Re: Talos Regression: Ts Shutdown, MAX Dirty Profile increase 0.94% on Leopard Firefox
Shawn Wilsher
Cheers,
Shawn
On 11/5/09 3:01 PM, Nobody wrote:
> Regression: Ts Shutdown, MAX Dirty Profile increase 0.94%
> <http://graphs.mozilla.org/graph.html#tests=[{"machine":169,"test":51,"branch":1},{"machine":170,"test":51,"branch":1},{"machine":171,"test":51,"branch":1},{"machine":172,"test":51,"branch":1},{"machine":173,"test":51,"branch":1},{"machine":174,"test":51,"branch":1},{"machine":175,"test":51,"branch":1},{"machine":176,"test":51,"branch":1},{"machine":177,"test":51,"branch":1},{"machine":178,"test":51,"branch":1},{"machine":179,"test":51,"branch":1},{"machine":180,"test":51,"branch":1},{"machine":181,"test":51,"branch":1},{"machine":182,"test":51,"branch":1}]&sel=1257358860,1257531660>
> on Leopard Firefox
>
> Previous results: 6943.47 from build 20091105080037 of revision
> d211f5c35d00 at 2009-11-05 10:12:00 on talos-rev2-leopard06 run # 0
>
> New results: 7008.68 from build 20091105082922 of revision bc3e54a0632b
> at 2009-11-05 10:21:00 on talos-rev2-leopard07 run # 0
>
> Suspected checkin range: from revision d211f5c35d00 to revision
> bc3e54a0632b <
> http://hg.mozilla.org/mozilla-central/pushloghtml?fromchange=d211f5c35d00&tochange=bc3e54a0632b>
>
>
>
> _______________________________________________
> dev-tree-management mailing list
> dev-tree-...@lists.mozilla.org
> https://lists.mozilla.org/listinfo/dev-tree-management
Marco Bonardo
> This may be real, but I am going to wait to see more data points to be
> sure.
>
> Cheers,
>
> Shawn
there is also a 3% on Linux, but actually these Ts Shutdown tests are
fluctuating in insane ways and it's hard to link results to changes.
Recall the fact some time ago we got a regression just moving some code
without actually changing what it was doing. And we got
only-partially-explainable 70% improvements.
Limits on Ts Shutdown are quite strict so this % is most likely less
important than stability fixes that this release is bringing us (1
crash, 1 overflow with crash, assertions, timers on win...)
Marco
Shawn Wilsher
> there is also a 3% on Linux, but actually these Ts Shutdown tests are
> fluctuating in insane ways and it's hard to link results to changes.
> Recall the fact some time ago we got a regression just moving some code
> without actually changing what it was doing. And we got
> only-partially-explainable 70% improvements.
> Limits on Ts Shutdown are quite strict so this % is most likely less
> important than stability fixes that this release is bringing us (1
> crash, 1 overflow with crash, assertions, timers on win...)
to see if they can reproduce, and possibly fix it.
Cheers,
Shawn
Shawn Wilsher
> fluctuating in insane ways and it's hard to link results to changes.
> Recall the fact some time ago we got a regression just moving some code
> without actually changing what it was doing. And we got
> only-partially-explainable 70% improvements.
synthetic (taking the max value over everything we collect on
places-stats, and using it to generate a database).
> Limits on Ts Shutdown are quite strict so this % is most likely less
> important than stability fixes that this release is bringing us (1
> crash, 1 overflow with crash, assertions, timers on win...)
This did get us the proper fix for a big stability issue, and fixed a
few other minor issues.
I have been talking with drh about this, and he suspects this has to do
with triggers given the logs that Marco generated. They'll look into
making those faster and will give us test builds so we can generate
numbers and see if they can improve it.
I feel that the wins this update gives us outweigh the slight
performance loss here and we should keep the upgrade. Can I get a
decision from someone with more authority though?
Cheers,
Shawn
Shawn Wilsher
Cheers,
Shawn