A message-passing library interface like MPI or PVM is only one interface between only two components the complete environment
seen be the user of a parallel system. Here we discuss a larger set of components and draw attention to the usefulness of
considering them separately. Such an approach causes us to focus on the interfaces among such components. Our primary motivation
is the efficient use of large clusters to run MPI programs, and we describe current efforts by our group at Argonne to address
some of the interface issues that arise in this context.
This work was supported by the Mathematical, Information, and Computational Sciences Division subprogram of the Office of
Advanced Scientific Computing Research, U.S. Department of Energy, under Contract W-31-109-Eng-38.