PowerNowd v.75

PowerNowd v.75

[John Clemens]

Yes, i released another cpufreq client daemon.  don't yell, i actually
wrote it months ago when everyone else was just writing their's too :-).

It's simple, and just works very well for me.  I'm not trying to step on
anyone's toes but i figured i wrote it so I may as well put it out there
in case someone else finds it ueeful or wants to use some of the ideas in
it.  not that it's rocket science...

http://www.deater.net/john/powernowd.html

enjoy,
john.c

-- 
John Clemens          http://www.deater.net/john
john@deater.net     ICQ: 7175925, IM: PianoManO8
      "I Hate Quotes" -- Samuel L. Clemens

Re: PowerNowd v.75

[Jeremy Fitzhardinge]

On Thu, 2003-07-17 at 12:58, John Clemens wrote:
> Yes, i released another cpufreq client daemon.  don't yell, i actually
> wrote it months ago when everyone else was just writing their's too :-).
> 
> It's simple, and just works very well for me.  I'm not trying to step on
> anyone's toes but i figured i wrote it so I may as well put it out there
> in case someone else finds it ueeful or wants to use some of the ideas in
> it.  not that it's rocket science...

This looks very similar in idea to speedfreq; my intention was to keep
things simple and make everyone else do the hard work of deciding what
policy to adopt.  It runs as a daemon, speedfreqd, but has a client
which can set new policies and also monitor state changes (either policy
changes or speed changes).  The client side is wrapped up into a
library, so it should be easy to use from other programs (gui interface,
etc).

For the "dynamic" policy, I simply calculate the amount of system idle
time by looking at /proc/uptime; I don't even bother looking at all the
processes.  It seems to work well enough, though it can't ignore niced
processes.

http://www.goop.org/~jeremy/speedfreq

	J

Re: PowerNowd v.75

[John Clemens]

Hi Jeremy,

I still think parsing /proc/uptime to get a load average is the wrong
heuristic to use.  load average is not cpu usage... That said i think you
have a nice daemon there with some good ideas.  If you like, you are
welcome to steal the /proc/stat parsing code from powernowd (which, in
turn, was stolen mostly from procps).  It even supports SMP for when
CPUfreq gets SMP capability... in theory :).

I really don't have much of a desire or the time to make a complete
product out of powernowd, but i thought it had some pretty cool ideas.  So
I just cleaned it up a bit and released it.. to allow others like you to
see those ideas and maybe get some use out of them.

john.c

On 19 Jul 2003, Jeremy Fitzhardinge wrote:

> ....

> For the "dynamic" policy, I simply calculate the amount of system idle
> time by looking at /proc/uptime; I don't even bother looking at all the
> processes.  It seems to work well enough, though it can't ignore niced
> processes.
>
> http://www.goop.org/~jeremy/speedfreq
> 	J

-- 
John Clemens          http://www.deater.net/john
john@deater.net     ICQ: 7175925, IM: PianoManO8
      "I Hate Quotes" -- Samuel L. Clemens

Re: PowerNowd v.75

[Jeremy Fitzhardinge]

On Sun, 2003-07-20 at 21:58, John Clemens wrote:
> I still think parsing /proc/uptime to get a load average is the wrong
> heuristic to use.  load average is not cpu usage...

I know.  It isn't getting load average (which is in /proc/loadavg), but 
idle time.  The two numbers in /proc/uptime are seconds of uptime and
seconds of idle uptime.  By polling it, you can pretty easily work out
how idle the CPU was in your last poll interval.

>  That said i think you
> have a nice daemon there with some good ideas.  If you like, you are
> welcome to steal the /proc/stat parsing code from powernowd (which, in
> turn, was stolen mostly from procps).

I'll have a look at it.  It might be useful to do things like ignore
niced processes, but on the other hand, who runs seti@home on a battery
powered laptop?

>   It even supports SMP for when
> CPUfreq gets SMP capability... in theory :).

The cpufreq core supports SMP, but I don't know if any of the CPUs do. 
What is your policy for SMP?  Do you compute overall system use and then
set all the CPUs to some power state which can handle it, or try to work
out how busy each CPU is and power-control them separately?  

> I really don't have much of a desire or the time to make a complete
> product out of powernowd, but i thought it had some pretty cool ideas.  So
> I just cleaned it up a bit and released it.. to allow others like you to
> see those ideas and maybe get some use out of them.

I don't have much more to do with speedfreq, except I've been meaning to
hack of the existing gnome applets into extracting and displaying some
info.

	J