Re: problem with cpu user on linux, 4.8.2

[Martin Pala]

I believe that monit shows correct information ;) It works  this way:

1.) each monit cycle reads the average cpu usage from the /proc/stat  
(total average across all cores as stated by linux kernel in the  
first line of the input)
2.) counts the deltas of the user, system, wait and total cpu usage  
between the current and last cycle and the appropriates percentage

monit is sampling according to the monit cycle length ... the value  
provides the 'immediate' average for the last cycle.

The difference between 'top' and 'monit' could be in the sampling  
method. i don't know the exact method which is used in 'top', but  
some tools, like for example solaris vmstat use the progressive  
increase to keep the trend and flatten the spikes ... it is based on  
exponential decay, so even though the load may be at 100%, the tool  
won't show the real usage immediately.

I'm running monit 4.8.2 on several DL380 with two dual core xeons and  
don't see any problems regarding the CPU usage statistics.

Martin

On Dec 1, 2006, at 12:59 PM, Aleksander wrote:

> Replying to myself, sorry.
>
> I thought about this a bit and maybe the problem is, that monit  
> checks for cpu user usage on ANY of the cores, not the average that  
> top shows by default. "1" shows each core and one of the cores  
> might have really been at 90+, but that's OK by me, when the other  
> 3 cores were mostly idle. The box was not overloaded.
>
> So how does it behave exactly, I didn't find this in the manual.
>
> Thanks,
>         Alex
>
>
> --
> To unsubscribe:
> http://lists.nongnu.org/mailman/listinfo/monit-general