[Paraview] Dual-headed output from a single GPU?

[Paul Melis]

Hi,

I'm trying to get paraview to handle rendering on a single GPU with
dual-monitor outputs. These outputs are run using different X screens on
the same X server. Using xinerama is not an option as this is a
preliminary test setup for driving a TPD using nodes that each drive 2
displays and the mullions can't be handled correctly with xinerama. The
goal is to run Paraview as described in
https://visualization.hpc.mil/wiki/Paraview_Tiled-Display_Mode.

The problem I have is to get multiple pvservers to run on the same host.
I'm basically following the procedure outlined in "Multiple GPUs Per
Node" on http://www.itk.org/Wiki/Setting_up_a_ParaView_Server. The
-bynode mpirun line (supported by OpenMPI) works, but it leads to the
different pvserver processes trying to listen on port 11111:

paulm at sara0143:/scratch/paulm/paraview-3.6.1-Linux-i686/bin$ mpirun
-bynode -np 1 ./pvserver -display :0.0 : -np 1 ./pvserver -display :0.1
Listen on port: 11111
Waiting for client...
Listen on port: 11111
ERROR: In
/home/kitware/ParaView-3.6/ParaView3/Servers/Common/vtkProcessModuleConnectionManager.cxx,
line 193
vtkProcessModuleConnectionManager (0x8216f60): Failed to set up server
socket.

As far as I understand it this port (11111) is the "combined server"
port that the client connects to, but the connections among different
pvservers will use different (arbitrary) ports. So it seems two
pvservers want to play the role of combined server entry point in my
case. The mpirun line above is very similar to the one shown in the
Setting_up_a_ParaView_Server wiki page:

mpirun -bynode -np 8 ./pvserver -display :0.0 : -np 8 ./pvserver
-display :0.1

Is there some information missing on that page that is needed to get
this specific setup to work? I.e. something in an MPI hosts file?

Thanks in advance,
Paul