A COMPARISON OF PETSC LIBRARY AND HPF IMPLEMENTATIONS OF AN ARCHETYPAL PDS COMPUTATION

Two paradigms for distributed-memory parallel computation that free th e application programmer from the details of message passing are compa red for an archetypal structured scientific computation-a nonlinear, s tructured-grid partial differential equation boundary value problem-us ing the same algorithm on the same hardware. Both paradigms, parallel libraries represented by Argonne's PETSc, and parallel languages repre sented by the Portland Group's HPF, are found to be easy to use for th is problem class, and both are reasonably effective in exploiting conc urrency after a short learning curve. The level of involvement require d by the application programmer under either paradigm includes specifi cation of the data partitioning (corresponding to a geometrically simp le decomposition of the domain of the PDE). Programming in SPMD style for the PETSc library requires writing the routines that discretize th e PDE and its Jacobian, managing subdomain-to-processor mappings (affi ne global-to-local index mappings), and interfacing to library solver routines. Programming for HPF requires a complete sequential implement ation of the same algorithm, introduction of concurrency through subdo main blocking tan effort similar to the index mapping), and modest exp erimentation with rewriting loops to elucidate to the compiler the lat ent concurrency. Correctness and scalability are cross-validated on up to 32 nodes of an IBM. SP2. (C) 1998 Published by Elsevier Science Lt d. All rights reserved.