Features/changes with performance regressions

[Martijn]

Question: are performance regressions acceptable when with UI features/changes?
I was hoping it was not, but with
https://bugzilla.mozilla.org/show_bug.cgi?id=437829 it seems like it
is.

Also, the new tab switcher performance is quite bad here on my 1.6GHz
PC (although usable). But I'm afraid it will be quite a drag on older,
slower computers.

Regards,
Martijn

--
Martijn Wargers - Help Mozilla!
http://weblogs.mozillazine.org/qa/
http://wiki.mozilla.org/Mozilla_QA_Community
irc://irc.mozilla.org/qa - /nick mw22

[Peter Weilbacher]

On 11.11.2008 14:02, Martijn wrote:
> Question: are performance regressions acceptable when with UI features/changes?
> I was hoping it was not, but with
> https://bugzilla.mozilla.org/show_bug.cgi?id=437829 it seems like it
> is.

Thanks for pointing out that bug! It was reported to me several times that on
OS/2 page-loading causes high CPU and always thought that was our problem. But
it really seems to be a general one (which is a relief as we cannot really run
profiling for the Mozilla code on OS/2). The workaround that I found was to
just remove the throbber from the main bar...

Peter.

[Boris Zbarsky]

Peter Weilbacher wrote:
> Thanks for pointing out that bug! It was reported to me several times that on
> OS/2 page-loading causes high CPU and always thought that was our problem.

Uh... UI changes that cause a Tp hit are generally considered
unacceptable, I'd think, without very special dispensation.

-Boris

[Mike Beltzner]

This isn't actually a UI change causing a Tp hit, it's a platform
change (rendering APNG seems to be very costly, CPU-wise) which was
tripped by the front-end change, but which would also occur any time a
user loads a page with an APNG on it.

See bug 463984 (https://bugzilla.mozilla.org/show_bug.cgi?id=463984)
which has been nominated for blocking.

On the actual issue of a Tp change, one was never noticed when we
checked the new throbber in to the new theme (and believe me, people
were watching perf numbers when we made that change) so are you sure
that this actually caused a significant Tp regression? Got a graph? :)

cheers,
mike

[Boris Zbarsky]

Mike Beltzner wrote:
> This isn't actually a UI change causing a Tp hit, it's a platform change
> (rendering APNG seems to be very costly, CPU-wise) which was tripped by
> the front-end change, but which would also occur any time a user loads a
> page with an APNG on it.

Fair point, though the throbber is there during all pageloads. But see
below.

> On the actual issue of a Tp change, one was never noticed when we
> checked the new throbber in to the new theme (and believe me, people
> were watching perf numbers when we made that change) so are you sure
> that this actually caused a significant Tp regression? Got a graph? :)

I didn't say there was a Tp change; Peter said that there were reports
of significantly increased CPU usage on OS/2 during pageload. That may
or may not translate into Tp issues for users (esp. if the browser is
not the only app running). More data is needed, if someone cares about
that issue. I realize that we didn't have a Tp change on our Tier-1
unloaded boxes of course. ;)

-Boris

[Peter Weilbacher]

I don't think that it caused any Tp hit on recent CPUs where you rarely get
100% usage because of this. Users who complained mostly had CPUs with < 1 GHz.
I guess the performance monitoring machines are quite new and fast. And I don't
think there is a graph for monitoring (mean/median/min/max) CPU usage during
the Tp tests. Perhaps that's something worth implementing?

Peter.