Overall system impact when 1 cpu core is spiking when the rest are not
 

I'm fairly new as a Systems Administrator, but I have a fair amount of experience as a programmer/analyst in IT.

One of the problems that we struggle with in my current position is when should a (Linux) system send us an alert that something is wrong?

Specifically I'm referring to CPU alerts. One of our (bare-metal) systems has 8 cpu cores. We have HPSIM setup to send us an alert when a CPU is above a certain threshold (80% or more over X amount of time). In this case, one core (cpu 7) is spiking at near 100% usage, but the rest of cores are more or less at rest.

The question is, is this really an issue? Is the rest of the system still fine? My gut instinct tells me yes, it is. I assume that the process that's hogging one CPU is probably not multi-threaded, and there is nothing we can do about that. However I'm guessing that if a new process comes along, it would go to any of the other 7 freer processors.

Curious what you all think,

--Shaun

I think that you paid for a CPU, and ideally it should be running at 60-80% (all cores) at all times. When it's not, you're wasting your investment.

So no, one core running at 100% while 7 cores are idle (12.5% CPU utilization) is not a problem.

I think if the multi-core scheduler is enabled in the kernel it should switch the process between cores to prevent 1 core from getting consistently hotter than the others. At least, this is what it does on my system when the process is not threaded.

That's correct.

I'd look to why either some software has been dedicated to that core or why it is performing above the other's.

Some applications would use a single core to act as a control for work sent to other cores. Many high end or special use apps could do that.

I find it odd that it is core 7 and not core 0 or 1 depending on naming.

Thanks all, I'll check to make sure multi-treading options in the kernel config are set. I imagine they are, it's a RHEL-4 system I believe.

I think cpu 7 was spiked because there was another process earlier that was spiking cpu 0. The new process probably went to cpu 7, and then by the time I looked it at again, the process on cpu 0 had finished. (I'm using `top` to see which cores are spiking).

This information will help me go back to the senior admin I work with and tell him it's ok, we don't need to be woken up at 2am because one core is spiking :)

--Shaun