Task management, virtual shared memory, and multithreading in a distributed memory implementation of Sisal

Abstract

This paper describes the design and initial performance of a runtime system for implementing Sisal [5] on a distributed memory multiprocessor. The runtime system provides support for task management and distribution, virtual shared memory, and multithreading. Tasks represent portions of code that can be executed in parallel, and we examine the performance effects of both flat and multi-level task distribution strategies. We introduce our virtual shared memory scheme, called VISA, and investigate a block cyclic mapping function using fixed and variable address translation. Finally, we study the effectiveness of multithreading in hiding latency for our current nCUBE/2-based implementation. In particular, we examine a machine-independent design for thread representation, thread switching, and split-phase transactions. We analyze the costs of multithreading and provide initial performance results.