Applying cutting-edge fluid dynamics at the exascale
Computational fluid dynamics (CFD) involves the application of numerical analysis and big data to assess and solve problems involving fluid flows. CFD is critical to a wide range of industrial areas, including aerodynamics and aerospace analysis, weather simulation and even visual effects for film and games. Faster and more powerful CFD calculations would therefore benefit a range of end users, through improving the accuracy and speed of computer models and simulations. To this end, the SCALABLE project, coordinated by CS GROUP in France, sought to develop and test a new class of computational methods, with a view to achieving this goal.
Simulating highly complex fluid flows
To do this, the SCALABLE project team applied what are known as Lattice Boltzmann Methods (LBMs), a CFD approach shown to be significantly faster than conventional CFD approaches. LBMs are numerical techniques that can be used to simulate highly complex fluid flows. Their strength lies in their ability to easily represent complex physical phenomena, ranging from multiphase flows to chemical interactions. In the context of high-performance computing (HPC), there is huge potential in combining LBMs with advanced supercomputer architectures. A key drawback however is the fact that LBM techniques consume a lot of energy compared to other numerical methodologies. This has limited their application in the HPC space. The goal of SCALABLE was therefore to find a way of applying LBMs in the context of HPC, to deliver the efficiency and scalability needed for upcoming European exascale computing systems.
Performance, scalability, efficiency
The project brought together a group of eminent industrial and academic partners from three European countries. The objective was to develop LBM-based CFD software, capable of achieving unprecedented performance, scalability and above all energy efficiency. The project team utilised the public domain research code waLBerla, which offers excellent performance and scalability, and is currently run on some of the largest HPC clusters in the world. The project also made use of industrial CFD software called ProLB (previously known as LaBS). This software solution, based on LBMs, delivers accurate aerodynamic modelling to help engineers make early design decisions. Although at a high level of maturity, there is still room for performance improvements.
Cutting-edge modelling technology
The SCALABLE team successfully combined the benefits of these two existing CFD tools, to deliver cutting-edge modelling technology tailored for exascale computing. In doing so, the project also helped to break down silos between the scientific computing world and that of physical flow modelling. The prototype software was trialled in a number of test cases involving convection and turbulence modelling. These simulations are critical to the design of things such as airplane engines, wind turbines and cooling systems. The project also specifically looked at modelling airplane landing gear. This provided a useful test case for assessing the precision and performance of the project’s CFD prototype software. The successes and breakthroughs achieved during SCALABLE will now be built upon, by the project’s academic and industrial partners. The SCALABLE project was carried out with support from the European High Performance Computing Joint Undertaking (EuroHPC JU), an initiative set up to develop a world-class supercomputing ecosystem in Europe.
Keywords
SCALABLE, EuroHPC JU, HPC, exascale, supercomputing, fluid dynamics, CFD, aerospace, automotive