Measured Capacity of an Ethernet: Myths and Reality

From: Waqas ur Rehman <wurehman_at_hep.caltech.edu>
Date: Mon, 18 Sep 2006 21:41:01 -0400

Amongst many purposed LAN technologies, Ethernet is the most successful.
Work has been done in past to predict the Ethernet performance
theoretically but because of the simplifying assumptions used in these
analyses people have misunderstood the results and have created myths
about Ethernet that do not reflect the reality. In this paper the author
attempts to dismiss those myths. To start with the author has defined
the parameters that effect performance e.g. parameters that are defined
in the specification (bit rate, propagation delay, slot time, packet
lengths) and those defined by the user (number of hosts, length of
cable). The author has highlighted that performance cannot be quantified
on a single scale, instead it includes measure like average delay,
throughput, channel capacity and fairness. Next the author has presented
the summaries of relevant studies and has shown how the people have
extracted the wrong lessons from these analyses.
The author then defines the characteristics of the Ethernet and in order
to present true picture of an Ethernet he establishes a test bed to
measure Ethernet performance. Using this test bed the author commuted
bit rate, standard deviation of bit rate, packet rate, standard
deviation of packet rate, transfer delay and exec delay. Through these
parameters the author has shown that

• Bit rate increases with increasing packet size but as the number of
hosts are increased it starts to fall down.
• Fairness increase with increasing number of hosts
• Average transmission delay increases linearly with increasing number
of hosts as apposed to previously held belief that it increases
dramatically when the load exceeds 1/e
• Network efficiency increases as collision resolution time decreases.

Based on these observations the author has suggested the following
guidelines for installing Ethernet

• Avoid long cables
• Limit the number of hosts on a single segment
• Use the largest possible packet size
• Implement the protocol correctly

In short this paper has dispelled the some of the misleading myths that
could mislead someone who is deciding between different LAN technologies
to implement for his LAN. Not only this paper provides some of the
guidelines that one can follow to tune the Ethernet according to ones
need and thus experience the performance promised by Ethernet.
Received on Mon Sep 18 2006 - 21:41:05 EDT

This archive was generated by hypermail 2.2.0 : Mon Sep 18 2006 - 21:41:29 EDT