Ron says:

> There's a lot of MPI you (well, I) don't need.

I would say you need almost none of it.  Around 1993 as part of my PhD
project I designed a parallel computing library based on the BSP model
with exactly six primitives (of which two are housekeeping and one
more is arguably redundant) - see attached man page.  The simplicity
meant that it could be implemented quickly and efficiently on a range
of platforms from workstation clusters to the Cray T3D, and users
found it easy to use and understand.

Sadly BSP was too unconventional (too simple?) to become popular.

-- Richard