Myths and Reality. Di Niu

This is a well-written and well-organized paper which helps people
increase their understanding into the capacity of an Ethernet
greatly. The paper has at least two strengths. First, it summaries
and clarifies most of the important theoretical studies pertaining to
the capacity of Ethernet. Second, it employs an appropriate
measurement methodology and covers a wide range of realistic
settings, and thus provides further insights into Ethernet as
compared to the previous theoretical studies.

As a highlight of the paper, it undermines the previously well-known
myth produced by the theory that an Ethernet is saturated at an
offered load of 37%. According to the author, this is because in
theoretical studies, the worst-case of a minimal packet size is
frequently assumed, while in reality packets lengths, host number
and cable lengths are highly heterogeneous. Specifically, new results
different from the previous theoretical studies are obtained, with
respect to different scenarios such as fixed length packets on a long
or a short net, bimodal distribution of packet lengths. Based on
these results, guidelines for Ethernet implementations are also
provided.

However, a weakness of the paper is that it fails to clarify the
specific reason that leads to the difference between the theory and
the measurement in this paper. For example, it does not clearly
explain what's the underlying reason for the "37% myth". It mentioned
just before its Section 3 that the possible justification of this
myth is because of the use of minimal-packet assumptions. However,
later in its measurement, it is not clear whether it is the length of
the packet or the cable length that caused the difference between the
theory and practice. In my point of view, extensive comparisons
should be offered to explain the effects of different parameters on
Ethernet performance.
Received on Mon Sep 18 2006 - 23:08:51 EDT