[Beowulf] Tyan S2882

[Florent.Calvayrac at univ-lemans.fr]

Quoting Constantin Charissis <cch at dataswift.fr>:

> Krugger a écrit :
>> To be solved:
>> - random kernel panics that take out the logging even when all debug
>> flags are set in the kernel, as it fails to sync the disc during the
>> kernel panic.
>>
>

we had the same problems, even with BIOS upgrades,
excepted on the head node which
had only 2GB of memory. After a memtest run which reported
myriad of errors the vendor (Alineos) concluded that memory
was fine and on-spec and accepted to change the motherboards
for another brand (Serverworks I think) which works fine
even with slightly inferior performance.

This time we were glad, in order to save time
(and because of lack of familiarity with AMD64 at the time)
to have bought a turnkey cluster and to have
run intensive tests in the first days, before we had actually paid.
I guess that the motherboard switch would have been harder else !
French law protects universities in that regard, for the first
three weeks we are supposed to test before paying, even if hardly
anybody does that. The vendor recognized that they should have
run intensive burn-in tests before delivery, as advertised.