Characterizing and representing workloads for parallel computer architectures

Experimental design of parallel computers calls for quantifiable methods to compare and evaluate the requirements of different workloads within an app lication domain. Such methods can help establish the basis for scientific d esign of parallel computers driven by application needs, to optimize perfor mance to cost. In this paper, a framework is presented for representing and comparing workloads, based on the way they would exercise parallel machine s. This workload characterization is derived from parallel instruction cent roid and parallel workload similarity. The centroid is a workload approxima tion that captures the type and amount of parallel work generated by the wo rkload on the average. The centroid is a simple measure that aggregates ave rage parallelism, instruction mix, and critical path length. When captured with abstracted information about communication requirements, the result is a powerful tool in understanding the requirements of workloads and their p otential performance on target machines. The workload similarity is based o n measuring the normalized Euclidean distance (ned) between workload centro ids. It will be shown that this workload representation method outperforms comparable ones in accuracy, as well as in time and space requirements. Ana lysis of the NAS Parallel Benchmark workloads and their performance will be presented to demonstrate some of the applications and insight provided by this framework. This will include the use of the proposed framework for pre dicting the performance of real-life workloads on target machines, with goo d accuracy. (C) 2000 Published by Elsevier Science B.V. All rights reserved .