I have access to my universities cluster system and was thinking about implementing a small render farm (128 cores).
I was wondering if anyone had any experience in working with Blender and MPI. I could sure use some advice and of course I plan on releasing the code back to the community.
Thanks for your time,
I'm also interested in Blender MPI support. I downloaded the Blender source, but the code-base is pretty large to "just jump in" and add MPI support. I'm looking for some direction on where to begin... In the mean time, I'm going to work on writing simple MPI applications to become familiar with MPI coding.
For a render farm, you don't need MPI. Look into the command line options of Blender; it's possible to run it on a .blend file and tell it a range of frames to render. That makes it possible to parallelize a render job fairly straightforwardly just by having different machines render different subsets of an animation. I think there are people out there who have written scripts to take advantage of this and make render farm software.
Blender already takes advantage of multiple cores on one machine.
Where MPI or something like it would be really nice is in the fluid simulation code, which, for good results, is more computationally intensive than rendering.