[procps] Re: C-States handling - new switch?
- From: Jaromir Capik <jcapik@xxxxxxxxxx>
- To: procps@xxxxxxxxxxxxx
- Date: Thu, 09 Feb 2012 07:02:18 -0500 (EST)
> >
> > 1.) with frequency set to 2.66GHz the load was approximately 34%
> > 2.) with frequency set to 1.73GHz the load was approximately 52%
> >
> > Whilst the load should be 34% in both cases ...
> >
> > I tested that with procps 3.2.8 and procps-ng 3.3.2 and both of
> > them
> > returned the same result.
>
> Hi Jaromir,
Hi Jim.
>
> I was referring to tics, not load.
procps users do not see ticks, they see just the load.
>
> And your loads look proper given the frequencies.
Like I've already written, for some people it might look correct,
for others not ... especially when the frequency might change
automatically without the user's intervention. Imagine cases
when character of the total load is causing periodic frequency
changes. The load graphs for processes which have a constant
resource consumption then must look like a rampart (jumping up
and down with each frequency change). That's what I meant when
I was writing about usability of the output for long-time monitoring.
This should definitely be configurable by a switch, because
it's impossible to sample the cpu frequency and procps output
atomically ... that means it's even not easily possible to
reliably stretch the results later in the monitoring platform
according to the freqency graph.
I'm not talking about some imaginary solution, I'm talking about
real usecase. At the moment we have really MANY users which
are affected and that's why I might look anxious (because I am).
>
> Jim
Jaromir.
Other related posts:
