Mini Symposia

Parallel Computing with FPGA's

  • Erik D'Hollander, University of Ghent
  • Dirk Stroobandt, University of Ghent
  • Abdellah Touhafi, Brussels University Association
  • Description: Field programmable gate arrays are emerging in many areas of high performance computing, either as tailor made signal processor, embedded hardware algorithm, systolic array, floating point accelerator or application specific architecture. However, the design complexity of FPGA's and their role as building block in a high performance computing system is subject to innovative research. Many diverse approaches exist, ranging from novel architectures, hardware-software codesign, evolvable hardware, evolutionary algorithms, reconfigurable computing, to the development of whole compiler, simulation and synthesis frameworks. Because of their versatility as an enabling component in parallel architectures, all areas of FPGA's in parallel computing are addressed, in particular: methodology, performance analysis, architectures, algorithms and applications.

    The Future of OpenMP in the Multi-Core Era

  • Barbara Chapman, University of Houston
  • Dieter an Mey, RWTH Aachen University
  • Description: OpenMP is an Application Programming Interface (API) which is widely accepted as a standard for high-level shared-memory parallel programming. It is a portable, scalable programming model that provides a simple and flexible interface for developing shared-memory parallel applications in Fortran, C, and C++. By September 2007 the upcoming Version 3.0 of the OpenMP specification will be in its public reviewing phase. It will include a novel tasking concept and improved support for nested parallelism. Features to make OpenMP ccNUMA-aware are under intense discussion. Another important approach to broaden the scope of OpenMP is Intel's Cluster OpenMP implementation.

    Among the speakers who have already accepted our invitation to present are Mark Bull, the leader of the OpenMP Language Committee, Larry Meadows, the main driving force behind Intel's Cluster-OpenMP implementation, Eduard Ayguade, the major developer of the well-known Nanos OpenMP research compiler, and Ruud van der Pas, a highly recognized expert in OpenMP application performance issues.

    Scalability and Usability of HPC Programming Tools

  • Felix Wolf, Forschungszentrum Jülich (chair)
  • Daniel Becker, Forschungszentrum Jülich
  • Bettina Krammer, University of Stuttgart
  • Dieter an Mey, RWTH Aachen University
  • Shirley Moore, University of Tennessee
  • Matthias Müller, Technical University of Dresden
  • Description: Facing increasing power dissipation and little instruction-level parallelism left to exploit, computer architects are realizing further performance gains by placing multiple "slower" processor cores on a chip rather than by building faster uni-processors. As a consequence, numerical simulations are being required to harness much higher degrees of parallelism in order to satisfy their growing demand for computing power. However, writing code that runs correctly and efficiently on large numbers of processors and cores is extraordinary challenging and requires adequate tool support for debugging and performance analysis. Unfortunately, increased concurrency levels place higher scalability demands not only on applications but also on software tools. When applied to larger numbers of processors, familiar tools often cease to work in a satisfactory manner (e.g., due to escalating memory requirements or limited I/O bandwidth). In addition, tool usage is often complicated and requires the user to learn a separate interface. This minisymposium will serve as a forum to discuss methods and experiences related to scalability and usability of HPC programming tools, including their integration with compilers and the overall development environment.

    DEISA: Extreme Computing in an Advanced Supercomputing Environment

  • Hermann Lederer, RZG/IPP, Germany (chair)
  • Denis Girou, IDRIS, France
  • Marc André Hermanns, Forschungszentrum Jülich, Germany
  • Gavin Pringle, EPCC, United Kingdom
  • Description: DEISA is a consortium of leading national supercomputing centres in Europe. Starting in 2004 it deploys and operates a persistent, production quality, distributed infrastructure for supercomputing applications. It connects eleven European supercomputing centres in eleven countries via a dedicated high bandwidth network, enabling seamless and transparent access to a Europeanwide shared global file system and coscheduling services. The coordinated operation of this environment is tailored to enable new, ground breaking applications in computational sciences. Scientists from all over Europe have successfully used this platform for extreme computing endeavors with concrete scientific impact. This minisymposium aims to highlight the scientific impacts achieved so far with the DEISA infrastructure, while also giving a forum for the extensive enabling work done for applications running in the DEISA Extreme Computing Initiative (DECI) and encouraging discussion of current limitations of distributed computing as well as future developments.

    Scaling Science Applications on Blue Gene

  • Bill Gropp, Argonne National Laboratory (Chair)
  • Wolfgang Frings, Forschungszentrum Jülich
  • Marc-André Hermanns, Forschungszentrum Jülich
  • Ed Jedlicka, Argonne National Laboratory
  • Kirk Jordan, IBM Deep Computing
  • Fred Mintzer, IBM Watson Research Center
  • Boris Orth, Forschungszentrum Jülich
  • Description: Massively parallel supercomputers like IBM's Blue Gene/L offer exciting new opportunities for scientific discovery by enabling numerical simulations on an unprecedented scale. However, achieving highly scalable performance is often not straightforward as the system's extraordinary level of parallelism and its specialized nodes present challenges to applications in many areas including: communication efficiency, memory usage, and I/O.

    This mini-symposium aims to highlight some of the remarkable scaling and performance results achieved, and bring together users and system experts to discuss possible solutions to yet unresolved issues. It will feature speakers whose applications have run at large scale on the 8K node system at John von Neumann Institute for Computing (NIC) at Research Center Juelich and the 20K node system at IBM's Watson Research Center. The talks will provide a sampling of the different applications and algorithms that have successfully run on Blue Gene. The speakers will discuss best practices, particularly with respect to scaling to tens of thousands of processes and challenges faced in using BG/L to massive scale, and they will showcase some of the breakthrough science that has already been achieved. In addition, IBM and Argonne National Laboratory will present early experiences with Blue Gene/P, the next generation of the IBM Blue Gene system.

    Forschungszentrum Jülich
    D-52425 Jülich