I think the following message from Robert Mulley pretty much sums it
up. I have just done what he suggests, hard lock the Cisco switch in
10/full and my timings compare favourably with something running in
full duplex (a Linux box...).
The moral is "believe Cisco".
Arrigo
------- start of forwarded message (RFC 934 encapsulation) -------
From: Robert Mulley <robert_at_gnsconsulting.com.au>
To: Arrigo Triulzi <arrigo_at_albourne.com>
Subject: Re: 8200's old 10baseT and Cisco 10/100 switch problems
Date: Fri, 29 Jan 1999 22:12:29 +1100
Hi,
I actually despaired for a while over this same fact (my DBA just
wouldn't stop on the point). As far as I can work out, ifconfig -a
always returns simplex. I read the ifconfig man page and anything I
could put my hands on. Even going as far as adding speed=200 everywhere
I could. My network guys and I decided eventually that the damn thing
was in full duplex mode, even though it didn't say it. This was on a
cluster of 2x8400's. Even had a digital guy say "I can probably fix it,
just let me take one node down". No way I was doing that considering
they were both fully maxed out (CPU wise) production machines.
Anyway, I was getting approx. 4000 imputs and 4000 outputs per second at
one stage. I reasoned that half duplex couldn't do that. My advise cisco
wise, lock the port into 10M full duplex, set the same for ewa* and you
should be right.
Robert Mulley
Digital Unix Administrator
GNS Consulting.
------- end -------
Received on Fri Jan 29 1999 - 11:22:07 NZDT