Evaluating the parallel performance of a heterogeneous system

Abstract

Heterogeneous parallel systems are becoming increasingly more common, especially with the increasing use of cluster computers, such as Beowulf systems, and networks of workstations for parallel computing. Measuring and evaluating the performance of such parallel systems is not straightforward. In particular, conventional techniques such as computing speedup and efficiency are not appropriate for evaluating the performance of a heterogeneous system, and even have their limitations for homogeneous systems. This paper looks at alternative ways of measuring and evaluating the parallel performance of a heterogeneous system, such as linear speed, and extends this with the concept of linear efficiency.