Remote Visualisation

Note: Tested on a standard compute node in HPC 3.0. Node already has CUDA 4.0 installed with a number GL based RPMs. Not sure if these are required:

yum install libXi-devel libXmu-devel gcc kernel-devel gcc-c++ freeglut.i386 freeglut.x86_64 freeglut-devel glibc-devel kernel-source

Installation Prerequisites

• Required libXaw and libXv

# For turbovnc and virtualgl rpms
yum install libXv libXaw

# for a desktop environment
yum groupinstall "GNOME Desktop Environment" "X Windows System"
yum install xorg-x11-server-Xorg xterm firefox

• Change runlevel to 5 and reboot. GDM needs to be running for vnc sessions to start up correctly.
• Setup xorg.conf using nvidia-xconfig

Install turbovnc and virtualgl

• Copies on PDD Applications (<file>\\srv-vfs2\PDD_DATA\Product Development\High Performance Computing\HPC Benchmarking\Applications|PDD Link</file>) or http://www.virtualgl.org/Downloads

rpm -ivh turbovnc-1.0.1.x86_64.rpm VirtualGL-2.2.1.x86_64.rpm

vglserver_config +glx
usermod -a -G vglusers david
/opt/VirtualGL/bin/glxinfo -display :0 -c

Setting up remote access

Use ssh for port forwarding:

• In this example
  • vhpchead is the headnode used to access the cluster
  • viscomp000 is the node running vncserver which is used for remote visualisation
  • 10001 is the port listening on your local desktop
  • 5901 is the port listening for vnc connections on viscomp000

# On one terminal
ssh -L 10001:viscomp000:5901 vhpchead

# Then on another local terminal
vncviewer localhost:10001

HP RGS vs VirtualGL/TurboVNC

RGS limitations:

• HP hardware only (a problem for customers who want to keep things ‘open’)
• Only supported on Windows & RHEL (not SLES)
• Single session per server
• No dynamic resolution

VirtualGL limitations:

• Does not fully support the OGL standard
• No dynamic resolution
• The way it is run (requires “virtualgl” in front of the application command