[eepro100] Stable kernel/driver configurations?

Ben Greear greearb@candelatech.com
Mon, 18 Jun 2001 14:21:51 -0700


Khyron wrote:
> 
> Has anyone documented they're configurations to the
> point where they can claim that "kernel rev 2.x.y + driver version
> x.y.z" is stable. By stable, I mean the driver doesn't experience
> lockups or other connectivity failures (outside of the normal
> duplex mismatches, etc.), doesn't experience "card reports no resources"
> errors, or other problems which affect its ability to communicate
> with the network.
> 

I have had good luck with eepros using the v1.09i-7
driver (found in kernel 2.4.3-pre3 or so).

In order for me to consider them good, they have to be
able to change the link states and auto-negotiate correctly.

I also run them for long periods at 10Mbps bi-directional or
higher, and I have stressed them at 50Mbps bi-directional
for shorter periods (like 1 hour or more) with no ill effects.

You can find my notes, with pictures of the boards (since just
being an EEPRO card still gives you a huge amount of cards to
choose from!) here:

http://www.candelatech.com/linux_ethernet.html

I'll be happy to add to this document if someone wants to send me information
and/or cards to play with! :)

I'm also going to try to formalize the document a bit more in the coming
weeks to better define the tests I've done.  If you have specific tests in
mind (ie bandwidth, duration), let me know..as I can add them to my
LANforge test scripts.

> Any and all comments welcome. I want to gauge where the support
> for this hardware really is and whether its worth my (and my
> company's) time to continue attempting to make this hardware work
> or investigate non-Intel networking alternatives.

I have found nothing better, though I hear 3-com cards are good.  I
do know that I've had problems with (older?) tulip cards.  If you find
any that you really like, to let me know!

One thing to note:  I set the driver tx-queues to at least 400 (default is 100),
and the default socket buffers to 512k or 1Mb (default is 64k) when doing
high-speed throughput tests.  That may mitigate some of the resource-contention
issues.

Enjoy,
Ben

-- 
Ben Greear <greearb@candelatech.com>          <Ben_Greear@excite.com>
President of Candela Technologies Inc      http://www.candelatech.com
ScryMUD:  http://scry.wanfear.com     http://scry.wanfear.com/~greear