The LaBS software is the R&D software developed within the framework of a research consortium bringing together CS GROUP, Renault, Airbus, the Ecole Centrale de Lyon and Aix Marseille University. CS GROUP industrializes and markets this software under the name of ProLB.
Website : http://scalable-hpc.eu/
Website : http://www.prolb-cfd.com/
LinkedIn link : https://www.linkedin.com/showcase/prolb/
Europe, through the EuroHPC joint initiative (https://eurohpc-ju.europa.eu/index_en), seeks to develop a world-class supercomputing ecosystem.
The European EuroHPC SCALABLE project, coordinated by CS GROUP, is being carried out with leading industrial and academic partners: CS GROUP (France), CERFACS (France), Neovia Innovation (France), Renault (France), Airbus Operations GmbH (Germany), Friedrich-Alexander-Universität Erlangen-Nürnberg (Germany)
This project is also supported in France by the National Research Agency (ANR).
The consortium partners are working together to improve the performance, scalability and energy efficiency of industrial computational fluid dynamics (CFD) software based on the Lattice Boltzmann Method (LBM). The LBM method, an alternative to Finite Volume Navier-Stokes methods, has already evolved in recent years to become a reliable alternative to conventional CFD. In several engineering applications, LBM has proven to be about an order of magnitude faster than Navier-Stokes approaches in a fair comparison and in comparable industrial simulation scenarios.
In the EuroHPC approach, LBM is particularly well suited to exploit advanced supercomputer architectures through vectorization, GPU gas pedals, and massive parallelization.
In the public domain research code waLBerla, superb performance and scalability have been demonstrated: waLBerla has for example been used to perform a computation on a mesh of more than one trillion (10^12) cells on one of the petascale supercomputers of the European PRACE network. However, waLBerla is not compatible with industrial applications due to the lack of simple management of complex geometries and the general complexity of implementation for non-HPC experts.
On the other hand, the industrial CFD software LaBS/ProLB already has such industrial capabilities with a proven high level of maturity, but its performance can be improved to fully exploit future supercomputers.
SCALABLE seeks to take the best advantage of these two existing CFD tools (waLBerla and LaBS/ProLB) to obtain the best compromise in terms of raw performance and constraints of an industrial code. The axes of the work done are the following:
– Audit and monitoring tools of the performance and energy consumption of each step and sub-function of the software
– Analysis and redesign of data structures to improve load balancing, memory access, etc.
Study of the adaptation and porting of LBM software on new HPC architectures such as ARM, GPU, …