Nagios Support Forum

The machines I have Nagios monitoring all have 24 cores, is just throwing the -r option by itself sufficient for check_load to give me an accurate idea of the load averages based on the actual number of cores?

I did a little googling first and found some posts where people mentioned using the formula:

y = c * p / 100

y = threshold value c: number of cores p: wanted load threshold

However this seems unnecessary and a waste of time to configure for each server (assuming I had servers with different numbers of cores), just wanted to verify.

Well, -r will divide by number of CPU's, that is one way you could do it. Load is always a bit tough to monitor when you have systems who's core count greatly differs. I believe that may be the option you are looking for though.

Ah duh, for some reason I thought -r would let it divide by total cores, not CPUs

Ok, I think I'll just specify the thresholds using the formula, thanks!

You could always write a wrapper around check_load that gets called every time the check is performed and calculates the proper value before passing it on to check_load...

eloyd wrote:You could always write a wrapper around check_load that gets called every time the check is performed and calculates the proper value before passing it on to check_load...

Thanks for the suggestion, I will definitely look into it.

Great! Let us know if you need help along the way, we'll keep the thread open.

Here's a related question, when determining the load values should virtual cores be counted the same as physical cores? All our boxes are hyperthreading enabled, and I initially configured check_load thresholds using the total physical+virtual cores.

You're in the weeds here. Basically, each core means "1" in a load average before the machine is bogged down. 4 cores = a load of 4 and still there is one process per processor. A load of 8 means 2 processes per processor which is still pretty good.

I would do something simple and see how it goes. If you're really sensitive to a load of "one process per processor" versus "1.1 process per processor" then I would start asking why.

eloyd wrote:You're in the weeds here. Basically, each core means "1" in a load average before the machine is bogged down. 4 cores = a load of 4 and still there is one process per processor. A load of 8 means 2 processes per processor which is still pretty good.

I would do something simple and see how it goes. If you're really sensitive to a load of "one process per processor" versus "1.1 process per processor" then I would start asking why.

Sorry, I didn't mean I configured it based on "one process per processor", but rather that for the purposes of determining the initial load thresholds (which I'm sure will need to be tweaked) I considered each virtual core to count the same as each physical core - just wanted to make sure that made sense.

Oh. Yup. It does to me. Each virtual core is the same as a physical processor.