This work is part of the exaFOAM project that aims to enable the open-source CFD software OpenFOAM to exploit massively parallel HPC architectures and overcome performance scaling bottlenecks. This benchmark represents an industrial application of the unsteady optimization infrastructure of OpenFOAM, using the Toyota CHR car geometry. When used with unsteady flows, the adjoint equations are integrated backwards in time, requiring the instantaneous flow fields to be available at each time-step of the adjoint solver, which noticeably increases storage requirements in large-scale problems. To avoid extreme treatments, such as the full storage of the computed flow fields or their recomputations from scratch during the solution of the adjoint equations and to reduce the re-evaluation overhead incurred by the widely used check-pointing technique, lossy compression techniques were implemented into the adjointOptimisation library of OpenFOAM, during the exaFOAM project. The re-computation cost is reduced by efficiently compressing the entire compressed flow series and storing it in memory. The current data-set includes the CFD mesh and a snapshot of the DDES flow solution (restart files), obtained after discarding 1 second of the transient part. Detailed information and case setup can be found in the README.md file contained in the case setup file.

exaFOAM, v2206

The case setup in DaRUS is a snapshot (commit 19c20ed19436e844001b8fa24e39ea42f3fbf552) from the code repository of the OpenFOAM High Performance Computing Technical Committee https://develop.openfoam.com/committees/hpc/-/tree/develop/incompressible/adjointOptimisationFoam/TMECar