Xpra: Ticket #2219: compare performance of gtk2 and gtk3 builds
In light of ticket:2211#comment:4 : Is GTK3 that much slower?
We need some hard numbers.
Let's run the full automated tests and compare.
Thu, 01 Aug 2019 14:53:35 GMT - Antoine Martin: owner, status changed
- owner
changed from Jonathan Anthony to Smo
- status
changed from assigned to new
Mon, 19 Aug 2019 21:04:25 GMT - Smo: attachment set
- attachment
set to test_gtk.tar.gz
comparison of gtk2 and gtk3
Mon, 19 Aug 2019 21:06:34 GMT - Smo: owner changed
- owner
changed from Smo to Antoine Martin
I've attached some logs, data, and charts. I did this test by changing the top of
/usr/bin/xpra
to be python3 instead of python2 to force make it try python3 first
I still have an issue with gtkperf the data looks all wrong and when I watch it running it appears to be going through the tests super fast so i'm not sure what is going on there.
Will just omit that one for now and try and see what the issue with it is.
Still some nice data here.
Tue, 20 Aug 2019 05:38:28 GMT - Antoine Martin: status changed; resolution set
- status
changed from new to closed
- resolution
set to fixed
... /usr/bin/xpra
to be python3 instead of python2 to force make it try python3 first
FYI: that should already be the case when installing from RPM.
Very nice!
I have uploaded the charts here: https://xpra.org/stats/gtk2-vs-gtk3/charts.html.
Overall, this is an unexpected decisive win for GTK3 in most key metrics:
- lower batch delay - except for x11perf
- lower damage latency - except for glxgears
- higher quality - except for glxgears
- much lower max batch delay and max damage latency, even memscroller is almost where it should be
- higher quality and more pixels-per-second sent
- more regions-per-second
- less server CPU! despite compressing and sending more data
The only metrics which are showing regressions are:
- 15% more memory usage on the server, 25% more on the client
- the user CPU is a bit higher, but that is mostly in line with the increased number of pixels that are processed (thanks to the better framerate)
- the encoding and decoding "pixels-per-second" - not too critical, but I will try to look into this as there is quite a big gap, maybe because:
It looks to me like gtk3 is using video encodings more aggressively: more threads, more cpu, etc. Not sure why that is.
Sat, 23 Jan 2021 05:45:22 GMT - migration script:
this ticket has been moved to: https://github.com/Xpra-org/xpra/issues/2219