[Paraview] client/server display problem

[Jack Gundrum]

I think I (you guys) figured it out.  I had used our display wall 
cluster with an Xserver on each node with no problem before.

The new cluster I am using is headless (no Xserver) and I need to 
recompile with use OSMesa option turned on and then use the
offscreen rendering flag and this will take advantage (be faster) than X 
forwarding.  If this is correct someone say YES!?


Kevin H. Hobbs wrote:

>On Tue, 2006-05-16 at 11:32 -0400, Jack Gundrum wrote:
>  
>
>>Hi Christoph,
>>
>>So when paraview running in parallel tries to open a process on other 
>>nodes it uses X forwarding? The X forwarding from my
>>cluster server to laptop client works fine if I dont run in parallel 
>>(i.e. mpirun) so X forwarding is working untill I use mpirun.
>>
>>    
>>
>
>What is the value of DISPLAY on the cluster frontend? If it's
>localhost:whatever, set by ssh, this isn't likely to make sense on the
>clients. If the display variable is set by ssh to the real hostname
>(internal network is fine) of the cluster front end then the compute
>nodes should be able to open windows. This is an sshd option. Can you
>run glxgears ( or xclock or whatever ) on the compute nodes?
>
>  
>
>>Should I be compiling paraview with offscreen rendering to avoid X 
>>forwarding among cluster nodes?
>>
>>    
>>
>
>This made my life MUCH easier. 
>
>I think it's even worse if you don't use offscreen rendering on the
>cluster or provide each node with a display ( X virtual frame buffer
>works. ). I think the compute nodes are trying to use the graphics
>hardware on your laptop... Not sure about that though.
>
>
>  
>