(no subject)

Paper Review: Measured Capacity of an Ethernet: Myths and Reality

This paper presents the results of the authors’ attempt to measure
capacity and fairness in Ethernet LAN technology.

The paper begins by explaining the considerable confusion that exists
as to the actual capacity of an Ethernet. Previous attempts to
determine the capacity, either by measurement or analysis, are said to
be flawed. One common myth, resulting from misreading theoretical
studies, is that Ethernet is saturated at an offered load of 37%.
The paper attempts to separate facts from the myths and show that
Ethernet is capable of supporting its nominal bandwidth and also
allocates it fairly.

The paper is organized into three major sections. First, a brief
review of theoretical studies is presented. Second, the paper
presents an extensive set of measurements collected from actual
implementations (experiments). Finally, the authors explain how to
apply the findings to real-world applications to achieve optimal
performance.

The author reviews several papers and highlights how the work fails to
properly analyze Ethernet LAN technology. The problems found in the
historical works fall under the following areas:

- Different protocol: Ethernet is 1-persistent, unslotted, CSMA-CD
with binary exponential
- False simplifying assumptions: balanced-star configuration, infinite
population, constant packet lengths and no buffering.
- Measuring the wrong variables which should be varied based on the
intended application
- Operating in the wrong regime (using worst-case offered loads)

The paper then presents a detailed description of its experiment which
adds significant value to its overall effect. Using their experiment,
several results are drawn both through measurement and analysis.
Based on these results, the authors have compiled a list of lessons
learned about Ethernet.

- Don’t install long cables; instead use bridges or routers
- Don’t put too many hosts on one cable
- Implement the protocol correctly
- Use the largest possible packet size
- Don’t mix serious real-time and serious bulk-data applications

The paper is well-organized and easy to follow. Sufficient background
information along with common terminology is first introduced in order
to support the analytical and experimental findings presented in the
paper. The extensive use of diagrams and graphs successfully aid the
delivery of the information. Overall this was a very well-written
paper and provides a strong argument in support of Ethernet LAN
technology.

-- Nadeem Abji
Received on Tue Sep 19 2006 - 00:35:27 EDT