Is Linux Power Management Getting Better Or Worse?

[ioannis]

It's not just about software improving but also hardware. 2.6.35 introduced the intel_idle driver, which makes a massive difference for CPUs that support it but none of those tested do.

I use the 2.6.35 (maverick), on an Acer One netbook. Don't know if it really uses the intel_idle instead of acpi_idle, but its idle power usage is abysmal. Powertop reports lots of load balancing ticks (which is another bug that is supposed to be fixed in 2.6.35) and generally the laptop is burning hot.

I wish the linux kernel developers where a bit more concerned on power usage. It's killing the netbook experience (makes Microsoft's solutions much better, which shouldn't be the case)

[ioannis]

mind you, it may well be the distro vendors' fault (Canonical in my case), that have not set up things properly, to take advantage of power saving features. Or as others here have mentioned, the CPU might not be the culprit, rather the GPU or some other subsystem.

[Chewi]

I use the 2.6.35 (maverick), on an Acer One netbook. Don't know if it really uses the intel_idle instead of acpi_idle, but its idle power usage is abysmal. Powertop reports lots of load balancing ticks (which is another bug that is supposed to be fixed in 2.6.35) and generally the laptop is burning hot.

I wish the linux kernel developers where a bit more concerned on power usage. It's killing the netbook experience (makes Microsoft's solutions much better, which shouldn't be the case)

intel_idle is chosen over acpi_idle if your CPU supports it. I gather the Atom does. /sys/devices/system/cpu/cpuidle/current_driver should report the name of the driver being used. I also use i7z to monitor its behaviour but I don't know whether this works on the Atom. Maybe give it a try.

[fewt]

Why didn't you test using Jupiter (http://sourceforge.net/projects/jupiter/)? You may have different results, and Fedora 12 and 13 are supported.

[locovaca]

Why didn't you test using Jupiter (http://sourceforge.net/projects/jupiter/)? You may have different results, and Fedora 12 and 13 are supported.

Because this wasn't a test for Jupiter, it was a test for baseline Fedora.

[fewt]

Because this wasn't a test for Jupiter, it was a test for baseline Fedora.

The baselines haven't really changed in the 14 years that I have used Linux, why would they magically change now?

[Jimbo]

One thing that hesitates me a lot about linux is the cpu usage under minimal desktop work. Totally idle system consumes around 2% CPU on my ubuntu 64 witch is great!, although a minimal work like:

- Switch desktop 10-20% cpu
- Minimize/maximize window 10-20% cpu
- Move window 10-20% cpu
- Opening a system monitor cpu usage graph: redrawing the graph itselfrs consumes 5-10% cpu
...

This doesn't seems much, but your cpu is switching from idle state to power on state under simple desktop operation, so power consuption / battery life degrades.

I am trying with kde 4.5 - radeon OSS drivers with and without compositing, and under minimal / none desktop effects. I am not sure if this is kde, xorg or radeon drivers fault ¿? but on windows this simple operations consumes no cpu.

[ioannis]

One thing that hesitates me a lot about linux is the cpu usage under minimal desktop work. Totally idle system consumes around 2% CPU on my ubuntu 64 witch is great!, although a minimal work like:

- Switch desktop 10-20% cpu
- Minimize/maximize window 10-20% cpu
- Move window 10-20% cpu
- Opening a system monitor cpu usage graph: redrawing the graph itselfrs consumes 5-10% cpu
...

This doesn't seems much, but your cpu is switching from idle state to power on state under simple desktop operation, so power consuption / battery life degrades.

I am trying with kde 4.5 - radeon OSS drivers with and without compositing, and under minimal / none desktop effects. I am not sure if this is kde, xorg or radeon drivers fault ¿? but on windows this simple operations consumes no cpu.

Although it's possible that there is a difference in the way Linux and Windows report CPU usage (system wide vs user space), you are right. I think it's a combination of the overall Xorg architecture (XRender etc) and drivers.

[ioannis]

intel_idle is chosen over acpi_idle if your CPU supports it. I gather the Atom does. /sys/devices/system/cpu/cpuidle/current_driver should report the name of the driver being used. I also use i7z to monitor its behaviour but I don't know whether this works on the Atom. Maybe give it a try.

thanks for tip, it does use the intel_idle. And no, i7z does not work on the Atom. It's for Nehalem based architectures.

[ioannis]

By the way, has anyone tried MeeGo on a netbook?

It uses kernel 2.6.33. I'm getting very few weakups per second (~35) and long periods in the lowest state (32ms!) with 99.5%. ACPI reports 8Watt and 7hours of battery life. The laptop runs cool. It doesn't even use intel_idle driver (?) Intel did a good job with their distro.

Compare that to Ubuntu 10.10 netbook edition, with 10x more weakups per second and less than 2ms in the lowest state (and ~45%), ACPI reports 9.6W and 5.6hours after taking all suggestions from powertop (and powering down the webcam). Same brightness level.

In comparison, Windows XP reports 6.5hours. MeeGo beats WinXP, as it should be. Those extra 1.6W Ubuntu takes are a shame!

Phoronix, you can do something to raise awareness. Do a test of various netbook distros on power. When people see that there is a huge difference between distros (since I think it's not simply the kernel's fault) maybe 'some' will try to rectify the situation...