[prev in list] [next in list] [prev in thread] [next in thread] 

List:       kde-devel
Subject:    Re: [PATCH] powerdevil-remove-cpufreq.patch
From:       Holger Macht <holger () homac ! de>
Date:       2010-03-08 11:12:04
Message-ID: 4B94DB84.3050006 () homac ! de
[Download RAW message or body]

On 03/08/2010 09:14 AM, Oswald Buddenhagen wrote:
> On Sun, Mar 07, 2010 at 11:57:37PM +0100, Holger Macht wrote:
>> Remove all CPU frequency scaling code from powerdevil and the kcm
>> module. You should never offer those options to the user because he
>> might not understand what the impact is. You only want a dynamic
>> frequency algorithm which is the default these days (ondemand
>> governor). A good explanation is provided in [1].
>>
> i disagree.
> 
> i use my laptop as a node in a compile cluster (not a particularly big
> one, heh). unlike my desktop, it is not water-cooled, i.e., loud when
> under heavy load. so if i want to compile something while, e.g.,
> watching a movie, i always set it to "powersave" manually.

You're taling about compile clusters, so you're obviously an experienced
user and technician, but people aren't. Nobody wants to hinder you from
setting things like that _manually_.

> the second point is that the ondemand governor doesn't give full
> performance. i'm running the node with max 3 concurrent jobs while it
> has only one core, so it is pretty much saturated all the time. but i
> still see it downclocking. of course this is a bug (hmm, i should
> upgrade the kernel on that machine ...), but "being robbed" of an easily
> accessible workaround would just suck.

Maybe it just reduces the frequency during the time where there is only
IO? Anyway, this would be a bug and it would be good to keep something
because of a bug.

> a third datapoint is the wrongness of the assumption that the workload
> is fixed and should just be executed as fast as possible. this certainly
> isn't so, for example in interactive applications which refesh "as often
> as possible" there is always somthing to do. downclocking the cpu is the
> only way to conserve power in such a scenario.

Can you give an example where an application refreshes that fast that
you would notice the impact? Frequencies are switched in microseconds
these days.

> also, perceived slowness of the machine has a psychological effect of
> not demanding too much of it, i.e., it changes the usage pattern to
> something more power-friendly.

You don't notice the difference from using ondemand vs. performance,
really. Working smoothly with a desktop isn't limited by processor
power, but rather by IO barriers.

Please continue the discussion in the review board to not having two
different places where discussion takes place.

Regards,
 Holger
 
>> Visit http://mail.kde.org/mailman/listinfo/kde-devel#unsub to unsubscribe <<
[prev in list] [next in list] [prev in thread] [next in thread]