Relay-Version: version B 2.10 5/3/83; site utzoo.UUCP
Posting-Version: version B 2.10.2 9/18/84; site cornell.UUCP
Path: utzoo!watmath!clyde!cbosgd!ihnp4!houxm!vax135!cornell!jqj
From: jqj@cornell.UUCP (J Q Johnson)
Newsgroups: net.lan
Subject: Re: Ethernet statistics
Message-ID: <1534@cornell.UUCP>
Date: Sun, 8-Dec-85 06:38:35 EST
Article-I.D.: cornell.1534
Posted: Sun Dec  8 06:38:35 1985
Date-Received: Tue, 10-Dec-85 06:20:31 EST
References: <1294@wucs.UUCP>
Reply-To: jqj@cornell.UUCP (J Q Johnson)
Organization: Cornell Univ. CS Dept.
Lines: 19
Keywords: vax unix ethernet

In article <1294@wucs.UUCP> br@wucs.UUCP (Bill Ross) writes:
>I'm trying to find rough figures for several numbers:
>	     Average ethernet packet length
>	     Average time between packets
>	     Anything else along these lines
The distributions of these numbers very non-normal and installation
dependent, so a simple average statistic is useless for almost all
practical purposes.  For example, packet length is sharply bi- or tri-
modal, with one mode at protocol headers+1 data byte, and another mode
at the typical bulk data size (headers+ 1K for bsd systems, headers+576
if you use IP and an MTU corresponding to the ARPA value, etc.).
Average time between packets does not capture burstiness, which is the
interesting performance limiter.

One traditional statistic of some interest is the percent of packets
transmitted with 1 or more collisions.  On 2 ethernets I manage, this 
average over the last 4 days is .4% on one net, .2% on the other.  More
typical weekday values are order 1 to 4%.  (net 1 has 4 vaxes, 10 suns, 
45 Xerox DLs, 5 Symbolics LMs; net 2 has about 16 BSD systems and 10 DLs.)