Mobile Users Beware: Linux Has Major Power Regression

[not.sure]

I am the Jupiter author, yes.

What does it actually do? The doc seems very, very vague. Does it do more than what powertop suggests?

[fewt]

What does it actually do? The doc seems very, very vague. Does it do more than what powertop suggests?

It implements the most common configuration changes from powertop when the power connection is broken, and sets them back to default when the connection is re-established. It also adjusts the processor governor.

It enables / disables things like SATA Link power management, USB power management, Audio card power management, scheduler power management, laptop mode (kernel parameter) etc.

With the Eee PC support package, it enables Asus's Super Hybrid Engine technology (already available in the kernel) and also overclocks / underclocks the GPU (on supported hardware).

There is plenty of documentation available in the Wiki.

I've been able to reach 11 hours on my 1015PEM, without Jupiter, 5 ish. There are thousands of users currently, it's been popular with Eee PC users for years now.

[not.sure]

It implements..

Thanks. So there's no magic involved. The eee stuff sounds nice though. I'll try that on my T91 (5h winxp vs 4h ubuntu).

[fewt]

Thanks. So there's no magic involved. The eee stuff sounds nice though. I'll try that on my T91 (5h winxp vs 4h ubuntu).

Right, no magic. Just a simple tool that takes all of the (manual) power saving best practices and automates them.

[Azpegath]

Broken sleep/hibernate

I had sleep and hibernation working on my Zepto notebook (intel graphics and dual-core cpu) after some fiddling around with swap, etc, since Ubuntu 8.10 or something like that. But Ubuntu 10.04 broke that completely and it still hasn't been fixed. I've debugged, tried looking for docs or howto:s, but it just seems like it's broken.

Perhaps it's time to give Fedora or some other distribution a shot. I've been running Gentoo for soon-to-be 10 years on my desktop computer, but all the compiling feels like a waste of time on a laptop. Talk about battery consuming =)

[evolution]

Kernel Timer Interrupts... Can they be related to the power regression issue?

Well, after recompiling my 2.6.35-lts-ck and 2.6.38-ck kernels with support for powertop (tested with version 1.13), I found the biggest difference between them being the kernel timer interrupts.

I tested this in my laptop (Core2Duo T7300@2GHz, fglrx, 2GB DDR2-800) by watching a video on youtube (with mplayer-vaapi) with google-chrome and the rest of the system "idle".

Here comes the screenshots:

2.6.35:


2.6.38:


As you can see, (ignoring the rescheduling interrupts, as they're always changing due to the system loading the video), there's a big difference in the [extra timer interrupts] section between the 2.6.35 kernel (1st image) and the 2.6.38 kernel (2nd image).

Furthermore, when I exited from youtube and put the system "idle" (only with openbox loaded), the big difference between [extra timer interrupts] in 2.6.35 and 2.6.38 didn't change significantly.

So that, after doing these tests, I'm suspecting one of the main power management issues might be related to (eventual) kernel timer interrupts code changes...

Cheers

[evolution]

p.s.: If you can't see the images uploaded on imageshack (click on them to see them in full-size):

Timer Interrupts 2.6.35:

Timer Interrupts 2.6.38:

[fewt]

Well, after recompiling my 2.6.35-lts-ck and 2.6.38-ck kernels with support for powertop (tested with version 1.13), I found the biggest difference between them being the kernel timer interrupts.

I tested this in my laptop (Core2Duo T7300@2GHz, fglrx, 2GB DDR2-800) by watching a video on youtube (with mplayer-vaapi) with google-chrome and the rest of the system "idle".

Here comes the screenshots:

2.6.35:


2.6.38:


As you can see, (ignoring the rescheduling interrupts, as they're always changing due to the system loading the video), there's a big difference in the [extra timer interrupts] section between the 2.6.35 kernel (1st image) and the 2.6.38 kernel (2nd image).

Furthermore, when I exited from youtube and put the system "idle" (only with openbox loaded), the big difference between [extra timer interrupts] in 2.6.35 and 2.6.38 didn't change significantly.

So that, after doing these tests, I'm suspecting one of the main power management issues might be related to (eventual) kernel timer interrupts code changes...

Cheers

Try repeating your test with PowerTOP 1.97

[evolution]

Try repeating your test with PowerTOP 1.97

Powertop 1.97 gives me this (do you have any idea which missing kernel config feature might be missing?):

Summary: -nan wakeups/second, -nan GPU ops/second and -nan VFS ops/sec

Cheers

[fewt]

Powertop 1.97 gives me this (do you have any idea which missing kernel config feature might be missing?):



Cheers

CONFIG_CPU_FREQ_STAT and CONFIG_X86_MSR