On Thu, Oct 21, 2010 at 7:36 PM, Laurence Tratt <[email protected]> wrote:
> 18 months ago I posted a patch to make apmd's -C and -A modes work
> half-sensibly on multi-processor machines:
>
>  http://marc.info/?l=openbsd-tech&m=123315164930014&w=2
>
> The patch went into the tree but was backed out because on some very slow,
> very old Sparc machines it apparently couldn't react quickly enough to
> changes in processor utilisation. Since I had no access to such a machine,
I
> couldn't debug it, and the patch died. That means that apmd -C has been
> close to useless on MP machines, since it needs all CPUs to be working hard
> before it increases hw.setperf.
>
> One day hopefully this functionality will move into the kernel (I know Ted
> and others are working towards that), but in the meantime someone asked me
> if I could resurrect the backed-out patch which I attach to the end of this
> e-mail. This patch cranks hw.setperf to 100 as soon as increased activity
is
> noticed and then gradually backs it off it can (in contrast to the old
patch
> which ramped up gradually; the new behaviour should increase battery life).
> It also reflects recent changes to apmd.c. Apart from that it's largely the
> same as before. It's been working for me over the last few days, but hasn't
> been extensively tested so YMMV. However, if you work off battery on an MP
> machine, you might find it helpful.

This has been working very well for me through the last two -current
snapshots running cool mode GENERIC.MP amd64.

Even though being a purely subjective observation my box seems more
responsive especially when switching into and out of browser windows.
And the battery consumption appears to be lowered a bit.

Thanks,
Rune

Reply via email to