load and power management

[Peter A]

On Wednesday 14 April 2010 10:34:08 Kwan Lowe wrote:
> On Tue, Apr 13, 2010 at 7:51 PM, Peter A <loony at loonybin.org> wrote:
> > Am I doing something wrong? My view is pretty simple but the few tests I
> > did seemed to all indicate the same thing...
> > 
> > How valuable is reporting the CPU usage if its not really shows how much
> > of maximum CPU power is used?
> > 
> > Wouldn't it make more sense to show CPU usage normalized to the maximum
> > frequency rather than the current frequency?
> > 
> > Does the load as display with uptime compensate for power management? I
> > went through some of the kernel code the scheduler and the cpufreq
> > kernel modules but didn't see anything.
> 
> I don't think so.  From my  understanding,  the load is computed by
> sampling at a given interval. The per process load is a function of
> how often a particular PID shows up in the sample.
> 
> http://www.teamquest.com/resources/gunther/display/5/index.htm
> 
> I believe this method is used because of limits in a processor knowing
> its "speed". I.e., every processor has different metrics, identity
> strings, etc.. We can calculate synthetic numbers like bogomips, but
> this won't be valid throughout the run of the system (system is not
> necessarily at full power during boot or vice versa).  To add to this,
> the metrics are computed in the kernel so there are performance and
> possibly other technical reasons that calculating those numbers can't
> be done.

Makes sense - cause if you have a run queue, then your CPU is anyway busy and 
it should clock higher... That kinda makes it a non issue...

Any ideas on the other 3 points though?

Peter.

-- 
Censorship: noun, circa 1591. a: Relief of the burden of independent thinking.