355457 - Ts hit between 2006-10-04 00:00 and 2006-10-04 12:00

Status: RESOLVED WONTFIX

(Core :: General, defect)

Description

[Adam Guthrie]

There seems to have been a small increase in Tp sometime between 2006-10-04 00:00 and 2006-10-04 12:00.

Check-ins: http://bonsai.mozilla.org/cvsquery.cgi?treeid=default&module=PhoenixTinderbox&branch=HEAD&branchtype=match&dir=&file=&filetype=match&who=&whotype=match&sortby=Date&hours=2&date=explicit&mindate=2006-10-04&maxdate=2006-10-04+12&cvsroot=%2Fcvsroot

argo perf Ts: http://graphs.mozilla.org/#spst=range&spstart=1159833600&spend=1160003820&bpst=cursor&bpstart=1159833600&bpend=1160003820&m1tid=27&m1bl=0&m1avg=0
gaius perf Ts: http://graphs.mozilla.org/#spst=range&spstart=1159750080&spend=1160000820&bpst=cursor&bpstart=1159750080&bpend=1160000820&m1tid=30&m1bl=0&m1avg=0

Looking at the raw data for the graphs, the regression seems to have occurred in the time that the patch for bug 349465 landed.

Comment 1

[Adam Guthrie]

s/Tp/Ts/ in the first sentence.

Comment 2

[Peter Van der Beken [:peterv]]

That's odd, because I don't see that code running on startup.

Comment 3

[Boris Zbarsky [:bzbarsky]]

Did anything else land in the relevant time period?

And I assume the XULElement changes in bug 349465 were just moving identical code from point A to point B?

Comment 4

[Olli Pettay [:smaug][[e-mail]]]

btek seems to have some tp hit too.

Comment 5

[Peter Van der Beken [:peterv]]

(In reply to comment #3)
> Did anything else land in the relevant time period?

The only thing that could affect Tp would be bug 333535 I think.
> 
> And I assume the XULElement changes in bug 349465 were just moving identical
> code from point A to point B?

Yes.

I guess I'll try to back out tomorrow.

Comment 6

[Peter Van der Beken [:peterv]]

Backing out the patch for bug 349465 didn't have any effect on the Ts on argo and gaius or on the Tp on btek, so I guess bug 333535 should be tried next.

Comment 7

[Simon Montagu :smontagu]

Backing out the patch for bug 333535 didn't have any effect either.

Comment 8

[Adam Guthrie]

Here are the numbers from the raw data for the graphs:

argo: 
901	MOZ_CO_DATE=2006:10:04:04:04
920	MOZ_CO_DATE=2006:10:04:05:21

gaius-perf:
1641	MOZ_CO_DATE=2006:10:04:03:12
1719	MOZ_CO_DATE=2006:10:04:06:00

Check-ins during that range only include the two bugs that we backed out to see if it'd change things, so I'm really not sure what's going on here...

Here is the check-in range backed up a bit: http://bonsai.mozilla.org/cvsquery.cgi?treeid=default&module=PhoenixTinderbox&branch=HEAD&branchtype=match&dir=&file=&filetype=match&who=&whotype=match&sortby=Date&hours=2&date=explicit&mindate=2006-10-03+21&maxdate=2006-10-04+12&cvsroot=%2Fcvsroot. There were about 7 builds on argo (~4 on gaius-pref) after Seth's check-in of those images, and at least two builds on each platform after the SVG check-in that didn't show any regression in Ts times, so unless this is some sort of latent manifestation, I'm stumped.

Comment 9

[Damon Sicore (:damons)]

Anyone heard anything about this at all?  Boris, thoughts?  I'm working through the nom queue for General, and this one is pretty old. 

Comment 10

[Boris Zbarsky [:bzbarsky]]

I don't really know.

Comment 11

[Damon Sicore (:damons)]

I can't say we'd block on this.  If anyone can figure it out, it'd be nice to get that perf back.