Remote Visualization

Brief introduction (see www.virtualgl.org for a comprehensive discussion)

Remote Visualization allows to perform visualization on a remote compute system (with dedicated capabilities, such as 3D hardware acceleration), and have the display efficiently sent to the local client workstation/PC.

The traditional remote X-Windows approach (ssh -X ...) is not generally suitable for remote visualization of large 3D datasets, since all of the OpenGL commands and 3D data are sent over the network to be rendered on the client machine. Given the large amounts of data to be transferred such an approach can easily be verified to be prohibitive for achieving interactive framerates, even on fast networks (wrt. bandwidth, latency). In remote visualization solutions, such as VirtualGL, the OpenGL commands and 3D data are instead redirected to a 3D graphics accelerator on the server machine, and only the rendered (and compressed) 3D images are sent to the client machine where they are displayed.
Besides tool-specific approaches (see, e.g., the client-server architecture implemented by VisIt and ParaView) a number of generic software solutions for remote visualization exist, which allow to employ any (OpenGL-based) graphics software for remote visualization.
 
 

Remote visualization  at RZG

Currently, RZG provides provisional remote visualization facilities on host kandinsky.ipp.mpg.de based on two open-source projects VirtualGL and TurboVNC. Two different modes are available (see www.virtualgl.org for a comprehensive description)

a) VGL Image Transport ("direct mode"):

• recommended mode in LANs: high bandwidth/low latency networks without significant firewall barriers (besides ssh, communication on ports 4242,4243 must be permitted)
• requires an X-server running on the client PC
• requires installation of VirtualGL on the client PC
• howto:
  • without ssh-tunneling: recommended for RZG, IPP clients
  1. on the client PC start: vglclient
  2. login to the visualization server: ssh -X -l <RZG account> kandinsky.ipp.mpg.de
  3. on the visualization server start application with vglrun <path to OpenGL application>, e.g. vglrun glxgears
  • with ssh-tunneling of ports 4242,4243, ...: recommended for MPA clients
  1. on the client PC start: vglconnect -s <RZG username>@kandinsky.ipp.mpg.de
  2. in the kandinsky terminal window start application with vglrun <path to OpenGL application>, e.g. vglrun glxgears
  3. when done, cleanup connection on the client PC with vglclient -kill

•  prerequisites:
  1. apply for an account on kandinsky.ipp.mpg.de by sending an informal E-mail message to visualization@rzg.mpg.de
  2. install VirtualGL on your desktop machine, downloadable from www.virtualgl.org

b) X11 Image Transport ("raw mode"):

• recommended mode in low bandwidth/high latency networks
• recommended mode for Windows clients
• does not require an X-server running on the client PC
• requires installation of TurboVNC on the client PC

• howto:

1. login to the visualization server with: ssh-l <RZG account>-L 5904:localhost:5904 kandinsky.ipp.mpg.de
2. on the visualization server start vncserver, e.g. with vncserver -geometry 1280x1024 :4
3. on the client PC connect with vncviewer localhost:4
4. within the virtual desktop popping up start application with vglrun <path to OpenGL application>, e.g. vglrun glxgears
5. when done, do not forget to shut down vncserver withvncserver -kill :4

   note: the last digit of the the selected port number 590X corresponds with the display number :X, as indicated by the red font colors in the examples above

•  prerequisites:

1. apply for an account on kandinsky.ipp.mpg.de by sending an informal E-mail message to visualization@rzg.mpg.de
2. install TurboVNC on your desktop machine, downloadable from www.virtualgl.org

local support:

visualization@rzg.mpg.de