SDL Highly Scalable Fluid & Solids Engineering

SDL Mission

The Simulation and Data Lab Highly Scalable Fluids & Solids Engineering (SDL FSE) aims at supporting users of the engineering sciences who have already developed parallel codes but need support for the use of massively parallel systems regarding high scalability, memory optimization, programming of hierarchic computer architectures, and performance optimization on computer nodes.

Introduction

Experimentally testing new designs in the field of engineering is often time consuming and expensive. Numerical simulations, in contrast, enable to easily vary model parameters and to explore a problem in all detail. To accurately predict the physics of a realistic problem, large-scale simulations are required that pose new challenges to simulation codes and that necessitate the power of supercomputers. The SDL contributes to solving theses challenges by increasing the parallel efficiency of simulation codes and establishing an interdisciplinary interface between engineers and the HPC community. It develops highly scalable software suited for HPC systems and pushes research in the fields of computational fluid dynamics, aeroacoustics, shape optimization, and machine learning. The SDL seeks solutions for questions such as: What is the origin of respiratory pathologies? How can airplane noise be reduced? How can pharmaceutical and chemical processes be optimized? It naturally integrates into the support and research structure of JSC and JARA-CSD and delivers efficient porting and tuning solutions for codes to run on current and future supercomputer architectures. Synergies arise from cooperations with JSC's and JARA-CSD’s Cross-Sectional Teams (CSTs). They are explored to find performance bottlenecks and to visualize scientific big data.

Know-How

The SDL has an interdisciplinary setup and has members with backrounds in Engineering and Computer Science. It hence gathers knowledge in various disciplines:

• finite-element, finite-volume, lattice-Boltzmann, and discontinuous Galerkin methods
• multi-level parallelization techniques
• Cartesian grid-based methods
• parallel mesh generation and partitioning
• shape optimization
• fluid-structure interaction
• multi-physics coupling
• machine learning algorithms