*BSD News Article 52366


Return to BSD News archive

Path: euryale.cc.adfa.oz.au!newshost.anu.edu.au!harbinger.cc.monash.edu.au!simtel!col.hp.com!news.dtc.hp.com!hpscit.sc.hp.com!hpax!cupnews2.cup.hp.com!raj
From: raj@cup.hp.com (Rick Jones)
Newsgroups: comp.unix.bsd.freebsd.misc
Subject: Re: Ethernet Card Benchmarks
Date: 6 Oct 1995 00:26:29 GMT
Organization: Information Networks Division: Hewlett Packard
Lines: 38
Message-ID: <451t3l$ev1@hpindda.cup.hp.com>
References: <DFzI5F.H20@ritz.mordor.com>
Reply-To: raj@cup.hp.com
NNTP-Posting-Host: hpindio.cup.hp.com
X-Newsreader: TIN [version 1.2 PL2]

Chris Mauritz (ritz@ritz.mordor.com) wrote:
: Has anyone done detailed benchmarking with various ethernet
: cards and FreeBSD?  

: I've been using 3COM 3C509's, but I was considering going
: with either (a) faster ISA cards or (b) looking into some
: 100mbit ethernet cards.

It should be possible to use netperf
(http://www.cup.hp.com/netperf/NetperfPagel.html) to get some
comparisons. There is a database of results at that URL - one which
could use a few more submittals ... :)

: This is mostly just to speed up NFS traffic and remote
: backups.
: Any tips/suggestions would be greatly appreciated.

Here is where things could be interesting - 100Mbit/s Fast Ethernet
does not increase the MTU of the link, so data transfer is no more
"efficient" from the persoective of the hosts on either end - they
still need as much CPU to move 1024 bytes of data as before. So, if
you are at/near CPU saturation with 10 Mbit/s you will not necessarily
see much of a change in performance (modulo switching from a
Programmed IO to a DMA card).

Also, unless your NFS traffic is primarily reads and writes of 8KB,
NFS is more latency sensitive than it is bandwidth sensitive. Things
like getattr and lookup typically are small (< 192 bytes), so their
performance is limited more by the host CPU path lengths and the
delays in I/O busses and interface cards than bit rate on the wire.

So, if you do get a chance to play with netperf, be sure to include
one the TCP_RR or UDP_RR test in your benchmarking.

happy benchmarking,

rick jones
http://www.cup.hp.com/netperf/NetperfPage.html