Parallel Performance Project Research Paper

Research Paper

Characterizing Shared Memory and Communication Performance: A Case Study of the Convex SPP-1000
Gheith Abandah and Edward S. Davidson
Technical Report CSE-TR-277-96, University of Michigan, Jan 96.

Abstract

The objective of this paper is to develop models that characterize the memory and communication performance of shared-memory multiprocessors which is crucial for developing efficient parallel applications for them. The Convex SPP-1000, a modern scalable distributed-memory multiprocessor that supports the shared-memory programming paradigm, is used throughout as a case study. The paper evaluates and models four aspects of SPP-1000 performance: scheduling, local-memory, shared-memory, and synchronization. Our evaluation and modeling are intended to supply useful information for application and compiler development.