There's a posting in the archives titled Re: [opendx-users] About DXMPI
that gives a fair overview of it; I also have a document I've been working
on slowly that I'll try to finish up eventually.
To answer your specific questions, I use OpenDX/MPI on Linux clusters as
well as AIX POE systems.
There are several rendering approaches available in OpenDX/MPI. One uses
OpenGL on the slave nodes to render, then reads back the RGB pixels and Z
buffer info from the graphics cards and uses a binary swap algorithm to
composite the final image. Another uses a hacked version of Mesa on the
slaves to render in software, then binary-swaps Z-sorted lists of
transparent and opaque surfaces to provide correct order-independent
transparency. Still another uses the parallel extensions to OpenGL in
Chromium and lets the Chromium environment handle the parallel rendering.
Finally, you can always bring the renderable geometry back to the master
node for rendering.
Greg
