Re: [g-a-devel] At-spi2-atk performance tests

[Alejo Pacín Jul <apacin igalia com>]

Hi people!

After a lot of serious problems to get newly compiled Accerciser, 
Gobject-introspection and others, I've finished the tests with Accerciser.

Results at 
http://www.fileserve.com/file/m9Xv8Dc/at-spi2-atk-performance-tests_v2_5.tar.gz. 
Also say that this is the final version (full and clean).

- On the sysprof case:

At this time, I've had to change a bit the approach. As in the previous 
results, I've had problems with the interference of other apps in the 
cpu time results. But, at this time, this other apps were different 
between cases (access. enable and access. enable + accerciser). So, it 
was impossible to make comparisons between them.

To solve this problem, I've left to stay the tested app cpu time and 
reinterpreted the results. So, now a cpu time descent is a performance 
degradation (if the cpu time is spent onanother application is not 
spenton the tested application; so it degrades the performance of the 
second). So, now it's better a bigger value on "after" and "master" 
branches respect to the "before" branch.

    - Nautilus: we have an improvement of a 20% on the "acc. enable" 
case. Also, we can see a degradation of a 40% between the "acc. disable" 
and the "acc. enable + accerciser" cases.
    - Rhythmbox: no changes. But we have a degradation of a 60% on the 
same cases.

- On the valgrind case:

    - Nautilus: improvement of a 35%. Degradation of a 300%.
    - Rhythmbox: improvement between 5%-10%. Degradation of between 5%-15%.

So, I can say that the results are as expected. May be the improvement 
on Rhythmbox is small, but it is just an improvement. On the contrary, 
the improvement on Nautilus is notorious.

Finally, say I'm finishing my internship in Igalia and it is possible 
that this is one of my last post on the list. I have to say it's been a 
pleasure to work with you.

Best regards,

                        Alejo.