*BSD News Article 77012


Return to BSD News archive

Path: euryale.cc.adfa.oz.au!newshost.carno.net.au!harbinger.cc.monash.edu.au!munnari.OZ.AU!news.ecn.uoknor.edu!news.wildstar.net!news.sdsmt.edu!news.mid.net!newsfeeder.gi.net!newsfeed.internetmci.com!news.sgi.com!news1.best.com!nntp1.best.com!shellx.best.com!not-for-mail
From: rcarter@best.com (Russell Carter)
Newsgroups: comp.unix.bsd.freebsd.misc
Subject: Re: 100BaseT tuning considerations?
Date: 27 Aug 1996 08:11:10 -0700
Organization: BEST Internet Communications
Lines: 37
Message-ID: <4vv36e$dqo@shellx.best.com>
References: <4vin1s$4um@server.cs.vt.edu> <4vsr44$iom@hpindda.cup.hp.com>
NNTP-Posting-Host: shellx.best.com

In article <4vsr44$iom@hpindda.cup.hp.com>, Rick Jones <raj@cup.hp.com> wrote:
>Carl Harris (ceharris@mal.com) wrote:
>: Has anyone experimented with tuning the kernel to get the most out
>: of 100BaseT?  Seems like it could do better than the ~40 mbit/s I
>: can get with a stock kernel config.  I'm experimenting with
>: transfers that involve no with no disk I/O on a not-so-loaded
>: system, equipped with a Pentium 120 and 32 megs EDO...
>
>How much CPU does it take for the system to drive 10 Mbit Ethernet to
>saturation? Since the MTU does not increase between 10BT and 100BT
>(nor 1000BT?), the host does not become any more "efficient" at
>sending bulk data.
>
>So, if you needed 25% of your CPU to go 10 Mbit/s, 40 Mbit/s with
>100BT seems quite reasonable to expect.
>
>However, you should also look at the window sizes you are using and
>see if there are any TCP retransmissions.
>
>rick jones
>http://www.cup.hp.com/netperf/NetperfPage.html

Actually, the maximum network bandwidth is much more strongly correlated
to main memory bandwidth, not CPU.  In order to peak out at almost 90 Mb/s TCP 
you need

1. a 66 MHz bus clock and pipelined burst SRAM, or
2. an unbroken P6 (and not because of CPU, either)

A properly built P5-100 will sustain over 85Mb/s.  A P5-100 with asynchronous
SRAM will get maybe 60% of that.

The 'stream' benchmark from McCalpin is an excellent measure of main memory
bandwidth.

Russell