Monitoring a remote linux server CPU usage

[jainwolverine]

Hi,

I would like to understand how to monitor a remote linux server CPU usage.

Linux server is an ubuntu instance running on amazon cloud.

Nagios server is also running on amazon cloud.

Right now if i go to

configure wizard,

I see CPU but i can only set threshold for IO/WAIT but not for cpu usage.

I would like to know how can we do this.

Thanks.

RJ

[yancy]

jainwolverine,

have you installed nrpe on your ubuntu machine yet? I'm not sure if your cloud instance has any limitations, but that's the first step to monitoring a linux machine.

http://www.omegaprojex.com/index.php/20 ... on-ubuntu/

regards,

-Yancy

[jainwolverine]

Hi Yancy,

Yes i have installed nrpe on my ubuntu server. I believe its working fine.

I am not able to see under CPU section, cpu load. I can see IO wait.

Please check out the attached screenshot.

Is Load option for CPU load?

thanks.

RJ

[jainwolverine]

Suppose i would like to set the following:

Warning if CPU reaches 15%

Criritcal if CPU reaches 40%

Its a 2 core processor.

What should be the values for load? I see there are three values.

for warning and errors

Thanks.

RJ

[nscott]

Those three values are for the 1,5,15 minutes average of the load respectively.

[jainwolverine]

sorry about my lack of understanding on this

So is this correct?

15, 15, 15 for warning

and

40, 40, 40 for critical.

Thanks.

RJ

[nscott]

Yes, that would work, however its generally better to have the warning and critical values go down as the load duration goes up.

For example, say you have a server, that experiences a transient load spike, like a someone grepping a large file that only lasts a minute and then back to normal. This would cause the 1 minute load to be very high, but since it was a short spurt, the 5 and 15 minute readings are not affected too much by it, so allowing a large spike for a short duration is ok. However, if the CPU Usage continues for a few more minutes, the 5 minute average will reflect a higher average, which means the load has been high for most of the five minute period, which is why setting the 5 minute average threshold to be lower is generally a good idea. The same logic is applied for the 15 minute average.

In summary, a computer under 30% CPU Usage for a 1 minute average does not necessarily mean the CPU is overtaxed, so it shouldn't cause a warning or critical. However, a CPU that is consistently about 30% for 5 or 15 would mean there is a process that is running consistently indicates that there is a problem that should be addressed.

[jainwolverine]

Thanks for this explanation.

We monitoring a node.js server and this sever is based on real time application.

We would like to monitor real time spikes. So basically receive an alert as soon as we see a jump in CPU.

5/15 we dont care much.

so basically do you think is this a good number for the load:

warning (1/5/15)

15 10 5

for critical (1/5/15)

40 30 20

Thanks.

RJ

[scottwilkerson]

It really depends on the system and how many CPU's you have.

Generally, for most of my systems, I would say this is what I would use on a 4 CPU system...