I have a server with multiple processors (2x dual core, technically 4 processors). The load average is often above 2.0 for the 1, 5, ond 15 minute averages. I was once taught that having a load average that is consistently above 2 meant it was time to upgrade (processors). I don't think this machine needs upgraded. Can anyone tell me if this load average only appears high because there are 4 processors (i.e. a consistent 8 would be time to upgrade) or at what consistent load average point a Linux box might need a processor upgrade? TIA

You've got it right. "load average" needs to be divided by the number of CPUs when analyzing things. The rule of thumb on many production servers is that the load is high when the load average exceeds "3 or so times the number of CPUs" on a frequent and continuous basis.

A load average is "how many processes are waiting to get time on a CPU". The more CPUs you have, the higher load average you can tolerate.

e.g., a 16 CPU machine is not necessarily considered CPU overloaded until you frequently and consistantly see load averages above 48 (16 * 3 = 48). Everybody has their own rules on this. It's just a guideline.

Your load average of 2 on a 4 CPU machine is roughly equivalent to a load average of 0.5 on a single CPU machine. In other words - not really a problem. You also need to look at load average over a longer period to see what your real long-term average is. Your machine might be very busy during one hour of the day, but that doesn't mean you necessarily need to get more CPUs. Maybe, maybe not. If the machine is loafing during the other 23 hours of the day you may not really care about the bad hour. Depends on how important that one hour high-load performance is to you.

Thanks for the detailed reply. I was actually trying to figure out where to set Nagios so it quits waking me up at night. What a lucky guess. =)

Not strictly - it's a combination of tasks on the runqueue and waiting I/O. Have a look at the man for proc, and check out the description for /proc/loadavgNo argument here - wish I had a few more ....

You may find this 2 part article informative:
http://www.teamquest.com/resources/g...ay/5/index.htm
http://www.teamquest.com/resources/g...ay/7/index.htm
FYI, on Redhat based systems, after loading top, hit the '1' on the main keybd and it'll show all cpus/cores. Try it on yours.

I monitor a lot of production servers using Big Brother (apparently a Nagios competetor, based on your context). I cut the notifications way back at night, so it has to be a blatant problem before it will start notifing people.

I'm very familiar with those automated middle-of-the-night notifications! It took a while for me to tune Big Brother's settings to allow me a bit of sleep. All is good now. I keep beefing up my scripts so that the people that cause the problems are the ones that Big Brother now pages. This allows me more sleep, and also tends to make people solve the problems they are causing.