Re: Organising an ARM performance drive?

[Mike Turquette <mturquette gmail com>]

On Wed, Nov 4, 2009 at 10:37 AM, Dave Neary <dneary gnome org> wrote:
> Hi all,
>
> As you all know, this list has been quiet of late, and
> gnome-embedded-list is retired.
>
> Recently during a couple of events I attended (OSiM World, the Maemo
> Summit and ELC), a number of people have signalled to me that they are
> interested in working together on a performance review of the GNOME
> stack on ARM. Others have indicated that they're concerned about
> performance of newer GNOME developments on ARM. This smells to me like
> an ideal opportunity to collaborate.
>
> Is this the kind of project which I could get buy-in and support from
> people here, and ideally empower half a dozen people here to be the
> drivers for the project (based on hardware availability & skills &
> resources - none of which I have in abundance)?

I'm always interested to help from the OMAP perspective.  I think some
hard performance data for real world software like GNOME has been
missing for a while on ARM.  I have hardware to test on and I can do
some testing myself.  Brian Cameron has two Zoom2 boards that can also
be used; I've CC'd him.

> My thoughts are that we could attach this in one of two ways - from the
> top down or from the bottom up. Either we look at applications using the
> GNOME stack, and try to profile them on ARM and x86 to identify
> bottlenecks, or we go the other way, and write a bunch of test programs
> to stress individual points in the GNOME stack and identify places we
> can make improvements that way.
>
> It seems best to my mind to start from the real-world performance
> problems and work down. Otherwise you end up gamind the profiling suite
> rather than optimising for real world use.

Do any GNOME benchmarking tools exist?  If so start with those.  I
agree that the ARM benchmarks need to be compared to x86 for any sort
of real relevance.  Is there a GNOME performance/benchmark/tools team
or project that we can get help from?  Having to write the tests and
do the testing is a lot of work to have to commit to.

> Some obvious opportunities for performance checking at the GNOME level
> are Gstreamer, Clutter, PulseAudio, GTK+, pango and perhaps even Xorg.
> Suggestions for best ways to find specific performance issues which we
> can then go about getting fixed are welcome.

An interesting place to start is the Always Innovating Touchbook
(http://www.alwaysinnovating.com/home/index.htm).  I hate to be a
broken record on this product but it is the only real ARM netbook that
I'm aware of.  It ships with Angstrom running a mostly-GNOME
environment with some Xfce thrown in.  The user experience on this
device is pretty bad.  Interactions feel slow and clutter in
particular is quite buggy using the OpenGL ES backend.  The
accelerometers attempt to rotate the desktop (ie, the apps and where
the panel is docked) if the user spins the screen around, and this
reveals some bugs in the software.  I don't think there are that many
x86 machines exercising these code paths.  The Touchbook might be a
good platform to try and test on since its the closest thing to a real
product we have.

All of that said, the first thing we need is a test suite/performance
benchmarking suite.  If that doesn't exist in a useful manner to us
already then we're at a dead stop until that gets complete IMO.

Regards,
Mike

> Cheers,
> Dave.
>
> --
> Dave Neary
> GNOME Foundation member
> dneary gnome org
> _______________________________________________
> mobile-devel-list mailing list
> mobile-devel-list gnome org
> http://mail.gnome.org/mailman/listinfo/mobile-devel-list
>