Measuring High Performance Computing Productivity

  • Authors:

  • Stuart Faulk;John Gustafson;Philip Johnson;Adam Porter;Walter Tichy;Lawrence Votta

  • Affiliations:

  • COMPUTER AND INFORMATION SCIENCE, UNIVERSITY OF OREGON, EUGENE, OR 97405, USA;ECS INC., PLEASANTON, CA 94588, USA;COLLABORATIVE SOFTWARE DEVELOPMENT LABORATORY, UNIVERSITY OF HAWAII, HONOLULU, HI 96822, USA;COMPUTER SCIENCE DEPARTMENT, UNIVERSITY OF MARYLAND, MD 20742, USA;INFORMATICS, UNIVERSITY OF KARLSRUHE, 76128 KARLSRUHE, GERMANY;SUN MICROSYSTEMS INC., MENLO PARK, CA 94025, USA

  • Venue:
  • International Journal of High Performance Computing Applications
  • Year:
  • 2004

Quantified Score

Hi-index 0.02

Visualization

Abstract

One key to improving high performance computing (HPC) productivity is to find better ways to measure it. We define productivity in terms of mission goals, i.e. greater productivity means that more science is accomplished with less cost and effort. Traditional software productivity metrics and computing benchmarks have proven inadequate for assessing or predicting such end-to-end productivity. In this paper we introduce a new approach to measuring productivity in HPC applications that addresses both development time and execution time. Our goal is to develop a public repository of effective productivity benchmarks that anyone in the HPC community can apply to assess or predict productivity.