Subject: Re: 3c590 and NetBSD 1.2?
To: None <port-i386@NetBSD.ORG>
From: Curt Sampson <curt@portal.ca>
List: port-i386
Date: 10/07/1996 20:22:15
This is just a note on the status of my explorations with my 3c905
cards.

I've now upgraded my 486 from 1.0 to 1.2, and learnt in the process
that when one is upgrading through compiling and installing source,
one should install the kernel before the userland. :-)

So this means I've now got two `working' 3c905 cards in two NetBSD
machines, a P133 and a 486/100 (both with 64M of RAM). The two 3Com
cards form their own subnetwork, connected with a cross-over cable,
and the two machines are also connected into the general LAN through
another pair of normal 10base-T cards (a cheapo in the P133 and an
SMC Elite in the 486).

I tried to replicate the problem with the cards and couldn't. It actually
looks like the cards, despite being set for 10 Mbps mode, are running
in 100 Mbps mode. However, something is very wrong somewhere. I get the
following performance figures from netperf:

    486 -> P133, 3Com		22 Mbps
    P133 -> 148, 3Com		1.5 Mbps (!)
    486 -> P133, 10base-T	5.5 Mbps
    P133 -> 486, 10base-T	6 Mbps
    486 -> 486, loopback	35 Mbps
    P133 -> P133, loopback	100 Mbps

The 10base-T figures seem a bit low to me, but then again the P133
does have a cheap 16-bit PIO network card in it, and the network
may be a bit more loaded than last time I tried this. (I generally
see 7-8 Mbps between 486s with the SMC card.)

That the 486 -> P133 3Com rate is only 2/3 the loopback rate would
seem to indicate (to me, at least, correct me if I'm wrong) that
the driver and/or card are not currently not as efficient as they
could be. (Part of this may be because the card is running in PIO
mode instead of DMA mode.) The asymmetry between the two directions
on the 3Com side is rather astonishing, though.

My first step is to see if I can't convince these cards to run in
10base-T mode. If worst comes to worst, I may be able to dig up a
hub to plug them in to.

I think my next step is to drop some statistical gathering at
certain points in the 3com driver to see what I can come up with.

Also, I'll be having a look at the Linux driver for this card,
which does use DMA mode to see what I can find out from that.  (See
<http://cesdis.gsfc.nasa.gov/pub/linux/drivers/vortex.html> for
more info on this.) I'm still waiting for my docs from 3com.

And oh, by the way, I don't actually have a clue what I'm doing,
so any help is welcome. :-)

cjs

Curt Sampson    curt@portal.ca		Info at http://www.portal.ca/
Internet Portal Services, Inc.	
Vancouver, BC   (604) 257-9400		De gustibus, aut bene aut nihil.