You are viewing a plain text version of this content. The canonical link for it is here.
Posted to user@guacamole.apache.org by Tim Worcester <ti...@gmail.com> on 2021/02/05 13:41:18 UTC

Guacd performance tuning

Greetings!

I have been messing around with guacd 1.3.0 and TurboVNC on Centos7.  My
use case is that inside the desktop I open Google Chrome and then load up a
map server (maps.google.com for example).  I have been unable to get this
to be performant.

On the desktop side I have implemented VirtualGL configured to run with an
Nvidia card with the following arguments:
/usr/bin/X :0 -novtswitch -sharevts -nolisten tcp +extension MIT-SHM -seat
1 vt7 +extension GLX +extension Composite -extension RENDER -dpi 96 &

When I connect directly to the desktop through a vncviewer the performance
is much better and very little rendering lag.
When going through Guacamole to the same desktop I don't see
the improvements that my GPU has provided me -- it seems just as performant
as it was before I added VirtualGL to the configuration.

I have tried two different guacd deployments (both containers on a
kubernetes 1.18 cluster).
Both of the following deployments have the guacd pod allocated with 4 cpu
and 4G ram in resources (requested & limit)
1) K8s worker host is a VM with 10 CPU
2) K8s worker host is a server with 56 CPU (performs much better, but still
a slight delay)

Are there any configuration knobs that I should be trying to change on
guacd to make this better?
I have limited knowledge on the encodings and their performance so I don't
know if tweaking that would make a difference?