On Thu, May 31, 2012 at 3:45 AM, Juan Francisco Cantero Hurtado
<i...@juanfra.info> wrote:
> Hi. I've been using OpenBSD on my netbook daily for a few months. I was
> using apmd with the -C setting. My netbook is slow and the battery life
> is important, so 800Mhz (apmd -C) or 1600Mhz (apm -A) is not a big
> difference for me.
>
> Now I have a desktop computer with support for cpu speed scaling. In
> this case the options of apmd are very limited. If I use "-A" my cpu is
> always at 2700Mhz, wasting energy and heating my room (I live in the
> very sunny Extremadura :) ). If I use "-C", apm rarely raises the speed
> and the cpu is almost always at 800Mhz.
>
> This is important because when I open a web page with a lot of
> javascript, the browser is very slow. Also when I compile something with
> "make -j1", apmd doesn't raise the speed of my CPU, I need use "make
> -j4" for raising the cpu speed to 2700Mhz.

What shows top, vmstat, systat about %sys, %usr, %idle during that
time? Because you can have 800MHz of CPU, but %usr and/or %sys can be
eg. only 20% so there's no reason to switch to higher frequency.

>
> I understand the problem with the limits of apmd. The developers can't
> generate a limits for each CPU and workflow. I've been playing with the
> values of PERFINCTHRES and PERFDECTHRES on usr.sbin/apmd/apmd.c and
> "checking" the changes of speed with "while true; do apm | grep cool;
> sleep 1; done". I'm using 20 and 60 respectively right now. Probably
> this values aren't the best values but are better than the defaults (for
> my CPU and workflow). The performance is very good and I can see the cpu
> speed raising and lowering. I've tested various values and 20/60 are a
> good compromise "performance vs energy consumption".
>
> In short, I have a suggestion for apmd. Add one option for to set
> PERFINCTHRES and other for to set PERFDECTHRES. Each user could to
> configure the behavior of "apm -C" on rc.conf.local. Example:
> apmd_flags="-C -I 20 -D 60". Obviously, "apm -C" would use the defaults
> if both options aren't used.
>
> The implementation is simple and it will not break any system.
> Unfortunately I can't code the idea (despite I've read the code), so I
> wanted share this with you. B Some developer interested? :)
>
> Cheers.
>
> --
> Juan Francisco Cantero Hurtado http://juanfra.info

Reply via email to