You'll also want to make sure that you either give the full path to pvserver when you start the mpirun job, or make sure that pvserver is in your default path.

Will distributed processing help you? That totally depends on the distribution of your dataset, speed of your network & processors, etc. If you try to distribute a "small" dataset across multiple nodes, you may actually make things slower. Same is true with distributed HW-based rendering. The bottom line disclaimer: Your actual mileage may vary.

David Doria wrote:
On Wed, Jul 15, 2009 at 6:34 PM, pat marion<pat.mar...@kitware.com> wrote:
Hi David,

You'll want to change your mpirun command a little bit-

on machine X: mpirun -np 3 -machinefile machinefile.txt pvserver

This will run 3 instances of pvserver where each instance runs on one of the
machines specified in machinefile.txt.  Machinefile.txt will list the
machine hostnames, one per line.  You'll need to setup password-less ssh
from machine X to machines Y and Z, otherwise it will prompt you for a
password for each machine.

Here's a link to a little test program you can try.  It will have each mpi
process print out its rank and hostname so that you can confirm that each
process really is running on its own machine-
http://www.slac.stanford.edu/comp/unix/farm/mpi.html

Pat


Thanks Pat, I'll try that in the morning. Do I have to do anything on
Y and Z to prepare them to be in the machinefile list (like somehow
start mpi listening on them)? Also, if starting the pvserver from X,
does X have to be in the machinefile list? Or is it included
implicitly?

I've read several opinions on openMPI vs mpich2, but has anyone tested
these using paraview type benchmarks?

Can anyone comment on whether or not setting up something like this
will actually help? I am current using a pretty new P4 with 4gb ram
and a decent video card - would switching to a group of 3 or 4 P3's
with 2gb ram actually make a difference? Or would it actually be
worse? Or can I also run an MPI pvserver on my machine so I can use
A+X+Y+Z? If that is the case, is A smart enough not to send messages
over the network and then back to itself (i.e. will it really be A + X
+ Y + Z (i.e. strictly better than just A), or will it be (A - a
decent amount) + X + Y + Z?)

I also have access to an 8 core SGI machine that has no video
hardware. How much does the video card actually do in the rendering?
Is 8 cores worse than a machine with one core and a good video card?

Thanks,

David
_______________________________________________
Powered by www.kitware.com

Visit other Kitware open-source projects at 
http://www.kitware.com/opensource/opensource.html

Please keep messages on-topic and check the ParaView Wiki at: 
http://paraview.org/Wiki/ParaView

Follow this link to subscribe/unsubscribe:
http://www.paraview.org/mailman/listinfo/paraview
_______________________________________________
Powered by www.kitware.com

Visit other Kitware open-source projects at 
http://www.kitware.com/opensource/opensource.html

Please keep messages on-topic and check the ParaView Wiki at: 
http://paraview.org/Wiki/ParaView

Follow this link to subscribe/unsubscribe:
http://www.paraview.org/mailman/listinfo/paraview

Reply via email to