An Implementation of the LPAR Parallel Programming Model for Scientific Computations

LPAR is a portable coarse-grain parallel programming model for non-uniform structured scientiic applications running on MIMD message passing architectures. Non-uniform applications, which include N-body methods and adaptive multilevel mesh methods, rely on complex dynamic data structures and are particularly diicult to implement on parallel computers. This paper introduces the LPAR programming abstractions and discusses some important implementation issues. We also present performance results on the Intel iPSC/860 and nCUBE/2 for a vortex dynamics application developed using LPAR.