Re: [VirtualGL-Users] Click-and-drag caused crashing of VNC

[DRC <dco...@us...>]

On 9/25/12 9:16 AM, James Wettenhall wrote:
> BTW, the reason why we sometimes use TigerVNC server instead of TurboVNC server is that some of our (virtual) servers don't have GPUs, but we still want to be able to test some OpenGL software on them.  For these servers without GPUs, we have found it easier to get the OpenGL software up and running with TigerVNC server than with TurboVNC server / VirtualGL.  But for servers which do have GPUs, we want to be able to make use of hardware acceleration, so we use TurboVNC viewer (on the client side) and TurboVNC server.  Does that sound reasonable?  Or do you think that with a bit of practice, it should be easy to get OpenGL stuff (e.g. glxgears and glxspheres) up and running quickly (but without hardware acceleration) using TurboVNC server, instead of TigerVNC server on our GPU-less virtual servers? That way we could use the same VNC server software on all of our physical and virtual servers.

Yes.  If you are using a recent Linux distro, download the latest Mesa 
source (ftp://ftp.freedesktop.org/pub/mesa/) and build it using:

   configure --enable-xlib-glx --disable-dri --without-gallium-drivers
   make

Then you can simply add {full_path_of}/Mesa-8.0.4/lib to your 
LD_LIBRARY_PATH environment variable to cause OpenGL applications to 
prefer your custom build of Mesa.  What you've done above is build a 
version of Mesa that doesn't require an X server.  It renders all of the 
OpenGL in-process and just spits out plain Xlib calls.

I found that on an older distro (CentOS 5.x), I had to use Mesa 7.5.2 
and build it with:

   configure --with-driver=xlib
   make

but it worked the same way as 8.0.4.

It wouldn't be all that difficult to put a software GLX extension in 
TurboVNC, but one reason I don't is that I specifically want 3D not to 
work, so it's obvious when VirtualGL is working.  Otherwise, if a user 
forgets to type vglrun, then their application is suddenly slow, and no 
one can figure out why.  The other reason is just that OpenGL changes 
more quickly than the other APIs we depend on, and it would be a pain to 
keep abreast of those changes.