We have a group of machines that are afflicted by this problem in our lab.
Cards affected are 3c590, 3c900, and 3c905 (all have 8K buffer). Under
consistent heavy load for long periods the machines will begin issuing
transmitter access conflict errors. These increase in frequency until the
machine finally ceases to transmit. It usually takes between 4 and 12
hours to get to this point. It can be fixed by downing the interface
("ifconfig eth0 down") and bringing it up again. However, the problem
recurs much faster (within 20 minutes) after doing this instead of
rebooting.
Oh, "heavy load" that we see causing the problem consists of about 250-300
transmitted packets a second every single second from boot until it drops
off the net. This isn't necessarily a lot of packets. Other machines we
have see much higher rates than this but not at such a consistent
sustained pace.
> Oh, someone pointed out that 3c59x v0.29c-all appeared on Becker's web
> site very recently. Can anyone confirm what this offers or fixes above
> v0.28c-all?
>
I believe this incorporates the patch that he sent me last week. I tried
it and it didn't seem to help. I sent him a report about its lack of
improvement yesterday. If you can help describe problems I'd suggest
maybe dropping him a line.
--Eric