Powermanagement improvements.

There's quite some discussion about powermanagement here at the Ubuntu Developers Summit in sunny Sevilla. There are quite some Intel people around here, providing lots of insight in powermanagement on a hardware and driver level. Here's a couple of points that have come up.

• First, the good news. KDE is in a pretty good shape when it comes to sleeping applications. Applications using timers can wake up the CPU from sleep states (or even prevent sleeping in the worst case). The CPU sleep states (usually named C-states) greatly reduce CPU power draining. The worst offenders are currently kopete (responsible for 40% of the wakeups on a Fedora test system) and HAL polling the CD-ROM drive every two seconds. Will Stephenson told me that kopete in KDE4 is fixed already, and that fixing the KDE3 version is on his radar. This makes KDE even more battery friendly. Generally, KDE's technical design makes it rather easy to fix those issues system-wide. Kudos to the people working on this!
• Tickless kernel. The new dynticks and hi-res timers make a *huge* difference. Ben Collins mentioned 30 - 40 minutes of extra battery life with dynticks and hi-res timers enabled. This stuff is new in 2.6.21, and already in the Gutsy kernel. dynticks also make it possible for the CPU to sleep much more. Another thing that's important in that respect is HPET, basically doing something similar, but less effective. Every bit helps, however.
• Intel has a nice tool to analyse userspace misbehaviour. It's called ticktool and tells you which applications keep the CPU from sleeping. Having identified those, stracing the process in question can give you a better idea of what's going on. Next step: Fixing it. Intel will release the tool on their website intellinuxpower.org (not up yet) shortly, rumours are "within a few days".
• Synchronising timer events can also help a lot. 5 timers spread over the second force the cpu to wake up much more often than 5 timers running all at once. Result: the CPU sleeps longer, you save battery life. There seems to be a patch doing this flying around in Oslo, although there are probably some things that need consideration before this can be widely deployed.
• Intel will be working on a profile manager that should make it easy for userspace to specify power-profiles for various scenarios. Switching of devices and putting devices into a lower powerstate when they're not used. Especially USB is a big offender, although recent kernels (2.6.21, especially) have mechanics to suspend those battery-drainers when they're not uses). A proposal will pop up on relevant mailinglists within a couple of weeks.
• Powermanager (the battery applet used in Kubuntu) will be ported to Qt4 for Gutsy. There's very little KDE specific stuff in there, so it should be possible without losing lots of functionality. Kpowersave is deemed to complicated from a user's point of view. Still, it would be interesting to know if kpowersave is being ported to KDE4, or if one should think about that functionality. powermanager is not distro-dependant, it does everything via HAL. I don't think a python application is a viable option to have in KDE4 default and running since it adds about 8MB of extra memory in use and takes a bit long to start, especially with cold caches.
• AMD/ATI's fglrx driver is a total pain in the ass. In fact, it is the one component driver that is most complained about. ATI just doesn't seem to care ("Increase your market share by a factor of 5 and we might fix the drivers"). I can only encourage people to actively avoid buying hardware containing ATI chips. Try to get an intel one. Given the bad quality of the fglrx driver and the lack of features, you're Free Desktop will be able to take advantage of much more eye-candy and powermanagement features.
  ATI's fglrx module is needed on my notebook, there's a x1300 chip in there, which is currently not supported by a free driver. Finding a kernel-version / drvier-version combination that actually does suspend with this blob loaded took me a week and probably two dozen recompiles of various kernel versions. ATI's engineers seem to actively ignore this issue. One of the latest releases contained a new user interface tool which makes it possible to change for example anti-aliasing setting for the video card. It doesn't support things like texture_to_pixmap (useful for compositing features), proper suspend support (refuses to resume properly in >1 of 10 cases), you often have to wait for it to support recent kernels (happened with 2.6.20, took roughly 2 months while a patch was in the wild). AMD/ATI engineers seem to completely ignore the user's needs. So by all means: Avoid buying ATI chips if you want your Free Operating system to run well.

[ Tue, 08 May 2007 17:24:23 +0200 ] permanent link

This weblog does currently not offer the option to comment. I would be happy to receive an email with your thoughts.

Weblog Archive