Simulation Software for Supercomputers 2009
as part of the First International Workshop on Computational Engineering - Special Topic Fluid-Structure Interaction, organised by the DFG Research Unit 493 Fluid-Structure Interaction in Herrsching (Aug 12-14, 2009).
Simulation Software for Supercomputers
Scientific Computing, and the Computational Sciences, in general, have emerged into a new discipline, which is driven by a multitude of demands that stem from applied mathematics, from the respective application fields in science and engineering, and also from high performance computing. Application scenarios have shifted from individual simulation runs to parameter studies and optimisation processes, from models of limited complexity to multiphysics problems, and from comparably simple parallel clusters to hierarchically organised HPC platforms that offer parallelism on all scales.
As a result, simulation software, in particular on supercomputers, faces a multitude of new and oftentimes conflicting challenges. As models grow in complexity, simulation software itself is growing in extent and complexity. The scenario of a single-purpose code, that can be implemented within the time frame of a single PhD project, is replaced by that of generic, more versatile, interacting frameworks that are developed as group efforts. Systematic software engineering, rigorous testing and verification, and more formal approaches to developing software, in general, have therefore gained importance in the scientific computing community. Recent trends in hardware development, such as the trend to multi- and manycore CPUs have added to this scenario, as modern HPC software has to consider efficient access to memory and communication resources in order to ensure acceptable performance. Finally, the increasing parallelism on all scales, including the use of special-purpose hardware (graphical processing units, accelerator processors) need to be tackled by HPC software.
In the light of the recent call "HPC Software für skalierbare Parallelrechner" for BMBF projects, Michael Bader, Miriam Mehl, and Gerhard Wellein organised a minisymposium with the First International Workshop on Computational Engineering. Many of the minisymposium speakers reported on first results of the respective BMBF projects, and thus reflect the wide-spread activities regarding Simulation Software in Germany:
- Ulrich Ruede, University of Erlangen-Nuremberg:
Massively Parallel Algorithms for Fluid-Structure Interaction with Moving Objects.
- Martin Buchholz, Technische Universität München:
Load Balancing of Heterogeneous Particle Distributions in Molecular Dynamics.
- Carla Guillen, Leibniz Supercomputing Centre, München; Lamia Djoudi and Michael Gerndt, Technische Universität München:
Performance Analysis for HPC Applications.
- Paul Willems, University of Wuppertal:
On extending the MRRR-algorithm for the tridiagonal symmetric eigenproblem.
- Stephan Schmidt, Department of Mathematics, Universität Trier:
A 2589 Line Topology Optimization Code Written for the Graphics Card.
- Matthias Bolten, University of Wuppertal:
A highly scalable multigrid algorithm for particle simulation
- Oliver Fortmeier, RWTH-Aachen University:
Parallel Two-phase Flow Simulations.
- Gerhard Wellein, University of Erlangen-Nuremberg:
Numerical simulation on recent multi-core designs: Some comments on programming effort and performance.