Server spinning out of control...

[Roger 'Rocky' Vetterberg]

Jamie wrote:
> 
>     That is a good idea, thanks. We did check that though. Went through
> each user's accounts checking their .forwards and procmaillrc files.
> 
>     We are running spamassassin 2.55, and in the global procmailrc file we
> call spamc which connects to a spamd running on another machine.
> 
>     Are you aware of any other system utilities that might be used to
> trace CPU consumption and trap problems? We've taken a lot of stabs in the
> dark with what it could be, and we'd like to try some solid diagnostic
> utils to shed more light.
> 
>    - Jamie

Try running systat -vm, that should give you a good overview over what 
happens when the load skyrockets.

I had a similar problem once, not as extreme as the one you describe 
but the symptoms where the same. A few times a day one of our servers 
reported load averages at about 5.0-5.5. By the time I got there (30 
second run to the serverroom) the server was always back to almost 
idle, avg around 0.2-0.5. The only thing that was different in this 
compared to most of the other servers was the nic. Since the onboard 
nic died we had to replace it with a low profile PCI nic. I cant 
remember the exact make and model, but it was probably something cheap 
from the nearest computer store.
Using systat I noticed that during the bursts of high loads the number 
of interrupts on the nic went skyhigh. We replaced the nic with a more 
wellknown brand, and the server flatlined its load average. Its still 
doing exactly the same tasks but rarely goes above 0.1.

--
R