Update on DeviceKit
hmacht at suse.de
Thu May 8 14:05:26 PDT 2008
On Do 08. Mai - 22:00:13, Matthew Garrett wrote:
> On Thu, May 08, 2008 at 10:56:45PM +0200, Holger Macht wrote:
> > On Do 08. Mai - 21:50:42, Matthew Garrett wrote:
> > > On Thu, May 08, 2008 at 10:38:43PM +0200, Holger Macht wrote:
> > >
> > > > With you're argumentation being true, it wouldn't make sense to have
> > > > something like an ondemand governor at all. Not to mention having an
> > > > up_threshold exported to userspace. Just always run at full speed.
> > >
> > > In an ideal universe, it wouldn't be. However, there's a (small) power
> > > cost involved in ramping up the voltage and frequency. If the amount of
> > > work to be done is sufficiently small, you can win (slightly) by staying
> > > at the lower frequency, and so ondemand makes sense.
> > So setting the up_threshold down, and thus staying a little bit longer at
> > a low frequency, makes sense too.
> Depending on your workload. Systems with bursty but short periods of
> work will benefit from different values when compared to systems doing
> longer periods of work. It can't be expressed as a linear scale, and at
> the point where you're exposing implementation details like that over an
> OS-agnostic abstraction layer, you've already failed.
So, usual desktop systems have what kind of workload? Compiling over a
longer period of time? No. Browsing to another web page? Yes.
More information about the hal