit ethernet, seen both in terms of
throughput and delay under real-world usage scenarios. Since
ethernet was, and still is, an incredibly common LAN protocol, this
is a question of more than academic significance, and the paper
reflects this by providing practical advice as well as their results.

One of the paper's strengths is the extensive review of existing
work, although the authors place too much emphasis on prior studies
painting ethernet in a negative light and do not explain adequately
the prior work that supports their results. They do make a strong
argument, however, that the existing body of work is incomplete.

After describing the pitfalls of previous theoretical studies that
had questioned the effectiveness of ethernet, the authors described
their own real-world scenario for testing. The argument that their
network topologies are representative of real-world scenarios is
strengthened by their use of an existing laboratory network, however,
while they suggest that their distribution of packets in both in
terms of time and size is realistic, and it seems believable, they
don't provide enough evidence to back this up. I would be more
satisfied had they used actual real-world data, which assumedly could
have been harvested from the other DEC lab users, although its
possible that doing so would have imposed an excessive overhead on
the applications and users they were monitoring.

Overall, I believe the paper is effective in arguing for the
effectiveness of ethernet as a LAN protocol, although more discussion
of token-ring networks, for comparison purposes, would have been
valuable. Additionally, the paper is instructive in that it shows
how a range of different scenarios and assumptions significantly
affect one's results and how simple questions often have much more
complex answers than are initially expected.
Received on Mon Sep 18 2006 - 21:41:29 EDT