KDE performance (again) Luboš Luňák 4.7.2010 | Tampere, Finland | Akademy 2010
Do we know our resource demands? ● Akonadi? ● Nepomuk? ● Plasma clock? ● KWin compositing? ● 4.4 to 4.5 change? ● … 4.7.2010 www.kde.org 2
Do we know our resource demands? ● Akonadi? ● Nepomuk? ● Plasma clock? ● KWin compositing? ● 4.4 to 4.5 change? ● … ● Do we care? 4.7.2010 www.kde.org 3
Rule of the big three ● “Choose two, you can't have the third” ● For performance: good, efficient, in time ● => performance is always last, unless it's bad enough to affect “good” 4.7.2010 www.kde.org 4
Benchmarking basics ● Use a proper tool ● Really measure what is intended to be measured ● Minimize noise ● Comparable conditions 4.7.2010 www.kde.org 5
System memory ● Forget classic tools like 'top' or 'free' ● KSysGuard ● Exmap ● xrestop ● Processes can be quite expensive 4.7.2010 www.kde.org 6
Application memory ● KSysGuard ● Massif ● kdesdk/kmtrace 4.7.2010 www.kde.org 7
CPU usage ● QTime ● Callgrind/KCachegrind ● Sysprof 4.7.2010 www.kde.org 8
I/O ● strace (-e file) ● SSD 4.7.2010 www.kde.org 9
Performance testsuite ● It would be useful to have a set of “standard” interesting measurements ● Everybody would be able to repeat ● Releases could be compared ● Regressions could be found ● Login time, memory after login, KFoo memory usage after doing XYZ,... 4.7.2010 www.kde.org 10
Recommend
More recommend