[Paraview] Running paraview with MPI

[Berk Geveci]

Hi,

> Running ParaView-MPI binaries (mpirun -np 2 ParaView-MPI) and using the
> Mandelbrot source (or glyphing filter on a sphere with high theta and 
> phi resolution values), a new vtk renderer window appears and renders 
> part of the display; if this window is somehow hidden, the display in 
> the main window is buggy or incomplete (I can send screenshots if needed).

This is expected. Unless off-screen rendering is enabled, ParaView will
create a new render window on each process. These windows should not be
covered by other windows in any way, otherwise, the parallel image 
compositing will not work. This is because the image buffer is
grabbed during compositing. There are a few ways of avoiding this:

1. send each window to a different display; this is the common way
of running paraview on a cluster

2a. On Unix/Linux, compile paraview with mesa or a combination of 
another GL and mangled Mesa and use off-screen rendering
2b. On Windows, use off-screen rendering 
To enable off-screen rendering, you have to define the PV_OFFSCREEN
env. variable. Note that this will only work with the Windows binaries,
on Unix/Linux, you have to recompile VTK and ParaView. I can point
you to some documentation if this is what you want.

> I then compiled a CVS version of ParaView in order to see if this
> problem was specific to PV-06 binaries, but this time, the additionnal
> windows appear directly at launch time (Might I have gone wrong in
> specifying MPI compilation options ?). Rendering and display present the
> same problems, and I get this error message :
> Process id: 1 >> ERROR: In
> /home/cea2002/een/Sources/2003-02-21/VTK/Rendering/vtkInteractorStyle.cxx,
> line 387
> vtkInteractorStyleJoystickCamera (0x9517d70): Timer start failed

You can ignore this message. It is harmless. We will fix it in the
near future (it might have already been fixed). Also, the CVS version of
ParaView is not very stable at the moment since there are a lot of
changes going on.

-Berk