1235910 - Investigate discrepancy between e10s & non-e10s responsiveness in Beta 44

Status: RESOLVED INVALID

(Firefox :: General, defect, P1)

Description

[Vladan Djeric (:vladan)]

The BHR hang rate for e10s is twice that of non-e10s, even for profiles without addons, and despite the BHR fix in bug 1234618, and despite the lack of child-process BHR reporting (bug 1228437):

https://github.com/poiru/e10s_analyses/blob/beta/beta/e10s_experiment_without_extensions.ipynb

Median difference in hangs over 100ms per minute is 19.84 (e10s 38.63 hangs/minute, non-e10s 18.79 hangs/minute)

Comment 1

[Vladan Djeric (:vladan)]

As, before we similar relative trends for profiles with at least 1 extension & only whitelisted extensions:

https://github.com/poiru/e10s_analyses/blob/beta/beta/e10s_experiment_with_extensions.ipynb
https://github.com/poiru/e10s_analyses/blob/beta/beta/e10s_experiment_with_whitelisted_extensions.ipynb

Summary of hang rates here:
https://docs.google.com/spreadsheets/d/1fUgRxStBLdwpv5bhqMP4OgRFvmsa6JKTVMLI7MM44pw/edit#gid=0

Comment 2

[Vladan Djeric (:vladan)]

We're going to need to continue validating the BHR measurement and I've also asked Birunthan to re-run these analyses with more data.

OThers can help us verify that BHR corresponds to reality by installing this toolbar extension: https://github.com/chutten/statuser/blob/gh-pages/dist/statuser-0.0.3.xpi It increments a counter whenever BHR records an event lasting more than 100ms on the main thread.

Comment 3

[Bill McCloskey [inactive unless it's an emergency] (:billm)]

It looks like the problem here is that bug 1234618 didn't land on beta when we thought it did. Thus, the BHR measurements there are not valid.

I wasn't able to tell which buildids are included in the Aurora analyses in comment 1, but I suspect they include builds lacking bug 1234618. So that data is also invalid.

I ran a simple analysis on Aurora including only buildids with the patch in bug 1234618. It's not as statistically valid because I just broke people up based on whether they have e10s enabled or not (i.e., they're allowed to choose whether to enable e10s, rather than Mozilla choosing for them). I also excluded people with add-ons.

http://people.mozilla.org/~wmccloskey/aurora-analysis.html

e10s: 1.43 hangs >100ms per minute
non-e10s: 11.35 hangs >100ms per minute

Comment 4

[Brad Lassey [:blassey] (use needinfo?)]

Ritu, which beta did this get into?

Comment 5

[Ritu Kothari (:ritu) (Inactive, please n-i to RyanVM, jcristau, or pascal)]

(In reply to Brad Lassey [:blassey] (use needinfo?) from comment #4)
> Ritu, which beta did this get into?

Hi Brad, if you are asking about the patch from bug 1234618 (cset https://hg.mozilla.org/releases/mozilla-beta/rev/2e553b97f22f), it landed in m-b on 12/29/15 and so should have been in 44.0b4.

Comment 6

[Brad Lassey [:blassey] (use needinfo?)]

Vladan, what are the next steps here?

Comment 7

[Vladan Djeric (:vladan)]

(In reply to Ritu Kothari (:ritu) from comment #5)
> (In reply to Brad Lassey [:blassey] (use needinfo?) from comment #4)
> > Ritu, which beta did this get into?
> 
> Hi Brad, if you are asking about the patch from bug 1234618 (cset
> https://hg.mozilla.org/releases/mozilla-beta/rev/2e553b97f22f), it landed in
> m-b on 12/29/15 and so should have been in 44.0b4.

Ritu: your link says this landed on m-b on December 31st, no? If that's the case, wouldn't it have only made it into beta5 or beta6 instead?

Comment 8

[Ritu Kothari (:ritu) (Inactive, please n-i to RyanVM, jcristau, or pascal)]

(In reply to Vladan Djeric (:vladan) -- please needinfo from comment #7)
> (In reply to Ritu Kothari (:ritu) from comment #5)
> > (In reply to Brad Lassey [:blassey] (use needinfo?) from comment #4)
> > > Ritu, which beta did this get into?
> > 
> > Hi Brad, if you are asking about the patch from bug 1234618 (cset
> > https://hg.mozilla.org/releases/mozilla-beta/rev/2e553b97f22f), it landed in
> > m-b on 12/29/15 and so should have been in 44.0b4.
> 
> Ritu: your link says this landed on m-b on December 31st, no? If that's the
> case, wouldn't it have only made it into beta5 or beta6 instead?

Hi Vladan, since this was pushed on 12/31 and we skipped b5 in 44 cycle, this should have been in 44.0b6.