Closed Bug 1311593 Opened 8 years ago Closed 3 years ago

WebRTC churns on CPU on Mac Book Pro

Categories

(Core :: WebRTC, defect, P1)

defect

Tracking

()

RESOLVED INCOMPLETE

People

(Reporter: Harald, Unassigned)

References

(Blocks 1 open bug)

Details

(Keywords: stale-bug)

MacBook Pro (Retina, 15-inch, Early 2013), macOS Sierra

Nightly (52.0a1 (2016-10-19) (64-bit))

Had a call on https://talky.io/ with one participant, causing my CPU to churn and fan spinning. Few tabs open but all inactive because I just reopened my browser. I started the Gecko Profiler and recorded a few seconds:

https://cleopatra.io/#report=29ff32294befd1f9effcfea497a8456b61bf0374
Follow up from our IRC convo.
Flags: needinfo?(mreavy)
Thanks, Harald, for follow up and filing.  

Nils -- When you get a chance (I know you're still at the conference), can you work with Harald to get a more detailed profile of what he's seeing -- or reproduce the issue yourself?  Thanks.
Flags: needinfo?(mreavy) → needinfo?(drno)
Rank: 19
Priority: -- → P1
So I looked at the cleopatra.io report, but even though it shows some sparks of high CPU usage it's not clear what's going on.

Talked to Harald briefly over lunch about this. He has so far only experienced it once. The other end of the call was on Talky.io iOS client.

I'll try to repro myself some time later this week.
Flags: needinfo?(drno)
Assignee: nobody → drno
cleopatra doesn't record most WebRTC CPU use
Harald do you know if e10s was on or off during your call?
Flags: needinfo?(hkirschner)
I tested around a little bit between my MacBook Pro and the talky.io app on my iPhone.

The CPU usage according to macOS Activity Monitor usually somewhere between 70 - 150%, which the higher end being the exception. I noticed that when I have e10s turned on, that my CPU fan kicks in as well, but that is probably just because the total CPU load with other apps running was somewhere above 200%. Nothing unexpected so far.

Now what I find strange is that when I join a room on my MacBook the CPU usage goes immediately close to 100% and slightly above, even though the call is not connected to anywhere and about:webrtc does not report any PeerConnection. That CPU load does not change at all when another party actually joins the room. I'm guessing that is caused by the extensive graphics takly uses for showing the rocket game while you are waiting for the other party to join?!
e10s was on, one content process
Flags: needinfo?(hkirschner)
Mass change P1->P2 to align with new Mozilla triage process
Priority: P1 → P2
Most likely the use of the integrated Intel graphics cards is causing most of this problem. Switching to the dedicated card in MBP's helps.

But I don't have the capacity any more to work on this any time soon.
Assignee: drno → nobody
Blocks: webrtc-perf

Setting this to P1 for investigation and analysis so we can determine its true impact to the user.

Priority: P2 → P1

Marking this resolved/incomplete after no relevant updates in 3 years.

Status: NEW → RESOLVED
Closed: 3 years ago
Resolution: --- → INCOMPLETE
You need to log in before you can comment on or make changes to this bug.