Bad CPU autoscaling is making games unplayable for a lot of users

[Roderick Colenbrander]

Hi Scott,

On Linux power management decisions are made by cpufreq governors. Most
distributions I have used recently use the 'ondemand' scheduler, which
makes decisions as you would guess based on cpu load. This scheduler has
various parameters you can tweak in
'/sys/devices/system/cpu/cpufreq/ondemand'. Its 'up_threshold' parameter is
typically set to 95, which means when the load on a CPU reaches 95% it is
time to increase the clock frequency to a higher level. Often this value is
too high to trigger a high clock frequency. Just for fun adjust it to let's
say 25 and see it change behaviour.

Figuring out the 'right' power settings is a very hard problem, because
there are different use cases of which battery life vs performance is
probably the most important one.

There is not much which can be done at the application level. If you were
to tune cpufreq settings from Wine this would interfere with other apps.

Maybe some settings like up_treshold should be tuned at the distribution
level, but finding the right balance is a problem and every user has
different preferences.

Thanks,
Roderick




On Fri, Sep 13, 2013 at 2:14 PM, Scott Ritchie <scottritchie at ubuntu.com>wrote:

> Over the past couple years I've been able to try out Wine games on many
> different environments -- laptops, desktops, even cloud servers.
>
> On many occasions, I've discovered that a game appears to run
> functionally but slowly, however upon further investigation I find that
> forcing the CPU to run at 100% "performance" mode can make the game
> playable.
>
> There are a few tools to do this (eg cpufreq-indicator for desktop
> users), however the long and short of it is that, for many users, I
> suspect they just assume Wine is slow and give up on that particular game.
>
>
> 1) Since Wine is truly CPU-bound here, why isn't it provoking the
> autoscaling to actually speed up the processor?  I've seen this happen
> on systems running with AC power and 100% battery life.
>
> 2) Is this a kernel issue?  A distribution power management
> configuration issue?  Or is there something we can do in Wine itself to
> be less courteous?
>
>
> Mostly I've seen this on Ubuntu, however I'm sure it's affecting other
> distributions that have any sort of power managment whatsoever as it can
> be quite tricky to get this right.
>
>
> Thanks,
> Scott Ritchie
>
>
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://www.winehq.org/pipermail/wine-devel/attachments/20130916/7b45566d/attachment.html>