HPF on Fine-Grain Distributed Shared Memory: Early Experience

This paper examines the performance of a suite of HPF applications on a network of workstations using two different compilation approaches: generating explicit message-passing code, and generating code for a shared address space provided by a finegrain distributed shared memory system (DSM). Preliminary experiments indicate that the DSM approach performs with usually a small slowdown compared to the message passing approach on regular programs, yet enables efficient execution of non-regular program