[Thinlinc-technical] VirtualGL experience?

[Carsten Rose]

Dear all

does anyone use VirtualGL (http://www.virtualgl.org/) in a productive 
environment together with Thinlinc?

We're wondering which GPU's
- are well tested, stable and could be recommended?
- deliver a good price/performance ratio?

And further:
- are there any impacts to take care of?
- is anyone using more than 1 GPU per bare metal?
- how do you use VirtualGL: local or remote as proxy?
- how many sessions do you run per GPU?


We did a very short test with a Gnome 3 / Ubuntu 16.04 installation and 
have been quite impressed by the result.

Without VirtualGL:

a) A Gnome 3 Desktop was really slow (old SunRay 1 feeling). The 
gnome-shell process (one per session) takes more than 50% CPU time (of 
one core) and this constantly.

b) A Unity Desktop (which by default comes up with unity-lowgfx in a 
Thinlinc environment) are much better due to the unity-lowgfx 
(http://www.whizzy.org/2016/09/unity-7-low-graphics-mode/ - check the 
video). *But* there is one compiz process per session, which runs 
constantly between 50 and 100%!

With VirtualGL:

a) There was *no* difference in the user experience, between a Gnome 3 
desktop session on a powerful notebook (standard graphic features 
enabled) and a Gnome 3 Thinlinc session (lot's of graphic features 
disabled, comparable to Unity-lowgfx). The gnome-shell process raises in 
the CPU time only (up to 100%), if there are movements on the screen - 
the CPU drops down in the same moment when the movement stops. We've 
tested with three Thinlinc session simultaneously.

b) Unity Session: not tested now.


BTW: Cendio provides some generic installation hints about VirtualGL:
https://www.cendio.com/resources/docs/tag/virtualgl.html

Thanks a lot and have a nice weekend.

CU
Carsten
-- 
Carsten Rose
IT Koordinator / Institut für Mathematik

*****
Universität Zürich
Carsten Rose
Institut für Mathematik, Y27-J40
Winterthurerstrasse 190
CH-8057 Zürich

+41 44 635 58 47 Telefon
+41 44 635 57 05 Telefax

www.math.uzh.ch
carsten.rose at math.uzh.ch

"There are 10 kinds of people in the world, those
that understand binary .. and those that don't"