Statistical performance modeling: Case study of the NPB 2.1 results

  • Erich Strohmaier
Workshop 16: Performance Evaluation and Prediction
Part of the Lecture Notes in Computer Science book series (LNCS, volume 1300)


With the results of version 2.1 a consistent set of performance measurements of the NAS Parallel Benchmarks (NPB) are available. Unchanged portable MPI code was used for this set of 269 single measurements. In this study we investigate how this amount of information can be condensed. We present a methodology for analyzing performance data not requiring detailed knowledge of the codes. For this we study several different generic timing models and fit the reported data. We show that with a joint timing model for all codes and all systems the data can be fitted reasonable well. The timing model also contains only a minimal set of free parameters. This method is usable in all cases where the analysis of results from complex application code benchmarks is necessary.


Unable to display preview. Download preview PDF.

Unable to display preview. Download preview PDF.


  1. 1.
    D. Bailey, J. Barton, T. Lasinski, and H. Simon (editors). The NAS parallel benchmarks. Technical Report RNR-91-02, NASA Ames Research Center, January 1991.Google Scholar
  2. 2.
    W. Saphir, A. Woo and M. Yarrow. The NAS parallel benchmarks 2.1 Results. Technical Report NAS-96-O1, NASA Ames Research Center, August 1996.Google Scholar
  3. 3.
    Horst D. Simon and Erich Strohmaier. Statistical Analysis of NAS Parallel Benchmarks and LINPACK Results. In Bob Hertzberger and Guiseppe Serazzi, editors, High-Performance Computing and Networking, pages 626–633, May 1995.Google Scholar
  4. 4.
    Raj Jain. The Art of Computer Systems Performance Analysis. Wiley, 1991Google Scholar
  5. 5.
    Strohmaier, Erich. Extending the Concept of Computational Similarity for Analyzing Complex Benchmarks. Technical Report 43, Rechenzentrum der Universitaet Mannheim, April 1995Google Scholar
  6. 6.
    Vipin Kumar et al.. Introduction to Parallel Computing: Design and analysis of parallel algorithms. Benjamin/Cummings, 1994.Google Scholar
  7. 7.
    Jurgen Brehm and Patrick H. Worley and Manish Madhukar. Performance Modeling for SPMD Message-Passing Programs. Technical Report TM-13254, Oak Ridge National Laboratory, June 1996 *** DIRECT SUPPORT *** A0008C42 00034Google Scholar

Copyright information

© Springer-Verlag Berlin Heidelberg 1997

Authors and Affiliations

  • Erich Strohmaier
    • 1
  1. 1.Computer Science DepartmentUniversity of TennesseeKnoxville

Personalised recommendations