All of lore.kernel.org
 help / color / mirror / Atom feed
* Re: [Powertop] v2.5-rc1 testing: accounting still very wrong
@ 2013-11-02 13:48 Marc MERLIN
  0 siblings, 0 replies; 3+ messages in thread
From: Marc MERLIN @ 2013-11-02 13:48 UTC (permalink / raw)
  To: powertop

[-- Attachment #1: Type: text/plain, Size: 6779 bytes --]

On Fri, Oct 18, 2013 at 11:09:26AM -0700, Kristen Carlson Accardi wrote:
> Please test.

Still the same power counting bugs than previous versions.
After a few minutes, on battery, it went from 12W to 20W because of an
rsync in the background, and when that rsync was finished (sigstop), I got this.

Somehow I very much doubt that e17 was using 1550W.

Clearly there is at least an assert missing that should ensure that the list
of power estimated values clearly exceeds the total power.

Even if I stop enligthenment, which seems to be most of the wakeups, bad
accounting is all wrong in many ways, see below:

--------------------------------------------------------------------------------
The battery reports a discharge rate of 20.3 W
The estimated remaining time is 3 hours, 48 minutes

Summary: 609.5 wakeups/second,  426.1 GPU ops/seconds, 0.0 VFS ops/sec and 5.7% CPU use

Power est.              Usage       Events/s    Category       Description
 1.55 kW     15.5 ms/s     526.8        Process        /usr/bin/enlightenment
   198 W    496.5 µs/s     295.9        Process        [btrfs-transacti]
  9.27 W    100.0%                      Device         Audio codec alsa:hwC0D0: thinkpad (Realtek)
  304 mW     22.2 µs/s       0.3        Process        xscreensaver
  242 mW      6.7%                      Device         Display backlight
 68.5 mW      5.4 ms/s      68.9        Interrupt      [42] i915(a)pci:0000:00:02.0
 62.6 mW     18.4 ms/s      32.2        Process        /usr/bin/X :0 vt7 -br -nolisten tcp -auth /var/run/x
 51.3 mW      1.1 ms/s      58.5        Process        [btrfs-endio-met]
--------------------------------------------------------------------------------

If I relaunch powertop, I now see:
--------------------------------------------------------------------------------
The battery reports a discharge rate of 18.4 W
The estimated remaining time is 4 hours, 6 minutes

Summary: 3245.9 wakeups/second,  404.1 GPU ops/seconds, 0.0 VFS ops/sec and 51.4% CPU use

Power est.              Usage       Events/s    Category       Description
  18.7 W    100.0%                      Device         Audio codec alsa:hwC0D0: thinkpad (Realtek)
  4.91 W    139.9 ms/s     3185.8       Process        /usr/bin/enlightenment
  309 mW    363.2 ms/s     401.2        Process        /usr/bin/X :0 vt7 -br -nolisten tcp -auth /var/run/x
 79.2 mW      1.9%                      Device         Display backlight
 59.0 mW      0.8 ms/s      11.8        Process        xfce4-terminal -T window13 --role=window13 --tab
 4.72 mW      5.7 ms/s       2.0        Interrupt      [42] i915(a)pci:0000:00:02.0
 653 µW     150.1 µs/s      16.8        Process        [rcu_preempt]
--------------------------------------------------------------------------------
No, alsa isn't using more battery than the total laptop either ^^^


Ok, let's wait a bit longer:
--------------------------------------------------------------------------------
The battery reports a discharge rate of 18.6 W
The estimated remaining time is 4 hours, 1 minutes

Summary: 614.7 wakeups/second,  257.4 GPU ops/seconds, 0.0 VFS ops/sec and 6.5% CPU use

Power est.              Usage       Events/s    Category       Description
  13.4 W    643.1 µs/s       7.3        Process        /usr/bin/pulseaudio --start
  287 mW     28.6 ms/s     588.0        Process        /usr/bin/enlightenment
  224 mW     26.8 ms/s     158.5        Process        /usr/bin/X :0 vt7 -br -nolisten tcp -auth /var/run/x
 58.2 mW      6.7%                      Device         Display backlight
 4.19 mW      5.6 ms/s      46.2        Interrupt      [42] i915(a)pci:0000:00:02.0
 3.54 mW     63.9 µs/s       2.1        Process        xfce4-terminal -T window13 --role=window13 --tab
 630 µW       0.8 ms/s      17.8        Interrupt      [46] snd_hda_intel
 492 µW     658.9 µs/s      26.7        Timer          tick_sched_timer
--------------------------------------------------------------------------------


Ok, so I killed pulseaudio, which I'm pretty sure is not to blame, and now I se
this. So you get the idea, 621 wakeups per second isn't good, but the accounting 
seems to be wrong.
--------------------------------------------------------------------------------
The battery reports a discharge rate of 18.8 W
The estimated remaining time is 3 hours, 57 minutes

Summary: 621.7 wakeups/second,  191.7 GPU ops/seconds, 0.0 VFS ops/sec and 7.5% CPU use

Power est.              Usage       Events/s    Category       Description
  13.4 W    100.0%                      Device         Audio codec alsa:hwC0D0: thinkpad (Realtek)
  357 mW     25.8 ms/s     620.3        Process        /usr/bin/enlightenment
 38.8 mW      6.7%                      Device         Display backlight
 20.3 mW     26.2 ms/s      35.6        Process        /usr/bin/X :0 vt7 -br -nolisten tcp -auth /var/run/x
 7.05 mW      9.4 ms/s      0.05        Process        /usr/bin/python /usr/sbin/iotop
 5.40 mW    106.8 µs/s       3.8        Process        xfce4-terminal -T window14 --role=window14 --tab
 5.09 mW      0.9 ms/s       9.7        Process        xfce4-terminal -T window11 --role=window11 --tab
 4.30 mW      5.7 ms/s      65.5        Interrupt      [42] i915(a)pci:0000:00:02.0
--------------------------------------------------------------------------------

If I kill -STOP enlightenment, I get this, 18.7W for alsa out of 15.9W total!
--------------------------------------------------------------------------------
The battery reports a discharge rate of 15.9 W
The estimated remaining time is 4 hours, 35 minutes

Summary: 59.0 wakeups/second,  0.0 GPU ops/seconds, 0.0 VFS ops/sec and 5.1% CPU use

Power est.              Usage       Events/s    Category       Description
  18.7 W    100.0%                      Device         Audio codec alsa:hwC0D0: thinkpad (Realtek)
 50.0 mW     45.8 ms/s       0.5        Process        /usr/bin/python /usr/sbin/iotop
 885 µW       0.8 ms/s      16.4        Process        [btrfs-transacti]
 430 µW     393.5 µs/s       8.6        Interrupt      [4] block(softirq)
 403 µW     369.5 µs/s       0.4        kWork          kcryptd_crypt
 302 µW     276.4 µs/s       2.7        Timer          hrtimer_wakeup
 267 µW     244.7 µs/s       2.0        Timer          tick_sched_timer
--------------------------------------------------------------------------------

-- 
"A mouse is a device used to point at the xterm you want to type in" - A.S.R.
Microsoft is to operating systems ....
                                      .... what McDonalds is to gourmet cooking
Home page: http://marc.merlins.org/                         | PGP 1024R/763BE901

^ permalink raw reply	[flat|nested] 3+ messages in thread
* Re: [Powertop] v2.5-rc1 testing: accounting still very wrong
@ 2013-11-02 19:50 Marc MERLIN
  0 siblings, 0 replies; 3+ messages in thread
From: Marc MERLIN @ 2013-11-02 19:50 UTC (permalink / raw)
  To: powertop

[-- Attachment #1: Type: text/plain, Size: 4264 bytes --]

On Sat, Nov 02, 2013 at 06:48:02AM -0700, Marc MERLIN wrote:
> On Fri, Oct 18, 2013 at 11:09:26AM -0700, Kristen Carlson Accardi wrote:
> > Please test.
> 
> Still the same power counting bugs than previous versions.
> After a few minutes, on battery, it went from 12W to 20W because of an
> rsync in the background, and when that rsync was finished (sigstop), I got this.
> 
> Somehow I very much doubt that e17 was using 1550W.
> 
> Clearly there is at least an assert missing that should ensure that the list
> of power estimated values clearly exceeds the total power.
> 
> Even if I stop enligthenment, which seems to be most of the wakeups, bad
> accounting is all wrong in many ways, see below:

And note that it's not always broken, after removing enough things,
including unloading alsa, I got this:
----------------------------------------------------------------------------
The battery reports a discharge rate of 13.0 W
The estimated remaining time is 5 hours, 12 minutes

Summary: 639.4 wakeups/second,  105.5 GPU ops/seconds, 0.0 VFS ops/sec and 8.5%

Power est.              Usage       Events/s    Category       Description
  948 mW      3.9 ms/s     178.2        Process        /usr/bin/enlightenment
  886 mW    393.4 µs/s      40.9        Process        procmeter3
  412 mW      6.4 ms/s     152.9        Process        /usr/local/opt/google/chr
  334 mW     36.7%                      Device         Display backlight
  323 mW      6.5 ms/s      88.7        Process        /usr/local/opt/google/chr
 59.2 mW     37.2 ms/s      48.9        Process        /usr/local/opt/google/chr
 25.7 mW     16.2 ms/s       9.9        Process        /usr/bin/X :0 vt7 -br -no
 10.8 mW    138.1 µs/s       1.4        Process        xfce4-terminal -T window9
 6.66 mW      4.2 ms/s       6.7        Process        /usr/local/opt/google/chr
 4.80 mW      3.0 ms/s      66.7        Timer          hrtimer_wakeup
----------------------------------------------------------------------------

The problem is that the only number I can seem to trust is the discharge
rate and likely the number of wakeups per second. The rest, especially
the power estimation value per device or process, seems to be pure
fabrication, and if you add them up, they never add up to the battery
discharge rate.

Am I really the only one to see this?

kernel 3.11.5-amd64 powertop from git synced yesterday.

And just as I'm writing this Email, the next update I get from powertop
(after restarting it because it crashed, trace sent in another email),
is the following. My backlight moved from 334mW to 8.58W (I know both are wrong).
----------------------------------------------------------------------------
The battery reports a discharge rate of 13.1 W
The estimated remaining time is 4 hours, 59 minutes

Summary: 735.1 wakeups/second,  83.1 GPU ops/seconds, 0.0 VFS ops/sec and 8.5% CPU use       

Power est.              Usage       Events/s    Category       Description
  8.58 W     40.0%                      Device         Display backlight
  992 mW    100.0%                      Device         USB device: Yubico Yubikey II (Yubico)
  761 mW    303.7 µs/s      32.7        Process        procmeter3
  455 mW      2.9 ms/s      86.8        Process        /usr/local/opt/google/chrome/chrome --
  433 mW      3.7 ms/s      46.5        Process        xfce4-terminal -T window11 --role=wind
  359 mW     11.7 ms/s     151.8        Process        /usr/bin/enlightenment
  198 mW      5.4 ms/s      98.7        Process        /usr/local/opt/google/chrome/chrome
  102 mW     27.6 ms/s      39.3        Process        /usr/local/opt/google/chrome/chrome --
 68.2 mW     18.8 ms/s      20.5        Process        /usr/bin/X :0 vt7 -br -nolisten tcp -a
 29.3 mW      1.6 ms/s      94.5        Interrupt      [42] i915(a)pci:0000:00:02.0
----------------------------------------------------------------------------

Marc
-- 
"A mouse is a device used to point at the xterm you want to type in" - A.S.R.
Microsoft is to operating systems ....
                                      .... what McDonalds is to gourmet cooking
Home page: http://marc.merlins.org/                         | PGP 1024R/763BE901

^ permalink raw reply	[flat|nested] 3+ messages in thread
* Re: [Powertop] v2.5-rc1 testing: accounting still very wrong
@ 2013-11-04 18:31 Kristen Carlson Accardi
  0 siblings, 0 replies; 3+ messages in thread
From: Kristen Carlson Accardi @ 2013-11-04 18:31 UTC (permalink / raw)
  To: powertop

[-- Attachment #1: Type: text/plain, Size: 1253 bytes --]

On Sat, 2 Nov 2013 12:50:15 -0700
Marc MERLIN <marc_powertop(a)merlins.org> wrote:

> On Sat, Nov 02, 2013 at 06:48:02AM -0700, Marc MERLIN wrote:
> > On Fri, Oct 18, 2013 at 11:09:26AM -0700, Kristen Carlson Accardi wrote:
> > > Please test.
> > 
> > Still the same power counting bugs than previous versions.
> > After a few minutes, on battery, it went from 12W to 20W because of an
> > rsync in the background, and when that rsync was finished (sigstop), I got this.
> > 
> > Somehow I very much doubt that e17 was using 1550W.
> > 
> > Clearly there is at least an assert missing that should ensure that the list
> > of power estimated values clearly exceeds the total power.
> > 
> > Even if I stop enligthenment, which seems to be most of the wakeups, bad
> > accounting is all wrong in many ways, see below:

Thanks for the detailed report.  I think we aren't going to be able to
address this for the 2.5 release, but I will put it on my list of
things to do to see if we can duplicate this issue on another machine
and start figuring out what is going on.  We'll try to put a test case
for something like this together for the 2.6 release just to confirm
that we are within the realm of possibilities for estimation.


^ permalink raw reply	[flat|nested] 3+ messages in thread

end of thread, other threads:[~2013-11-04 18:31 UTC | newest]

Thread overview: 3+ messages (download: mbox.gz follow: Atom feed
-- links below jump to the message on this page --
2013-11-02 13:48 [Powertop] v2.5-rc1 testing: accounting still very wrong Marc MERLIN
  -- strict thread matches above, loose matches on Subject: below --
2013-11-02 19:50 Marc MERLIN
2013-11-04 18:31 Kristen Carlson Accardi

This is an external index of several public inboxes,
see mirroring instructions on how to clone and mirror
all data and code used by this external index.