ZURUECK HOCH VOR INHALT SUCHEN

» Back to overview
Proposing Institution

Lehrstuhl für Numerische Methoden der Plasmaphysik,TUM
Project Manager

Dr. Katharina Kormann
Boltzmannstraße 2
85748 Garching
Abstract
Numerical simulations are of key importance to improve the understanding of plasmas in nuclear fusion devices which are prone to nonlinear instabilities such as turbulence. The fundamental model in plasma physics is a kinetic description by a 6d phase space distribution function solving the Vlasov-Maxwell equations, or, for low-frequency phenomena, the Vlasov-Poisson equations. Due to the high memory footprint of 6d grids, an efficient parallelization that scales to large core numbers on HPC systems is essential for the success of grid-based algorithms to solve such models.Large-scale simulations on grids of at least 128^6 points are highly relevant to plasma turbulence simulations and will become state-of-the-art, soon.Within the framework of the SeLaLib library, we are developing a high-performance implementation of a semi-Lagrangian description for the Vlasov equation. To tackle the memory challenge we have recently implemented a distributed memory solver based on a domain decomposition approach using MPI. On the MPI process level the code is parallelized via OpenMP directives. Recently, the code has been demonstrated to show excellent scaling on up to 20480 cores on the HPC system Hydra of the Max Planck Society.To further prepare and optimize the code in terms of performance and scalability we would like take the great opportunity to join the 2017 "LRZ Scaling Workshop: Emergent Applications". Therefore, we would like to kindly apply for a "test project" on SuperMUC.

Impressum, Conny Wendler