Talos Regression: Ts Shutdown, MAX Dirty Profile increase 32.83% on Linux Firefox3.5

[Nobody]

Regression: Ts Shutdown, MAX Dirty Profile increase 32.83% on Linux Firefox3.5

Previous results: 4988.47 from build 20091102030746 of revision 592d82bb8df8 at 2009-11-02 04:03:00 on talos-rev2-linux15 run # 0

New results: 6626.11 from build 20091102091945 of revision 40f1c74ccf91 at 2009-11-02 10:21:00 on talos-rev2-linux07 run # 0

Suspected checkin range: from revision 592d82bb8df8 to revision 40f1c74ccf91

[Mike Beltzner]

How reliable are the Ts Shutdown dirty profile tests? These are pretty big swings, and there's stuff in the regression range that looks like it could have an effect on shutdown time, specifically:

https://bugzilla.mozilla.org/show_bug.cgi?id=521750

and maybe even https://bugzilla.mozilla.org/show_bug.cgi?id=524462

These are the topcrash fixes, but wasn't expecting them to have this sort of perf hit.

cheers,

mike

_______________________________________________
dev-tree-management mailing list
dev-tree-...@lists.mozilla.org
https://lists.mozilla.org/listinfo/dev-tree-management

[Shawn Wilsher]

Everything in this range only landed on the relbranch, so it wouldn't be
tested anyway, right?

Cheers,

Shawn

On 11/3/09 4:36 AM, Nobody wrote:
> Regression: Ts Shutdown, MAX Dirty Profile increase 32.83%

> <http://graphs.mozilla.org/graph.html#tests=[{"machine":190,"test":51,"branch":3},{"machine":192,"test":51,"branch":3},{"machine":195,"test":51,"branch":3},{"machine":197,"test":51,"branch":3},{"machine":188,"test":51,"branch":3},{"machine":201,"test":51,"branch":3},{"machine":193,"test":51,"branch":3},{"machine":194,"test":51,"branch":3},{"machine":196,"test":51,"branch":3},{"machine":203,"test":51,"branch":3},{"machine":205,"test":51,"branch":3},{"machine":191,"test":51,"branch":3},{"machine":198,"test":51,"branch":3},{"machine":202,"test":51,"branch":3},{"machine":200,"test":51,"branch":3},{"machine":204,"test":51,"branch":3}]&sel=1257099660,1257272460>

> on Linux Firefox3.5
>
> Previous results: 4988.47 from build 20091102030746 of revision
> 592d82bb8df8 at 2009-11-02 04:03:00 on talos-rev2-linux15 run # 0
>
> New results: 6626.11 from build 20091102091945 of revision 40f1c74ccf91
> at 2009-11-02 10:21:00 on talos-rev2-linux07 run # 0
>
> Suspected checkin range: from revision 592d82bb8df8 to revision
> 40f1c74ccf91 <

> http://hg.mozilla.org/releases/mozilla-1.9.1/pushloghtml?fromchange=592d82bb8df8&tochange=40f1c74ccf91>

[Ted Mielczarek]

On Tue, Nov 3, 2009 at 11:41 AM, Shawn Wilsher <sdw...@mozilla.com> wrote:

Everything in this range only landed on the relbranch, so it wouldn't be tested anyway, right?

 No, pushlog doesn't differentiate by named branch currently, so the build machines will build + test things on any named branch. (bug 464029 )

-Ted

[Axel Hecht]

> <https://bugzilla.mozilla.org/show_bug.cgi?id=464029> )
>
> -Ted
>

They build, but they may update to "default" instead of given revision.
At which point they'd not build the landed change, but the previous one.
tbpl would say, though, as the ident the builds put into TinderboxPrint
won't lie, I hope.

Axel

[Justin Dolske]

On 11/3/09 7:32 AM, Mike Beltzner wrote:
> How reliable are the Ts Shutdown dirty profile tests?

From looking at the linked graph, I'd say not very reliable at all.
Actually, it looks like there's a bimodal distribution... Results seem
to bounce between either ~5000 or ~6800. The test would probably be ok
if we could make it stick to one of those levels.

Justin