[vtkusers] paralleliso isnt.

[Dave Semeraro]

I have a question about the pipeline execution of the paralleliso example. 
That example uses
several processors to calculate an isosurface. It uses input and output 
ports. Process 0
builds several input ports and pipes their output to an appendpolydata 
filter. I have no doubt
that when the pipeline executes geometry is generated on each process. I am 
not convinced
that it is done concurrently. I think that when the appendpolydata filter 
updates it updates
each of its inputs sequentially. The inputs to the filter are the output if 
the input ports. When
this update occurs it waits for the part of the pipeline to execute and 
unwind for each port
sequentially. That would mean that the filter would wait for each process 
to do its thing
before moving to the next input. Is that what is really happening or am I 
reading the code
incorrectly. I have done some timing of this example and get no speedup. My 
platform is
MPI on a cluster of workstations. The program executes correctly and I get 
the iso surface
but there is no speed up when I increase the number of processors. For the 
program to
execute in parallel there would need to be some asynchronous update 
mechanism and I
cant seem to find it. Of course it is entirely possible that I am mistaken. 
I know there are
several parallel users out there. Anyone experience the same problems?

Best Regards,
Dave Semeraro
Dave Semeraro Ph.D.
Site Scientific Visualization Lead
ASC MSRC Wright Patterson Air force Base

Visualization and Virtual Environments Group
NCSA University of Illinois
605 E. Springfield Ave.
Champaign, IL 61820
Semeraro at ncsa.uiuc.edu
(217) 244-1852