Trace-Based Parallel Performance Overhead Compensation (original) (raw)

Abstract

Tracing parallel programs to observe their performance introduces intrusion as the result of trace measurement overhead. If post-mortem trace analysis does not compensate for the overhead, the intrusion will lead to errors in the performance results. We show that measurement overhead can be accounted for during trace analysis and intrusion modeled and removed. Algorithms developed in our earlier work [5] are reimplemented in a more robust and modern tool, kojak [12] , allowing them to be applied in large-scale parallel programs. The ability to reduce trace measurement error is demonstrated for a Monte-Carlo simulation based on a master/worker scheme. As an additional result, we visualize how local perturbation propagates across process boundaries and alters the behavioral characteristics of non-local processes.

Preview

Unable to display preview. Download preview PDF.

Similar content being viewed by others

References

  1. Lamport, L.: Time, Clocks and the Ordering of Events in a Distributed System. CACM 21(7), 558–565 (1978)
    MATH Google Scholar
  2. Fagot, A., de Kergommeaux, J.: Systems Assessment of the Overhead of Tracing Parallel Programs. In: Euromicro Workshop on Parallel and Distributed Processing, pp. 179–186 (1996)
    Google Scholar
  3. Hollingsworth, J., Miller, B.: An Adaptive Cost System for Parallel Program Instrumentation. In: Euro-Par Conference, August 1996, vol. I, pp. 88–97 (1996)
    Google Scholar
  4. Kranzlmüller, D., Reussner, R., Schaubschläger, C.: Monitor overhead measurement with sKaMPI. In: Margalef, T., Dongarra, J., Luque, E. (eds.) PVM/MPI 1999. LNCS, vol. 1697, pp. 43–50. Springer, Heidelberg (1999)
    Chapter Google Scholar
  5. Malony, A.: Event Based Performance Perturbation: A Case Study. In: Principles and Practices of Parallel Programming (PPoPP), April 1991, pp. 201–212 (1991)
    Google Scholar
  6. Malony, A.: Performance Observability. Ph.D. thesis, University of Illinois, Urbana-Champaign (1991)
    Google Scholar
  7. Malony, A., Shende, S.: Overhead Compensation in Performance Profiling. In: Danelutto, M., Vanneschi, M., Laforenza, D. (eds.) Euro-Par 2004. LNCS, vol. 3149, pp. 119–132. Springer, Heidelberg (2004)
    Chapter Google Scholar
  8. Malony, A., Shende, S.: Overhead Compensation in Parallel Performance Profiling. In: Parallel Processing Letters (2005) (to be pubished)
    Google Scholar
  9. Message Passing Interface Forum. MPI: A Message Passing Interface Standard, Chapter 8, Profiling Interface, Juni (1995), http://www.mpi-forum.org
  10. Sarukkai, S., Malony, A.: Perturbation Analysis of High-Level Instrumentation for SPMD Programs. In: Principles and Practices of Parallel Programming (PPoPP), May 1993, pp. 44–53 (1993)
    Google Scholar
  11. Song, F., Wolf, F., Bhatia, N., Dongarra, J., Moore, S.: An Algebra for Cross-Experiment Performance Analysis. In: Proc. of the International Conference on Parallel Processing (ICPP), Montreal, Canada (August 2004)
    Google Scholar
  12. Wolf, F., Mohr, B.: Automatic performance analysis of hybrid MPI/OpenMP applications. Journal of Systems Architecture 49(10-11), 421–439 (2003); Special Issue Evolutions in parallel distributed and network-based processing
    Article Google Scholar
  13. Wolf, F.: EARL - API Documentation. Technical Report ICL-UT-04-03, University of Tennessee, Innovative Computing Laboratory (October 2004)
    Google Scholar
  14. Wolf, F., Mohr, B.: Specifying Performance Properties of Parallel Applications Using Compund Events. Parallel and Distributed Computing Practices 4(3) (September 2001); Special Issue on Monitoring Systems and Tool Interoperability
    Google Scholar

Download references

Author information

Authors and Affiliations

  1. Innovative Computing Laboratory, University of Tennessee,
    Felix Wolf
  2. Department of Computer and Information Science, University of Oregon,
    Allen D. Malony, Sameer Shende & Alan Morris

Authors

  1. Felix Wolf
  2. Allen D. Malony
  3. Sameer Shende
  4. Alan Morris

Editor information

Editors and Affiliations

  1. Department of Computer Science, St. Francis Xavier University, Antigonish, Canada
    Laurence T. Yang
  2. School of Computer Science/Welsh eScience Centre, Cardiff University, UK
    Omer F. Rana
  3. Dipartimento di Ingegneria dell’ Informazione - Second, University of Naples - Italy, Real Casa dell’Annunziata - via Roma, 29 81031, Aversa (CE), Italy
    Beniamino Di Martino
  4. Computer Science Department, University of Tennessee, 37996-3450, Knoxville, TN, USA
    Jack Dongarra

Rights and permissions

© 2005 Springer-Verlag Berlin Heidelberg

About this paper

Cite this paper

Wolf, F., Malony, A.D., Shende, S., Morris, A. (2005). Trace-Based Parallel Performance Overhead Compensation. In: Yang, L.T., Rana, O.F., Di Martino, B., Dongarra, J. (eds) High Performance Computing and Communications. HPCC 2005. Lecture Notes in Computer Science, vol 3726. Springer, Berlin, Heidelberg. https://doi.org/10.1007/11557654\_72

Download citation

Keywords

Publish with us