[Paraview] Failure -> 16 windows -> Mangled Mesa

[Kevin H. Hobbs]

On Tue, 2006-03-21 at 15:27 -0500, Berk Geveci wrote:
> The problem is that whatever mpi implementation is using is setting
> the display to be your client display (I am guessing through ssh). You
> need to force it to use the local display. I usually set DISPLAY to :0
> in .XXXrc script.
> 
> On 3/15/06, Kevin H. Hobbs <kevin.hobbs.1 at ohiou.edu> wrote:
> > I compiled Paraview with the system GL and off screen mesa with very
> > poor results. I recompiled Paraview without any off screen rendering and
> > got very good test results but if I run a server on many nodes I get
> > many windows on the client display. It was really slow. Are these
> > windows competing for the graphics hardware on my workstation?
> >
> > Now I'm working on using mangled mesa on the cluster to try and get over
> > this.
> >
> >

The list of possible ways to work the rendering is starting to boggle my
mind.  It may be time to go home.  

To get a valid DISPLAY for paraview from the compute nodes we had to
tell ssh to not set localhost:blah but the real hostname of the login
node ( the node to which I ssh from where I'm sitting ). Then we had to
tell the batch system to copy all shell variables to the execution host.
mpiexec copies it to all the nodes. So yeah they are all getting that
valid DISPLAY from my workstation.

I suppose now that I know I can get client/server working I'll try
DISPLAY=:0. There's no X server running on them though so I didn't think
that'd work. They all do have a graphics card for each pair of
processors.

In the mean time I recompiled paraview with pure Mesa which seems to do
a lot better. CMake still doesn't like to be told the GL can do
offscreen mesa calls.
-------------- next part --------------
A non-text attachment was scrubbed...
Name: not available
Type: application/pgp-signature
Size: 191 bytes
Desc: This is a digitally signed message part
Url : http://public.kitware.com/pipermail/paraview/attachments/20060321/e6088805/attachment.pgp