Summary: Measured Capacity of an Ethernet

D.R. Boggs, J.C. Mogul, C.A. Kent, Measured Capacity of an Ethernet: Myths and
Reality, ACM SIGCOMM 1988.

This paper makes the case that Ethernet has unjustifiably been given a bad
name by both faulty empirical testing and misunderstood analytic results. It
explains that most empirical tests ended up measuring either a limitation of
the host system or Ethernet implementation, rather than a true limitation of
the protocol. Analytic results tended to be misapplied because of
unrealistic "simplifying assumptions", and also simply because practitioners
misunderstood the applicability of the analytic results.

The authors provide a what appears to be a pretty thourough summary of the
then-current analytic results, along with some of the common pitfalls
practitioners fall into when interpreting the results. Next, they describe a
test environment carefully designed to separate the limitations of Ethernet
from the rest of the system. Using this environment, they measured a wide
variety of things while adjusting the packet size distribution, and the
number of hosts attached to the network.

They conclude by stating that the majority of problems seen with Ethernet are
due to faulty implementations or installations, and that the protocol is
capable of excellent performance for many tasks and under a wide variety of
operating conditions.

This paper appears to be an excellent blend: it has an excellent literature
review, a carefully designed experiment, a measurement section that explores
much of the parameter space, and clearly drawn conclusions. However, I
wonder how applicable this material is given that switched Ethernet is pretty
much the standard. Now that Ethernet commonly uses store-and-forward rather
than collision detection and retry, it would appear that many of the
conclusions are no longer valid.
Received on Mon Sep 18 2006 - 23:32:44 EDT