> If you missed the 2.2.0-pre6 post, here is an update. :/ I compiled
> 2.2.0-pre9 and epic100.o again as a module, upon booting I ran scp as a
> non-root user, and proceeded to copy a very large file from one machine to
> another... After 470MB the sending machine rebooted... So the problem remains,
> I have not found the source, but I am looking at the cards low-power mode, and
> if thats not it I guess I will need to inspect the last packet sent to the
> receiving machine to see if there is ANY clue there...
>
> can anyone else run a similar test to verify this problem. Send at least one GB
> before being sure it works, as mine did not fail until after 470MB, which
> didnt take all that long really...
> also I should note tha the throughput nearly doubled when compiling
> into the kernel, and not as a module? modules shouldnt have that
> great a performance hit? Could someone familiar with _efficient_
> network drivers please look over the epic100 code as soon as
> possible :)
I apologize for the delay in responding to this note but I am catching
up on +1600 pieces of e-mail after being gone for a week.
I have had a lot of experience with the SMC Etherpower II cards based
on the EPIC chipset and I, quite frankly, would not recommend trying
to run these cards in any type of production environment.
I'm not sure if the driver is problematic or whether the hardware is
plain buggy but we haven't been able to make them stand up under load.
They refuse to talk to older switches that old SMC/Tulip cards and
Intel EtherExpress cards have no problems with. They have
been consistently hanging under load which require the interface to be
taken down and brought back up to restore function.
We have tested under both 2.0.35 and 2.0.36 kernels in uni and SMP
configurations. The problems seem to be pretty much invariant across
the last ' versions of the EPIC driver. The drivers, for whatever
reason, seem more stable within the confines of a 2.0.36 kernel. We
were experiencing solid machine lockups with these cards in 2.0.35.
We have tried 8 different cards and they are all acting the same way
which rules out problems with a single card. We are presently
stripping these cards out of our production servers and replacing them
with new Intel EtherExpress PRO/100+ cards. Our problems have been
ceasing as soon as we do this.
> Adam Scislowicz | Fingerprint
Greg
}-- End of excerpt from Adam Scislowicz
As always,
Dr. G.W. Wettstein Enjellic Systems Development - Specialists in
4206 N. 19th Ave. intranet based enterprise information solutions.
Fargo, ND 58102
Phone: 701-281-1686 EMAIL: greg@wind.enjellic.com
------------------------------------------------------------------------------
-
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@vger.rutgers.edu
Please read the FAQ at http://www.tux.org/lkml/