COMMUNICATION OPTIMIZATIONS FOR IRREGULAR SCIENTIFIC COMPUTATIONS ON DISTRIBUTED-MEMORY ARCHITECTURES

This paper describes a number of optimizations that can be used to sup port the efficient execution of irregular problems on distributed memo ry parallel machines. These primitives (1) coordinate interprocessor d ata movement, (2) manage the storage of, and access to, copies of off- processor data, (3) minimize interprocessor communication requirements , and (4) support a shared name space. We present a detailed performan ce and scalability analysis of the communication primitives. This perf ormance and scalability analysis is carried out using a workload gener ator, kernels from real applications, and a large unstructured adaptiv e application (the molecular dynamics code CHARMM). (C) 1994 Academic Press, Inc.