[Paraview] Running Paraview on an SMP machine

[Berk Geveci]

> Hi - I am new to the use of parallel visualization, and wanted to check it
> out a bit. I have several questions about using Paraview on an SMP (dual
> proc or otherwise) machine:
> 
> - When starting paraview in parallel - is the typical usage to start in
> client-server (so pvclient and pvrenderserver) mode with a multiprocess
> server or to start in distributed stand-alone mode (paraview)? Are there
> particular reasons to choose one over the other?

If I am running on an smp machine, I usually use:
mpirun -np 2 paraview
If you are running on one machine (no client/server separation), I
don't see why you would use client/server or client/render server/data
server.

> - The older online docs mention that if several processes share a display,
> the render windows may overlap and the resulting image may be corrupt. what
> is the typical way around this?

If your geometry is not too large, you will notice that this is not an
issue. This is because paraview renders small geometry on the first
node (or client). How small small is is controlled by a threshold
parameter on the 3D view setting page. If you are running on the same
machine, I don't see why you should use compositing. You will be using
the same graphics card and you might as well render on the first node.
Just disable "Composite" from 3D view settings.
 
> -  In goofing around, I tried adding the --use-offscreen-rendering flag to
> pvrenderserver, but still render windows appeared for the processes. Is this
> a sign that I don't have mesa installed or set up properly?

Yup.

> - In order to take advantage of the parallel rendering - would I need to
> feed decomposed data files into the program, or does it handle this
> internally? Also - Other than checking speedup in rendering, how can one
> check that the rendering is actually being done in parallel vs. duplicated
> on each process?

See comments above.