305356 - Tp (pageload) performance regression between 26-July-2005 07:00 and 09:00

Status: RESOLVED FIXED

(Core :: Graphics: ImageLib, defect)

Description

[Joonas]

Pageload times for pacifica, beast and creature all increased in July 26:
http://build-graphs.mozilla.org/graph/multiquery.cgi?tboxes=pacifica%2Cbeast%2Ccreature&testname=pageload&points=&autoscale=1&units=&ltype=&avg=1&days=40

Looking at tinderbox I think the regression window is around 2005-07-26 07:50
and 08:50: 
http://tinderbox.mozilla.org/showbuilds.cgi?tree=Firefox&hours=1&maxdate=1122392800&legend=0

Here are the checkins between 07:50 and 08:50:  
http://bonsai.mozilla.org/cvsquery.cgi?treeid=default&module=all&branch=HEAD&branchtype=match&dir=&file=&filetype=match&who=&whotype=match&sortby=Date&hours=2&date=explicit&mindate=2005-07-26+07%3A50&maxdate=2005-07-26+08%3A50&cvsroot=%2Fcvsroot

Bug 268798 and possibly bug 297131 look suspicious.

Bug 299231, bug 275960 and bug 293388 were checked in after 08:40, and I think
beast regressed before that.

Comment 1

[Joonas]

First trying bug 268798, which might have caused this Tp regression. Cc'ing
people from that bug.

Comment 2

[Mats Palmgren (inactive)]

Boris, can you look at this Tp regression please?
My guess would be bug 293388.

Comment 3

[Peter van der Woude [:Peter6]]

(In reply to comment #2)
> Boris, can you look at this Tp regression please?
> My guess would be bug 293388.

Mats, read https://bugzilla.mozilla.org/show_bug.cgi?id=268798#c11
Bug 268798 was the only checkin for beast for that build and that's when Tp +
started

Comment 4

[Mats Palmgren (inactive)]

Ok, I don't know to what degree one can trust the local clocks on these machines
though, are all machines within mozilla.org running NTP?
My impression from using Bugzilla/Bonsai/Tinderbox/CVS is that clocks can be
out-of-sync by several minutes.

Comment 5

[David Baron :dbaron: (⌚️UTC-4, no longer working on Mozilla)]

You can always look at the checkin log to see what they pulled.

Comment 6

[David Baron :dbaron: (⌚️UTC-4, no longer working on Mozilla)]

The build log, I mean.

Comment 7

[Doug Turner (:dougt)]

I suggest that someone (me maybe) back out each checkin locally on the tbox to
identify exactly which check cause this (or if any did).

Comment 8

[Boris Zbarsky [:bzbarsky]]

Given that this increase only happened on Windows tinderboxes, I'd say that
local backout (as a test) of bug 268798 is indeed the place to start.

Comment 9

[Joonas]

Doug T, do you have time to test if local backout of bug 268798 fixes this
10-13% Tp regression? Thanks

Comment 10

[Boris Zbarsky [:bzbarsky]]

Looks like that backout helps.  r+sr=bzbarsky on the backout.

Comment 11

[Asa Dotzler [:asa]]

a=asa

Comment 12

[Doug Turner (:dougt)]

Backed out from 1.8:
Checking in modules/libpr0n/src/imgCache.cpp;
/cvsroot/mozilla/modules/libpr0n/src/imgCache.cpp,v  <--  imgCache.cpp
new revision: 1.18.2.1; previous revision: 1.18
done

Backed out from trunk:
Checking in modules/libpr0n/src/imgCache.cpp;
/cvsroot/mozilla/modules/libpr0n/src/imgCache.cpp,v  <--  imgCache.cpp
new revision: 1.19; previous revision: 1.18
done

Comment 13

[Peter van der Woude [:Peter6]]

verified Tp inprovement on branch (Pacifica)
-> fixed1.8