33. Massively Parallel Simulation of Plasma Turbulence with the Sparse Grid Combination Technique
Authors: Mario Heene (University of Stuttgart)Alfredo Parra Hinjosa (Technical University Munich)Dirk Pflüger (University of Stuttgart)
Abstract: The problem sizes for the solution of higher-dimensional PDEs, such as for the simulation of plasma turbulence in a fusion device, often are very limited due to the exponential increase in the degrees of freedom with the dimensionality. With the Sparse Grid Combination Technique (SGCT) we can mitigate this so-called curse of dimensionality and push the computational limits of high-dimensional simulations significantly.
Offering a second level of parallelism, the SGCT minimizes the need for global communication and ensures scalability on future (exascale) HPC systems. First experimental results demonstrate the scalability of our distributed algorithm up to 180,225 cores on the supercomputer Hazel Hen. Furthermore, the inherent data redundancy of this technique enables fault-tolerance on the algorithmic level at low cost without the need of checkpointing or process replication. These are the key issues we investigate within the project EXAHD of Germany's priority program "Software for Exascale Computing" (SPPEXA).
Two-page extended abstract: pdf