*BSD News Article 43325


Return to BSD News archive

Xref: sserve comp.os.386bsd.questions:16966 comp.os.386bsd.development:3288
Path: sserve!newshost.anu.edu.au!harbinger.cc.monash.edu.au!bunyip.cc.uq.oz.au!munnari.oz.au!news.hawaii.edu!ames!elroy.jpl.nasa.gov!decwrl!vixie!nnrp!paul
From: paul@vix.com (Paul A Vixie)
Newsgroups: comp.os.386bsd.questions,comp.os.386bsd.development
Subject: correction (Re: ethernet throughput)
Date: 1 Mar 95 03:02:33
Organization: Vixie Enterprises
Lines: 26
Message-ID: <PAUL.95Mar1030233@sager.vix.com>
References: <3it7m8$o39@clavin.uprc.com> <3itrg4$pfj@crl9.crl.com>
NNTP-Posting-Host: sager.home.vix.com
In-reply-to: ggrant@crl.com's message of 27 Feb 1995 16:41:40 -0800

>in reality , due to CS/MA gaussian backoff, a normal ethernet cable 
>saturates at about 3.5 Megabits per second... or 450-500 KBytes per second..

Not even close.  Boggs and Mogul at DECWRL published a tech report back in
1988 that showed an ethernet that gracefully load shared for up to 30 hosts
(DECWRL Titans, early RISC machines) such that a fair share was still given
(roughly 1/N to each of N hosts) and that the aggregate data rate for N hosts
was in the 90+% range even with 30 hosts banging away at full speed).

In my own less ambitious test environment (driver development for BSD/OS), I
regularly get 1150KB/s TCP throughput between two mediocre PC hosts using
cheap $100 3COM cards with tiny 2K FIFO's.  That's the theoretical maximum
for a collision free ethernet, which mine was that day.  Now, I don't know
what would happen if I put 30 of my cheap PC's on a wire (and I never shall
know, either), but after helping Boggs and Mogul set up their test net and
after reading their tech report, I am a strong believer in Ethernet.

<wrl-techreports@decwrl.dec.com> is a mailbot which can be coerced into 
sending you postscript for the Boggs/Mogul report if you ask it for "help".

Disclaimer: I don't work for DEC any more.
--
Paul Vixie
La Honda, CA
<paul@vix.com>
decwrl!vixie!paul