dcommander
commented
7 years ago Thanks for sharing your configuration, but it is extremely specific. If you're asking whether I could provide such an example in the User's Guide, then the answer is no, because there is a high likelihood that such an example wouldn't work on any system other than yours. I would be willing to provide a wiki page on VirtualGL.org describing how to configure your specific GPU for headless VGL operation, but such a wiki page would need to act as a supplement to the User's Guide. See http://www.virtualgl.org/Documentation/HeadlessNV for an example.
If you want the User's Guides to be improved, then please approach the problem from the following angle:
- What specific problems did you encounter while following the User's Guide instructions?
- How specifically could those instructions be improved? That is, what additional information would make TurboVNC and VirtualGL easier to configure?
For instance, I could certainly write some general guidelines for headless configuration, but in terms of xorg.conf, it would be much better to discuss how a headless xorg.conf file differs from a "normal" xorg.conf file rather than providing a complete example, since not all parts of the example would apply to all systems.
TurboVNC should be simple to install. You should be able to just install the packages and run /opt/TurboVNC/bin/vncserver to start a session. If that procedure fails on a specific Ubuntu release, then that is a bug that needs to be fixed. Please provide details of the failure. It should work properly on Ubuntu 16.04 without MATE, but you need to have the GNOME Flashback package installed if you aren't using MATE. If GNOME Flashback isn't installed by default, then perhaps I need to make that more clear, or provide a link in the documentation to the window manager compatibility guide.
VirtualGL is a bit more complex to install. Generally the installation assumes that you already have a working X server running with GPU acceleration on the VGL server machine, and that seems to be where you ran into problems. I've added a brief section to the User's Guide that clarifies those prerequisite steps, but unfortunately the specifics of those steps are going to vary wildly from system to system, so it would be impossible for me to document all of them. The vast majority of VirtualGL installations use either nVidia or AMD GPUs. Intel GPUs have historically not provided anywhere near the same level of performance, stability, or ease of configuration, which is why I cannot recommend their use with VirtualGL. As I said above, however, I am willing to provide wiki instructions regarding how to configure specific GPUs for headless operation, if those instructions can be expressed in the most general way possible.
ddavidebor
I've spent a riddiculous amount of time getting virtualgl and turbovnc to work on ubuntu with mate with an integrated intel graphics on a completely headless system.
Mostly because i'm really pissed of by the fact that apparently a remote workstation is a thing only on windows.
My user case: I Work in 6 different location, all of them with a good internet connection and want a remote workstation with a complete OS to do CAD (in vmware) with 3d acceleration, software development etc etc.
I think the documentation really lacks an immediately implementable tutorial.
The following procedure works for ubuntu 16.04 server on a Xeon E3 1245 v5 https://www.supermicro.com/products/system/3U/5039/SYS-5039MS-H12TRF.cfm?parts=SHOW
The system is for trusted users only and i've yet to test it with multiple user at the same time.
I'd like to know what you think of the following procedure. I was not able to get gnome to work. I had to cherrypick some parts from the TightVNC docs.
The xorg config is necessary otherwise VirtuaGL will try to use one of those crappy server graphics, which of course does not support opengl, and will fallback to software rendering.
Example instructions
example turbovnc security conf
Example sysconfig
Example xorg.conf
testing