On Saturday 16 February 2008 00:27, John McCabe-Dansted wrote:
> After reading
>  http://www.bughost.org/pipermail/power/2007-May/000166.html
> I've filed a bug about Ubuntu's use of ignore_nice_load:
>  https://bugs.launchpad.net/ubuntu/+source/sysvinit/+bug/192303
> 
> I understand that due to the "Race to idle", it would actually save
> power to run nice tasks at full speed. Thus, except in edge cases,
> such as when the nice task caught in a loop, the ignore_nice_load
> option is pointless.
> 
> Would you agree that ignore_nice_load should not be set by default?

Venki should probably speak to ignore_nice_load.

> Secondly, as setting the CPU frequency saves little power during idle,
> why do you recommend ondemand rather than performance?

There are many different types of systems.
Some have a highly optimized idle that saves non-linear power
and rewards race-to-idle.  Some do not.

Some save more power in idle when entered at a lower "frequency".
(Actually it is the voltage associated w/ the frequency that helps).
Others lower the voltage on entering idle automatically, and thus
don't care what P-state they were in when they went idle.

Some systems consume a non-linear amount of additinal power
when running at maximum frequencey, and thus race to idle
isn't necessarily always an energy savings.

> Also what up_threshold do you recommend? (Ubuntu uses 31% by default,
> but AFAICT, race to idle would mean that a lower up_threshold would
> actually save power)
> 
> Finally, are there older CPUs for which the "conservative" governor
> would make sense?

AFAIK, conservative is not useful on any Intel processors.
I've heard that it is well suited for some older AMD processors,
but I don't know first hand.  We've totally ignored the conservative
governor for all of our tuning on Intel processors.

Davej threatened to delete conservative last year and make it
a mode for ondemand instead, but that hasn't happened yet.

cheers,
-Len

> If so, could you tell me off the top of your head which ones they are?
> 
> With thanks,
> 

_______________________________________________
Power mailing list
[email protected]
http://www.bughost.org/mailman/listinfo/power

Reply via email to