MPI-version versus Radiance with rpiece

It is possible to obtain a copy of the MPI version of Radiance from LBNL by
signing a license agreement for the source code. This version uses
parallelization method known as message passing. It was written for a Cray
and can be modified to run on an SGI without too much difficulty by someone
familiar with parallel programming. This is, however, not necessary because
the current version of Radiance can make use of multiple processors to speed
up rendering by using the rpiece function. This is explained in Chapter 15
of the Radiance text. There was a also an email exchange on this subject in
June which can be found in the Radiance-general archives in June under
(rpiece and parallel processing). The same approach will work on clusters of
PC's running Linux. I hope to report on a comparison between running
Radiance on a multi-processor SGI machine versus running it on a cluster of
Linux PC's soon. Note that rpiece speeds up rendering with rpict by
distributing the work to of rendering an image to multiple processors. It
does not speed up creation of the octree, however. Octree creation can be
speeded up by breaking the scene into several rad files and creating an
octree separately for each and then combining them with the -i option in the
oconv function. The different sub-octrees could be created in parrallel on
different processors.

A final word about the MPI version: it does not correspond to the current
version of Radiance but an earlier one. It is not clear that the MPI version
would be any faster than the current version using rpiece. I have had very
good results.




Message: 1
Date: Fri, 26 Jul 2002 10:03:15 -0700
From: Randolph Fritz <>
Subject: Re: [Radiance-general] MPI-Version of radiance?

On Tue, Jul 23, 2002 at 12:25:39PM +0200, wrote:
> I just found that there has been done some work on a mpi-radiance in
> the past. This seams to be quite an interesting development but
> AFAIK only a very old version of rpict has been modified to make use
> of mpi. Are there any new versions of this, or is there anybody
> working on a "parallel-radiance"? (I ask this after having rendered
> some pictures on a sun 880, an 4xultrasparc - what still took
> days... ;-).

It's a good question...I worked the with the MPI version a few years
ago. Later on, I found out that LBL had a parallelized version
already but that it wasn't published. Anyone willing to speak more
directly to this? What's the status?

Also, what are people's favorite clusters?



Radiance-general mailing list

End of Radiance-general Digest