Project Candle bug triage (11th October 2015)

[Nicholas Nethercote]

Greetings,

This is Project Candle bug triage email. The process used is described at
https://wiki.mozilla.org/Performance/Project_Candle#Bug_triage_process.

At the time of writing there are 27 unprioritized Power bugs, viewable at
http://mzl.la/1KuHf1J. In this thread I am discussing 10 of them.


* https://bugzilla.mozilla.org/show_bug.cgi?id=571394
* Clumping wakeups in TimerThread can reduce wakeups

This is an old idea, and one of those devil-in-the-detail ones. It's a great
idea if it can be done without breaking the web. I'm not sure how hard that is.

Any suggestions?


* https://bugzilla.mozilla.org/show_bug.cgi?id=894372
* Gstreamer backend don't use hardware accelerate (VAAPI)

I suspect bsmedberg would suggest P3 for this because it's Linux, and Linux
laptops are a tiny fraction of our userbase. Any disagreements?


* https://bugzilla.mozilla.org/show_bug.cgi?id=926215
* Geeksphone Keon runs out of power within 24 hours

The configuration and data in this bug is ancient history. I suggest closing
it.


* https://bugzilla.mozilla.org/show_bug.cgi?id=947942
* Evaluate power consumption of HwComposer and OpenGL composition on gonk

I don't anything to say about this one.

P2?


* https://bugzilla.mozilla.org/show_bug.cgi?id=952238
* [meta] Reduce power consumption during audio P2P calls on FxOS

Sounds bad but there's not much data. Marked as a meta-bug but there are no
blocking bugs.

P2 for want of a better suggestion?


* https://bugzilla.mozilla.org/show_bug.cgi?id=958566
* gfxFontCache::WordCacheExpirationTimerCallback causes activity during idle

I suggest P3 because this code isn't enabled at the moment. It's assigned to
jdaggett.


* https://bugzilla.mozilla.org/show_bug.cgi?id=961475
* High CPU usage on mynoise.net/NoiseMachines/whiteNoiseGenerator.php

I suggest P3 because the comments indicate that the site it doing things in a
very sub-optimal way, i.e. it's more of an evangelism bug.


* https://bugzilla.mozilla.org/show_bug.cgi?id=965392
* BackgroundHangManager causes many wakeups

I couldn't see any sign of BackgroundHangManager when I tried profiling this.
I've needinfo'd rvitillo.

I suggest we close this unless rvitillo can provide additional information.


* https://bugzilla.mozilla.org/show_bug.cgi?id=993775
* High CPU usage on a jsfiddle example

On Mac we're slightly worse than Chrome and a lot worse than Safari. Despite
being on jsfiddle it seems to be HTML/CSS/painting stuff that's the problem.

P2? I have no sense of what the problem here is and therefore how important it
is.


* https://bugzilla.mozilla.org/show_bug.cgi?id=1211565
* Unnecessary repainting on sohu.com (#8 site in China)

This is a spin-off from bug 984506, which is about general power consumption on
sohu.com, which is a Power:P1.

I suggest making this a P1 as well, given that this looks to be most of the
problem. Having two P1 bugs that are possibly about the same issue isn't ideal
but I can live with it.

[Richard Newman]

* https://bugzilla.mozilla.org/show_bug.cgi?id=961475
* High CPU usage on mynoise.net/NoiseMachines/whiteNoiseGenerator.php

I suggest P3 because the comments indicate that the site it doing things in a
very sub-optimal way, i.e. it's more of an evangelism bug.

I'm a myNoise supporter, so I'm happy to point this out to Stephane and try to move this forward. P3 sounds fine.

Agreed on all of your other resolutions.

[Kan-Ru Chen (陳侃如)]

+1

[Gabriele Svelto]

On 12/10/2015 02:53, Nicholas Nethercote wrote:
> * https://bugzilla.mozilla.org/show_bug.cgi?id=571394
> * Clumping wakeups in TimerThread can reduce wakeups
>
> This is an old idea, and one of those devil-in-the-detail ones. It's a great
> idea if it can be done without breaking the web. I'm not sure how hard that is.
>
> Any suggestions?

I filed a similar bug some time ago [1] but never got around prototyping
a solution. It might be worth testing first how many timers we could
aggregate in common usage scenarios before deciding if we want to pursue
this.

> * https://bugzilla.mozilla.org/show_bug.cgi?id=926215
> * Geeksphone Keon runs out of power within 24 hours
>
> The configuration and data in this bug is ancient history. I suggest closing
> it.

Closing is fine, this is most likely a dup of one of the many
battery-draining bugs we've had in FxOS.

> * https://bugzilla.mozilla.org/show_bug.cgi?id=947942
> * Evaluate power consumption of HwComposer and OpenGL composition on gonk
>
> I don't anything to say about this one.
>
> P2?

Tests we did on the leo devices suggested that using the HWC during
video playback cut power consumption by 20% which is *very* significant.
Current hardware trends [2] suggest that this gap is only going to widen
so I don't think we'll ever be able to afford not using it w/o suffering
significantly lower battery life than competing Android devices.

> * https://bugzilla.mozilla.org/show_bug.cgi?id=965392
> * BackgroundHangManager causes many wakeups
>
> I couldn't see any sign of BackgroundHangManager when I tried profiling this.
> I've needinfo'd rvitillo.
>
> I suggest we close this unless rvitillo can provide additional information.

The BHM caused a significant battery regression on FxOS some time ago
which we tracked down in bug 1173140 [3] (see comment 15) and which we
fixed by disabling the BHM entirely in gonk (bug 1180533 [4]). However
in my testing I noticed that some of the BHM's clients were abusing it's
API causing unnecessary wakeups so I've filed bug 1180677 [5] to
investigate further. I never had the time to test it again but I think
this is worth investigating.

Gabriele

[1] Evaluate aggregating timers to reduce the number of times we wake up
the CPU for serving them
https://bugzilla.mozilla.org/show_bug.cgi?id=1107626

[2] Modern PowerVR GPUs now have fully dedicated paths for compositing
and in some ARM Mali offerings the compositor is being moved into the
display controller so that one can turn off the entire GPU when only the
HWC is being used).

[3] [Shinano][Aries] Battery life regression
https://bugzilla.mozilla.org/show_bug.cgi?id=1173140#c15

[4] Disable BackgroundHangMonitor on gonk
https://bugzilla.mozilla.org/show_bug.cgi?id=1180533

[5] Verify that BackgroundHangMonitor clients are not abusing it
https://bugzilla.mozilla.org/show_bug.cgi?id=1180677

[Andrew McCreight]

On Sun, Oct 11, 2015 at 5:53 PM, Nicholas Nethercote <n.neth...@gmail.com> wrote:

Greetings,

This is Project Candle bug triage email. The process used is described at
https://wiki.mozilla.org/Performance/Project_Candle#Bug_triage_process.

At the time of writing there are 27 unprioritized Power bugs, viewable at
http://mzl.la/1KuHf1J. In this thread I am discussing 10 of them.

* https://bugzilla.mozilla.org/show_bug.cgi?id=571394
* Clumping wakeups in TimerThread can reduce wakeups

This is an old idea, and one of those devil-in-the-detail ones. It's a great
idea if it can be done without breaking the web. I'm not sure how hard that is.

Any suggestions?

This can probably be a P2. It would help, but I wouldn't imagine this is going to be in the same ballpark of power usage as our various "we're drawing something invisible at 60fps" issues.
 

* https://bugzilla.mozilla.org/show_bug.cgi?id=894372
* Gstreamer backend don't use hardware accelerate (VAAPI)

I suspect bsmedberg would suggest P3 for this because it's Linux, and Linux
laptops are a tiny fraction of our userbase. Any disagreements?

* https://bugzilla.mozilla.org/show_bug.cgi?id=926215
* Geeksphone Keon runs out of power within 24 hours

The configuration and data in this bug is ancient history. I suggest closing
it.

* https://bugzilla.mozilla.org/show_bug.cgi?id=947942
* Evaluate power consumption of HwComposer and OpenGL composition on gonk

I don't anything to say about this one.

P2?

This bug is super old and the last comment says it isn't as bad any more. I'd needinfo Sotaro to see if it can be closed.
 

* https://bugzilla.mozilla.org/show_bug.cgi?id=952238
* [meta] Reduce power consumption during audio P2P calls on FxOS

Sounds bad but there's not much data. Marked as a meta-bug but there are no
blocking bugs.

P2 for want of a better suggestion?

I would just close this as incomplete, if nobody replies after posting in the bug if it can be closed. This bug is 2 years old, and it looks like nobody did anything except a very basic initial investigation. It doesn't seem like a priority for anybody. Although it might be interesting to get at least a basic sense of WebRTC power usage on desktop, where we have some tools to measure power easily at hand.
 

* https://bugzilla.mozilla.org/show_bug.cgi?id=958566
* gfxFontCache::WordCacheExpirationTimerCallback causes activity during idle

I suggest P3 because this code isn't enabled at the moment. It's assigned to
jdaggett.

* https://bugzilla.mozilla.org/show_bug.cgi?id=961475
* High CPU usage on mynoise.net/NoiseMachines/whiteNoiseGenerator.php

I suggest P3 because the comments indicate that the site it doing things in a
very sub-optimal way, i.e. it's more of an evangelism bug.

* https://bugzilla.mozilla.org/show_bug.cgi?id=965392
* BackgroundHangManager causes many wakeups

I couldn't see any sign of BackgroundHangManager when I tried profiling this.
I've needinfo'd rvitillo.

I suggest we close this unless rvitillo can provide additional information.

* https://bugzilla.mozilla.org/show_bug.cgi?id=993775

* High CPU usage on a jsfiddle example

On Mac we're slightly worse than Chrome and a lot worse than Safari. Despite
being on jsfiddle it seems to be HTML/CSS/painting stuff that's the problem.

P2? I have no sense of what the problem here is and therefore how important it
is.


* https://bugzilla.mozilla.org/show_bug.cgi?id=1211565
* Unnecessary repainting on sohu.com (#8 site in China)

This is a spin-off from bug 984506, which is about general power consumption on
sohu.com, which is a Power:P1.

I suggest making this a P1 as well, given that this looks to be most of the
problem. Having two P1 bugs that are possibly about the same issue isn't ideal
but I can live with it.

I think that sounds good.
 

_______________________________________________
dev-power mailing list
dev-...@lists.mozilla.org
https://lists.mozilla.org/listinfo/dev-power

[Nicholas Nethercote]

Thank you to all those who responded to this thread.

On Sun, Oct 11, 2015 at 5:53 PM, Nicholas Nethercote
<n.neth...@gmail.com> wrote:
>
> * https://bugzilla.mozilla.org/show_bug.cgi?id=571394
> * Clumping wakeups in TimerThread can reduce wakeups
>
> This is an old idea, and one of those devil-in-the-detail ones. It's a great
> idea if it can be done without breaking the web. I'm not sure how hard that is.
>
> Any suggestions?

P2 per mccr8's suggestion.

> * https://bugzilla.mozilla.org/show_bug.cgi?id=894372
> * Gstreamer backend don't use hardware accelerate (VAAPI)
>
> I suspect bsmedberg would suggest P3 for this because it's Linux, and Linux
> laptops are a tiny fraction of our userbase. Any disagreements?

P3 because nobody disagreed.

> * https://bugzilla.mozilla.org/show_bug.cgi?id=926215
> * Geeksphone Keon runs out of power within 24 hours
>
> The configuration and data in this bug is ancient history. I suggest closing
> it.

gsvelto agreed. Closed.

> * https://bugzilla.mozilla.org/show_bug.cgi?id=947942
> * Evaluate power consumption of HwComposer and OpenGL composition on gonk
>
> I don't anything to say about this one.
>
> P2?

gsvelto said it was important, but mccr8 thought not and suggested
asking Sotaro in the bug. I've done that and left it unprioritized for
now

> * https://bugzilla.mozilla.org/show_bug.cgi?id=952238
> * [meta] Reduce power consumption during audio P2P calls on FxOS
>
> Sounds bad but there's not much data. Marked as a meta-bug but there are no
> blocking bugs.
>
> P2 for want of a better suggestion?

I resolved it as INCOMPLETE as per mccr8's suggestion.

> * https://bugzilla.mozilla.org/show_bug.cgi?id=958566
> * gfxFontCache::WordCacheExpirationTimerCallback causes activity during idle
>
> I suggest P3 because this code isn't enabled at the moment. It's assigned to
> jdaggett.

I did P3 because nobody commented on this one.

> * https://bugzilla.mozilla.org/show_bug.cgi?id=961475
> * High CPU usage on mynoise.net/NoiseMachines/whiteNoiseGenerator.php
>
> I suggest P3 because the comments indicate that the site it doing things in a
> very sub-optimal way, i.e. it's more of an evangelism bug.

This was resolved WORKSFORME during the week, thanks to rnewman's efforts.

> * https://bugzilla.mozilla.org/show_bug.cgi?id=965392
> * BackgroundHangManager causes many wakeups
>
> I couldn't see any sign of BackgroundHangManager when I tried profiling this.
> I've needinfo'd rvitillo.
>
> I suggest we close this unless rvitillo can provide additional information.

I ended up resolving it WORKSFORME, but there's some discussion in
comment 10 that might be worth taking a look at w.r.t. which channels
we should have the BHM enabled on.

> * https://bugzilla.mozilla.org/show_bug.cgi?id=993775
> * High CPU usage on a jsfiddle example
>
> On Mac we're slightly worse than Chrome and a lot worse than Safari. Despite
> being on jsfiddle it seems to be HTML/CSS/painting stuff that's the problem.
>
> P2? I have no sense of what the problem here is and therefore how important it
> is.

I did P2.

> * https://bugzilla.mozilla.org/show_bug.cgi?id=1211565
> * Unnecessary repainting on sohu.com (#8 site in China)
>
> This is a spin-off from bug 984506, which is about general power consumption on
> sohu.com, which is a Power:P1.
>
> I suggest making this a P1 as well, given that this looks to be most of the
> problem. Having two P1 bugs that are possibly about the same issue isn't ideal
> but I can live with it.

I did P1.