Custom Hot Spot Analysis of HPC Software with the Vampir Performance Tool Suite

Conference paper
First Online:

Abstract

The development and maintenance of scalable, state-of-the-art applications in High Performance Computing (HPC) is complex and error-prone. Today, performance debuggers and monitors are mandatory in the software development chain and well established. Like the applications, the tools themselves have to keep track of the developments in system and software engineering. Prominent developments in this regard are for example hybrid, accelerated, and energy aware computing. The ever increasing system complexity requires tools that can be adjusted and focused to user specific interests and questions. This article explains how the performance tool Vampir can be used to detect and highlight user-defined hot spots in HPC applications. This includes the customization and derivation of performance metrics, highly configurable performance data filters and a powerful comparison mode for multiple program runs. The latter allows to keep track of the performance improvements of an application during its evolution.

Keywords

High Performance Computing Trace File Event Match Program Phase Context View 
These keywords were added by machine and not by the authors. This process is experimental and the keywords may be updated as the learning algorithm improves.
This is a preview of subscription content, log in to check access.

References

  1. 1.
    Brunst, H., Knüpfer, A.: Vampir. In: D. Padua (ed.) Encyclopedia of Parallel Computing, pp. 2125–2129. Springer US (2011). DOI  10.1007/978-0-387-09766-4_60
  2. 2.
    Burau, H., Widera, R., Hönig, W., Juckeland, G., Debus, A., Kluge, T., Schramm, U., Cowan, T., Sauerbrey, R., Bussmann, M.: PIConGPU: A fully relativistic particle-in-cell code for a GPU cluster. Plasma Science, IEEE Transactions on 38(10), 2831–2839 (2010). DOI 10.1109/TPS.2010.2064310CrossRefGoogle Scholar
  3. 3.
    Hackenberg, D., Schöne, R., Molka, D., Müller, M., Knüpfer, A.: Quantifying power consumption variations of HPC systems using SPEC MPI benchmarks. Computer Science - Research and Development 25, 155–163 (2010). URL http://dx.doi.org/10.1007/s00450-010-0118-0. 10.1007/s00450-010-0118-0
  4. 4.
    Kogge, P.: The tops in flops. IEEE Spectrum 48, 48–54 (2011). DOI 10.1109/MSPEC.2011.5693074. URL http://ieeexplore.ieee.org/stamp/stamp.jsp?tp=&arnumber=5693074 Google Scholar
  5. 5.
    Pennycook, S.J., Hammond, S.D., Jarvis, S.A., Mudalige, G.R.: Performance analysis of a hybrid MPI/CUDA implementation of the NASLU benchmark. SIGMETRICS Perform. Eval. Rev. 38(4), 23–29 (2011). DOI 10.1145/1964218.1964223. URL http://doi.acm.org/10.1145/1964218.1964223
  6. 6.
    Skamarock, W.C., Klemp, J.B.: A time-split nonhydrostatic atmospheric model for weather research and forecasting applications. Journal of Computational Physics 227(7), 3465–3485 (2008). DOI 10.1016/j.jcp.2007.01.037MathSciNetMATHCrossRefGoogle Scholar
  7. 7.
    William, T., Berry, D.K., Henschel, R.: Analysis and optimization of a molecular dynamics code using PAPI and the Vampir toolchain. In: Cray User Group 2012. CUG, Stuttgart (2012). URL https://portal.futuregrid.org/sites/default/files/2012-CUG-pap142.pdf

Copyright information

© Springer-Verlag Berlin Heidelberg 2013

Authors and Affiliations

  1. 1.Center for Information Services and HPC (ZIH)Technische Universität DresdenDresdenGermany

Personalised recommendations