Some experiences with mpi-povray

[Heiko Bauke <Heiko.Bauke@Student.UNI-Magdeburg.DE>]

Hi,

some days ago I asked for a good program to demonstrate the power
of beowulf clusters. Now i've tried mpi-povray.
http://www.verrall.demon.co.uk/mpipov/

I compiled mpi-povray on a cluster of Sun workstations (3 single-cpu-
machines and 3 dual-cpu-smp-machines) with lam 6.3.2 and gcc and on a 
Linux cluster (144 Pentium III) with lam 6.5.1 and gcc. mpi-povray is
fine but there are some little problems:

* mpi-povray divides the picture in sub pictures. If there are more
  mpi processes than sub pictures mpi-povray crashes.
* If you are really interested in fast rendering ray tracing pictures
  and not in a public relation show don't use the +D option. Displaying
  the picture while rendering slows the computation down.
* Interpreting the input file (is not parallel) can make a significant 
  part of rendering time.


	Heiko

-- 
-- Den Namen des Rechtes würde man nicht kennen, wenn es das Unrecht
-- nicht gäbe. (Heraklit, gr. Philosoph, 540-480 v. Chr.)
-- Supercomputing in Magdeburg @ http://tina.nat.uni-magdeburg.de
--                 Heiko Bauke @ http://www.uni-magdeburg.de/bauke