Hi,

 

some time ago I inquired about peoples experience with x2go in a LAN environment. I have now built a TCE live system (TFTP-only, no NFS) and a new Jessie server and I am still not very satisfied. I tried almost all available options. Here is the setup:

 

1GBit/s LAN on both sides

Host: AMD 4.4GHz 8-core CPU with 32GB RAM

Client: Pentium D 2.8GHz, 1GB RAM, nVidia 6200 with 2 screens@1200x1600

 

I tested MATE desktop with Firefox 51 (tried XRender on/off) and Chromium 55. I usually set the browser to fill half of the screen, that mean roughly the same number of pixels as a Full-HD (1920x1080) screen. I usually open https://www.heise.de and scroll up and down a bit.

 

nopack: almost the best of it all, responsiveness good, Chrome and Firefox without XRender laggy

 

rfb-hextile/tight: better for small screen updates, worse for large screen updates, scrolling in Firefox with XRender active is *very* smooth

 

rdp: almost identical to rfb

 

adaptive: middle ground, everything is laggy but at least there's always the same amount of lag for small and big screen updates. Unacceptable though because almost all text has JPEG artifacts no matter whether the quality is set to 1, 7 or 9. Graphics like Munin graphs are really ugly.

 

16m-png: good quality, mostly lag-free except Chrome and Firefox when XRender is off, becomes very unresponsive with lots of screen updates

 

16m-jpeg: artifacts, speed like adaptive

 

 

I believe nopack or rfb would give better performance if it was possible to circumvent going through SSH. I have set SSH to use arcfour (this is a trusted, isolated LAN) and performance has been slightly better.

 

Is there any way of making the initial authentication and startup through SSH but then having the NX connection going directly to the client? That would save a lot of processing power! (I'm thinking similar to the "tunnel" feature for sound.)

 

Regards,

 

Stefan