Re: [SLUG] Comuting CPU usage from /proc

[Tony Green <tgreen@xxxxxxxxxxx>]

On 27/09/2004, at 11:53 AM, Carlo Sogono wrote:
> The values of load average refers to the average number of
> CPUs running for the past 1, 5 and 15 mins and yes, I have
> looked at man 5 proc before. Polling the loadavg for the past
> min is not so flexible as polling, say, every 1 second so you
> can switch states quickly on demand. Your 1 min load *average*
> might be <1 but your average for the past few seconds might be
> a lot higher and the cpufreq daemon might not be able to
> switch immediately and would still have to wait 1 min until
> the avg >1.

You can't observe something without changing it's state.  The closer 
you watch your CPU usage, the higher it will be.

Also, the load average numbers give the number of jobs in the run queue 
(state R) or waiting for disk I/O (state D) averaged over 1, 5 and 15 
minutes.  I agree that it's not a great way of gaging actual CPU 
utilisation.

SNMP might be worth looking at.

TG
--
Tony Green <tgreen@xxxxxxxxxxx>