Abstract

The performance of the Global Array shared-memory nonuniform memory-access programming model is explored on the I-WAY, wide-area-network distributed supercomputer environment. The Global Array model is extended by introducing a concept of mirrored arrays. Latencies and bandwidths for remote memory access are studied, and the performance of a large application from computational chemistry is evaluated using both fully distributed and also mirrored arrays. Excellent performance can be obtained with mirroring if even modest (0.5 MB/s) network bandwidth is available. 1 Introduction We are interested in studying the performance of current massively parallel chemistry codes on distributed computing resources, in particular the variable latency, limited bandwidth, distributed environment of I-WAY (described below). Our applications employ the Global Array (GA) non-uniform memory-access (NUMA) programming model [1-3] and reference data in blocks to increase data locality in order to reduce se...

Citations