WP4 – Evolution
Lead beneficiary: University of Zagreb (FSB)
- Improve OpenFOAM features that are identified as HPC bottlenecks to fully exploit the evolving HPC hardware.
- Implement strongly coupled solution algorithms resulting from the discretisation of sets of coupled linear and non-linear partial differential equations that appear in physics models of industrial interest.
- Include the use of novel and state-of-the-art numerical algorithms for the iterative solutions of sparse linear systems specifically designed for massive parallelization on exascale platforms, which are beyond the presently commonly used preconditioned Krylov subspace and Algebraic Multigrid (AMG) linear equation solvers.
- Seek synergies with open-source components, such as open-source packages for linear algebra solvers, or open-source libraries for massively parallel I/O.
- Improve code performance in reactive flow simulations to fully exploit the new capabilities of accelerated hardware, as co-processor by moving relevant compute-intensive operations to efficient hardware.
- Use lossy compression algorithms to reduce the amount of memory required by each checkpoint, so as to considerably increase the number of checkpoints and, thus, reduce the cost of recomputations in gradient-based aerodynamic optimization of transient flows, assisted by the adjoint approach.
- Optimize the parallel efficiency of the non-conforming mesh interface methodology (GGI/AMI) used to merge disconnected parts in a single simulation space, by exploiting optimised messaging in the novel HPC hardware.
- Implement an efficient load balancing strategy for simulations where computational load significantly varies during operations, such as dynamic adaptive mesh refinement/coarsening simulations.