Supercomputing for Simulation in Science and Engineering

Ulrich Ruede
Lehrstuhl für Systemsimulation
Universität Erlangen-Nuernberg
Erlangen, Germany


Abstract:

Science and engineering rely increasingly on computer simulation as a universal methodology. This simulation based engineering must be founded on solid physical principles, well-designed mathematical models, and the skilful use of modern computer technology. In this talk I will illustrate the potential of advanced simulation methods with examples taken from finite element computations and flow simulation. The software has been designed and implemented carefully to support parallel scalability. I will report scalability tests on very large systems, such as the HLRB-II at LRZ in Garching, an SGI Altix with almost ten thousand compute cores. More recently, we have been able to explore the parallel behavior on an IBM Blue Gene, the JUGENE system at Jülich with more than hundred thousand processor cores.