Performance Evaluation of Unified Parallel C for Molecular Dynamics

Abstract

Partitioned Global Address Space (PGAS) integrates the concepts of shared memory programming and the control of data distribution and locality provided by message passing into a single parallel programming model. The purpose of allying distributed data with shared memory is to cultivate a locality-aware shared memory paradigm. PGAS is comprised of a single shared address space, which is partitioned among threads. Each thread has a portion of the shared address space in local memory and therefore it can exploit data locality by mainly doing computation on local data. Unified Parallel C (UPC) is a parallel extension of ISO C and an implementation of the PGAS model. In this paper, we evaluate the performance of UPC based on a real-world scenario from Molecular Dynamics.