MapReduce-MPI library features and non-features

• open-source C++ library
• C interface provided, thus callable from any hi-level language: C++, C, Fortran, scripting language, etc
• Python wrapper provided, so MapReduce programs can be written in Python
• small, portable code base
• operates in serial or on any parallel platform that supports MPI

• no fault tolerance
• no out-of-core capability

Fault tolerance is really an MPI issue, which future versions of MPI may address. Since big parallel machines never crash, nor do their processors die, hopefully it's a non-issue for performing MapReduce operations.

We plan to add an out-of-core capability to the MR-MPI library, though the fact that many big parallel machines do not have a disk per processor is a complicating factor.