++ 03/02/05 14:45 -0600 - <Miles O'Neal>:
Hi,
> Ken Teh said...
>
> |NICs are activated in order eth0, eth1, ... What gets called eth0, eth1,
> |... is determined by the order in which they are "discovered" which is
> |usually just the PCI order on the motherboard.
>
> But therein lies the problem. The
> discovery order is at least somewhat
> random (or close enough that it's
> irrelevant whether it truly is).
>
> We have several servers with two
> to four NICs. Let's say that on
> first boot of a 4 NIC system, the
> NICs come up, bottom to top, as 0,
> 1, 2, 3. After a prolonged power
> failure that drains the UPS, they
> come up 0, 2, 1, 3. We figure this
> out and move the cables. A couple
> of months later, the system locks up
> when the PS dies. New PS, and they
> come up 3, 2, 1, 0. We discover this
> and move the cables. A few months
> later, we have to add more RAM and
> disk. This time they come up 0, 1,
> 2, 3 again.
>
> This is one real life example. The
> numbers aren't necessarily exactly
> what we saw, but they're representative.
> Another four NIC system rebooted
> fine 2 or 3 times with the same order
> before showing similar behavior.
>
> I don't believe this was the case
> with RedHat 7.1, but I wouldn't swear
> to it.
No, you can swear :)
This is how we found the problem migrating from 7.2 to RHEL3.
> I have bought a 4 port card from
> Soekris to see if having all the
> ports on one card solves the problem,
> but haven't gotten to test it yet.
I won't think it would because it a matter of IRQ at boottime handling, right?
Which port gets the first IRQ is going to get eth0, etc.
John
##############################################
# John Goebel <jgoebel(at)slac.stanford.edu> #
# Stanford Linear Accelerator Center #
# 2575 Sand Hill Road, Menlo Park, CA 94025 #
############################################ #
|