Name: Gennady Pekhimenko
Summary.
This paper describes two very useful problems for the Ethernet
(CSMA/CD and 10 Mbit/sec mostly):
-- how to make theoretical research that will be good for practice
using.
-- how to measure Ethernet characteristics and what to measure.
And also gives us some advices how to make high-performance Ethernet
networks.
As to the first problem, authors say us that there were a lot of
research papers that added considerable confusion in understanding
Ethernet capacity and even created some myths about it's performance.
To understand the real situation we need to read theoretical studies more
precisely. For helping us with this, paper gives us main fixed (bit
rate, max jam time, min/max packet length etc.) and non-fixed
(determined by the user) characteristics that we should keep in mind
while reading these studies. Also, we should be very careful with
different assumptions that most of the theoretical papers did. The
problem is that sometimes this assumptions are very far from real
Ethernet ("balanced star" topology, fixed packet size and some other
assumptions).
Then authors describe their way of performance measuring and give a
brief description of the main theoretical works on this topic.
After that we can see the methodology that they've used for
measurements and the results (in few figures with explanations).
The last part of this article gives few good advices on what to do or
not to do for getting high-performance Ethernet :
-- Do not use long cables
-- Do not put to many host on one cable
-- Use largest possible packet size etc.
The description for problems with few Ethernet implementations can be
very useful for "implementers".
I think that this paper is well written. It has both theoretical and
practical point of views on the subject. And gives a right conclusion
that we can get a good performance for typical applications if we'll
use their advices. The only think I want to mention that it's
sometimes not good to criticize theoretical research for their
assumptions, because they can predict some future problems. For
instance, it's good to analyze modern implementations for a more
"quick" networks (i.e. 1Gbit Ethernet) that we can see problems as
soon as possible.
Received on Mon Sep 18 2006 - 15:17:44 EDT
This archive was generated by hypermail 2.2.0 : Mon Sep 18 2006 - 15:30:04 EDT