I’ve been playing around with my Gutsy install on the laptop. Since it uses a tickless kernel (2.6.22), I decided to give powertop a chance and see what’s ticking away my computer. To do a somewhat informed analysis, I tried a couple of scenarios, all with wireless transmitting and without to see how that affects things. Here is what I’ve found so far, in a couple of powertop pastes:

Scenario 1. Default boot with no wireless data:

Wakeups-from-idle per second : 117.8    interval: 15.0s
Power usage (ACPI estimate): 18.7W (2.5 hours)
Top causes for wakeups:
52.5% ( 60.2)       uhci_hcd:usb1, yenta, fglrx
19.1% ( 21.9)       uhci_hcd:usb2, ipw3945, HDA Intel

Note: My how usb misbehaves even with nothing plugged in. And what is yenta doing there, since I have no card in the pcmcia slot ? Also, somebody give fglrx a valium.

Scenario 2. Default boot - wireless full steam

Wakeups-from-idle per second : 482.7    interval: 15.0s
Power usage (ACPI estimate): 20.1W (2.3 hours)
Top causes for wakeups:
87.4% (881.6)      uhci_hcd:usb2, ipw3945, HDA Intel
6.0% ( 60.2)        uhci_hcd:usb1, yenta, fglrx

As expected, full steam wireless transfers brings up the wakeup count.
Powertop optimizations: powertop provides a series of suggestions that supposedly save power. Here is what powertop suggested and I willingly complied to:

• Wireless powersaving
• Usb autosuspend
• Vm dirty writeback from 5 to 15 seconds
• laptopmode 5
• Remount the filesystem with noatime

And next is what I found.

Scenario 3. Powertop optimizations and no wireless

Wakeups-from-idle per second : 111.2    interval: 15.0s
Power usage (ACPI estimate): 17.4W (2.6 hours)
Top causes for wakeups:
48.8% ( 60.2)      : uhci_hcd:usb1, yenta, fglrx
33.0% ( 40.7)      : uhci_hcd:usb2, ipw3945, HDA Intel

After all the optimizations, only a few less wakeups per second. That weird, because usb and yenta are still up there, along with wireless (ipw3945) and sound (HDA Intel) for some odd reason. fglrx is on steroids.

Scenario 4. Powertop optimizations and wireless

Wakeups-from-idle per second : 367.0    interval: 15.0s
Power usage (ACPI estimate): 19.1W (2.4 hours)
Top causes for wakeups:
83.3% (599.7)       : uhci_hcd:usb2, ipw3945, HDA Intel
8.4% ( 60.2)         : uhci_hcd:usb1, yenta, fglrx

Now here is something interesting. Probably wireless powersaving, but it saves around 115 wakeups per second. This is nice.
Afterwards I started removing stuff that just seams to take up space.

rmmod uhci_hcd pcmcia yenta_socket

This is because I don’t have pcmcia cards or usb stuff connected to the laptop.

Scenario 5. Powertop optimizations, Modules removed, no wireless

Wakeups-from-idle per second : 96.5     interval: 15.0s
Power usage (ACPI estimate): 17.1W (2.6 hours)
Top causes for wakeups:
63.1% ( 60.3)      : fglrx
14.9% ( 14.3)      : ipw3945, HDA Intel

Now we are getting somewhere. About 20 wakepus less. That seems to be misbehaving modules (pcmcia, usb). fglrx is still jumpy (how sad).

Scenario 6. Powertop optimizations, Modules removed, Wireless

Wakeups-from-idle per second : 259.7    interval: 15.0s
Power usage (ACPI estimate): 18.2W (2.4 hours)
Top causes for wakeups:
73.4% (324.5)      : ipw3945, HDA Intel
13.6% ( 60.2)       : fglrx

Now that is a boost. About 100 less wakeups per second. Strange, but nice. What does usb/pcmcia have to do with wireless transfers ?
So I took it another notch and removed the two Intel modules - ipw3945 and HDA Intel (sound).

rmmod snd_hda_intel ipw3945

Scenario 7. Powertop opts, more modules removed, wireless removed

Wakeups-from-idle per second : 69.6     interval: 15.0s
Power usage (ACPI estimate): 17.6W (1.8 hours)
Top causes for wakeups:
84.2% ( 60.2)     : fglrx

Nice, but maybe just for offline usage. Comparing this with the other results, well, it seems that even on idle wireless takes up space. Down fglrx, down boy!

Now here is the final twist : Stopped X and remove that nasty fglrx

Scenario 8. Powertop opts, modules removed, wireless removed, fglrx removed

Wakeups-from-idle per second :  4.8    interval: 15.0s
Power usage (ACPI estimate): 19.3W (1.6 hours)
Top causes for wakeups:
23.3% (  1.3)   kernel core: queue_delayed_work_on
22.1% (  1.3)   acpi

A comfortable 4.8 wakeups per second. I just wish my laptop was like this while on idle (with all my apps running) …

Scenario 9. Powertop opts, clean boot, wireless stopped, running the vesa driver instead of fglrx:

Wakeups-from-idle per second :  6.1     interval: 15.0s
Power usage (ACPI estimate): 15.3W (3.0 hours)

That’s really nice. On a clean boot, with vesa, wakepus go down. Nor USB nor pcmcia stick out like a sore thumb. It seems almost magical. Those 6 wakeups are with X server running. But this made me wonder. Did I foul up with fglrx ? I redid the test, in the same conditions as the previous (#9), but with fglrx.

Scenario 10. Powertop opts, clean boot, wireless stopped, fglrx:

Wakeups-from-idle per second : 69.8     interval: 15.0s
Power usage (ACPI estimate): 16.8W (2.7 hours)

That’s more than a 1W of power draining towards fglrx. I noticed that powertop seems to bundle things. Maybe the yenta stuff is fine, and fglrx reports it with along with flgrx, since it probably can notice the difference from the bunch who is causing the wakeps, just like ipw3945 and the snd_hda modules.

Final thoughts

• Intel, what a twist! Powertop is really useful,. But in a twist of fate, the modules that take up more wakeups, and keep my intel cpu from going into C3 state are Intel modules. This could should be fixed. This might be a buggy wireless driver. Also there can be lack of means for powertop to tell the difference between ipw3945 from snd_hda_intel wakepus, since using just the snd_hda_intel module (with no wireless), the wakeups stop.
• I noticed that using the touchpad or the keyboard yield about 500-800 wakepus per second. Is this really necessary ?! Or is there a fix for this. I seen a fix for the macbook on www.linuxpowertop.org, maybe my thouchpad needs something of the kind… Here is a sample:

47.4% (494.2)      PS/2 keyboard/mouse/touchpad

• The ATI Video Driver (fglrx) is high on caffeine. It does a constant 60 wakeups per second. It’s not on behalf of any X app, because I killed most of them while testing it (by advice of powertop’s online tips). There is probably a bug in there somewhere that makes this very annoying and jumpy. But I haven’t complained, since this might have a _solution_ in the near future… Going with Vesa saves me over a 1 Watt of power.

This is just using stock ubuntu packages. I’m still to follow some of powertop’s tips, that are online. Really nice when running powertop+vesa driver, this really gives me more battery time. I started out with 18.7 and ended up with 16.8, almost 2 watts of power, which is something to notice.

UPDATE:

Updating the fglrx binary driver to the lastest version 8.41 fixes the jumpiness and dramatically cuts down the wakeups per second. Don’t believe me ? Here’s the result with all the powertop suggestions activated:

Wakeups-from-idle per second :  7.5     interval: 30.0s

Pretty damn sweet! Bring on the Gutsy Gibbon.