Hi,

when I start OpenDX on a shared-memory machine (eg. SGI Origin, IBM SP3,
Linux-SMP)
I see in the startup messages that multiple worker processes are started by
default
(typically two - this can be changed using the '-processors' option).
I'd like to make use of those when running a given DX network. My question is:
how
do I do that ?

I realize that many (which ones ?) of the DX modules are parallelized already,
and that I can use the Partition module to split a single dataset into chunks
for
parallel processing. I tried this with a simple example of calculating
isosufaces
from a big dataset, and it worked fine when running it from within the VPE.
However, when I then tried to run the saved *.net file directly from the command
line in script mode it failed with this error message:

   1:  worker here [7364] 
   0:  worker here [7365] 
  child process 1 (7364) exited, status = 0
  -1:  cleaning up and exiting 

  parent exiting

Is there something special I have to do when running a DX network in script mode
on a parallel machine ?


Apart from the intra-module parallelism I'd like to know whether OpenDX is also
able to run independent data paths in a given network in parallel. My example:
I have written an import module to read a remote HDF5 file. In my visualization
network I need to import data from three different HDF5 files. So I'd like to
run my three import modules independently and in parallel in order to hide the
latencies for the remote file access.
I tried to use exexution groups but didn't get very far with this. Specifically
I didn't know what hosts (other than 'localhost') I should assign my execution 
groups. Are execution groups meant to exploit an SMP machine ?


Thanks in advance for any hints and comments !


Ciao,
Thomas.

-- 
=========================================================================
Thomas Radke
Max-Planck-Institute for Gravitational Physics, Albert-Einstein-Institute
Am Muehlenberg 1, 14476 Golm, Germany

fon +49 331 567-7329  fax +49 331 567-7298  http://www.aei.mpg.de/~tradke
=========================================================================

Reply via email to