VI-HPS Tutorial: IEEE Cluster 2010

Heraklion, Crete, Greece
24 September 2010

Practical parallel application performance engineering using innovative tools

Presenters: Level: Introductory 60% Intermediate 25% Advanced 15%

Abstract: This tutorial presents state-of-the-art tools for engineering performant parallel applications on computer clusters with MPI and/or OpenMP. The suite of tools developed by the Virtual Institute for High Productivity Supercomputing (VI-HPS) are introduced, including Scalasca, Vampir and Periscope. The tools support automated and manually-customizable measurement and analyses with hardware counter metrics as well as communication and synchronization overheads.  A series of hands-on exercises are included which participants are encouraged to follow on their notebook computers using a provided Live-DVD with a bootable typical HPC cluster Linux environment.  This will offer practical experience using the tools and help prepare participants to apply modern methods for locating and diagnosing performance bottlenecks in real-world parallel applications up to the largest scales.

Goals

This full-day tutorial will:
  • give an overview of the VI-HPS programming tools suite
  • explain the functionality of individual tools, and how to use them effectively
  • offer hands-on experience using the tools with a LiveDVD

Day 1 Friday 24th September
08:30 (registration & set-up of notebook computers)
09:00 Welcome & Introduction [Wylie, JSC]
  • VI-HPS & POINT parallel productivity tools
  • Building and running the NPB-MPI-BT example code
    • 09:45 Introduction to parallel application engineering [Gerndt, TUM]
    10:30 (break)
    11:00 Scalasca performance analysis toolset [Wylie, JSC]
  • Scalasca hands-on tutorial exercises
  • Scalasca case studies
    • 12:30 (lunch)
    14:00 Vampir performance analysis toolset [Nagel, TUD-ZIH]
  • Vampir hands-on tutorial exercises
    • 15:30 (break)
    16:00 Periscope automatic performance analysis tool [Gerndt, TUM]
  • Periscope hands-on tutorial exercises
    • 17:15 Review of day
    17:30 (adjourn)

    Participants are expected to use their own notebook computers for the tutorial using a Live-DVD provided by us for hands-on exercises: it may be possible to arrange alternatives for those who don't have access to an x86-compatible notebook computer with DVD drive, if the organizers are informed in advance.

    VI-HPS Tools

    • MARMOT is a free correctness checking tool for MPI programs developed by TUD-ZIH and HLRS.
    • PAPI is a free library interfacing to hardware performance counters developed by UTK-ICL, used by Periscope, Scalasca, VampirTrace, and multiple other tools.
    • Periscope is a prototype automatic performance analysis tool using a distributed online search for performance bottlenecks being developed by TUM.
    • Scalasca is an open-source toolset developed by JSC that can be used to analyze the performance behaviour of parallel applications and automatically identify inefficiencies.
    • Vampir is a commercial framework and graphical analysis tool developed by TUD-ZIH to display and analyze trace files.
    • VampirTrace is an open-source library for generating event trace files which can be analyzed and visualized by Vampir.

    Hardware and Software Platforms

    The VI-HPS tools support and are installed on a wide variety of HPC platforms, including: After completing the tutorial, participants will be prepared to use the VI-HPS tools to analyze and tune their own application codes.

    Further information and registration

    Tutorial participation is included with registration for IEEE Cluster 2010, which must be done on the conference website.