Headless installation of virtualgl.

After installation of virtualgl on CentOS 6.2 one need to be sure that main X.org is starting without monitor.
Without that the vglrun throws an error unable to connect to display :0
The reason is xorg is not started due to the monitor. The monitor is detached.
You cannot keep one monitor per cluster node if you want to run rendering farm of 1000 GPUs.
So the solution is to run nvidia driver in headless mode!!!
After Nvidia official driver installation run following:

 
nvidia-xconfig -a --use-display-device=None --virtual=1280x1024

reboot it.
Then the in the vnc connection simply run:

vglrun glxinfo| grep -i renderer 

You should see your GPU string information: in my case is Tesla 2050