On Thu, Jul 15, 2021 at 1:33 PM <res1@embeddedjava.de> wrote:

I noticed that the desktop performance is relatively slow, especially also for direct connection to the local PC (X2Go-Client on Local machine connects to X2Go-Server on Local machine).

Connecting to a running desktop means catching any changes and sending them over the wire. This is - by principle - less efficient than running an own X server (x2goagent) and will always be slower. A real x2goagent session will not track changes but will intercept the drawing calls of the applications and efficiently forward them to the x2goclient side.

For that reason this mode of operation is kind of an "unloved child".

My expectation is that such connection (possible only over the loopback device) should run very fast, but isn't. So why is this the case? The same issue I observed in the local network (at 1GBit speed).

That means that connections over the global net can't be fast, if it is not possible with direct connections or local net, if there are some speed issues than the remote desktop is not usable anymore.

I don't think that my PC is the bottleneck (Ryzen 3700 + 32GB RAM + PCIe SSD).

No, as I wrote: it is in the nature of the local desktop mode (which is called "shadowing" internally). There's almost nothing you can do about it.

One approach for improvement is running the local desktop as x2go session, too. So you always run an x2go session, regardless of your location (in front of the machine or remote). Then you will always have a fast session. There are drawbacks with this approach, too: You'll lose hardware acceleration and you'll have to figure out how to set up such a session in your environment. I am constantly working like that but I am using an own (proprietary, not publicly available) script for that.

Uli