1073152 - 4% Linux ts, paint regression on inbound (v.35) sept 22 from push 2d6ffa903e38

Status: RESOLVED WONTFIX

(Core :: Audio/Video, defect)

Description

[Joel Maher ( :jmaher ) (UTC -8)]

ts, paint is a noisy test, but the trends show a pattern:
http://graphs.mozilla.org/graph.html#tests=%5B%5B83,131,33%5D,%5B83,131,35%5D,%5B83,131,25%5D,%5B83,131,31%5D%5D&sel=1411068232590,1411673032590&displayrange=7&datatype=running

for linux 32/64 and win7/8, there is a smaller regression seen around Sep 22.  I did some retriggers to understand this better:
https://tbpl.mozilla.org/?tree=Mozilla-Inbound&jobname=Ubuntu%20HW%2012.04%20mozilla-inbound%20talos%20other_nol64&startdate=2014-09-22&enddate=2014-09-23

the retriggers on linux32 help point out https://hg.mozilla.org/integration/mozilla-inbound/pushloghtml?changeset=2d6ffa903e38 as the culprit.

Comment 1

[Joel Maher ( :jmaher ) (UTC -8)]

This has been idle for a couple of week, Sotara, can you take a look at this?

Comment 2

[Sotaro Ikeda [:sotaro]]

(In reply to Joel Maher (:jmaher) from comment #1)
> This has been idle for a couple of week, Sotara, can you take a look at this?

jmaher, why do you think  Bug 1052206 cause the regression? It made very little change only to audio playback. I do not think it could cause paint regression.

Comment 3

[Joel Maher ( :jmaher ) (UTC -8)]

Sotaro, I did a lot of retriggers on your push and before/after:
https://tbpl.mozilla.org/?tree=Mozilla-Inbound&jobname=Ubuntu%20HW%2012.04%20mozilla-inbound%20talos%20other_nol64&startdate=2014-09-22&enddate=2014-09-23

you can see the values for ts_paint before your push have a good sample of reported numbers <930 with a range of 911-950, but starting with your push they stay in the 930-960 range.  This isn't an exact science.

the push prior to yours was a mochitest test change.  If it isn't your change then it could be the one after your push:
https://hg.mozilla.org/integration/mozilla-inbound/rev/0c8fa599e889

either way, we see an increase in these times.  The test itself measure the overall time to start the browser.

Comment 4

[Hannes Verschore [:h4writer]]

The revision you are talking about, doesn't change anything for x86/x64. It only disables a speed improvement added a few days earlier for ARM chips.

IIUC you are measuring on x86/x64. So patch is a NOP.

Comment 5

[Joel Maher ( :jmaher ) (UTC -8)]

this looks like it is on beta now!  From further analysis it seems to be related to win7 ts,paint as well.

Comment 6

[Joel Maher ( :jmaher ) (UTC -8)]

this is well over 6 months with no activity- we have already shipped with this.