The group utilises and develops several in–house and open source CFD packages, actively developing new numerical methods. A particular focus has been the development of very high order accurate schemes which can be applied to compressible flows at all speeds in industrial geometries with multiple species and be easily parallelisable.
A visualisation from a simulation of a compressible turbulent mixing layer initialised from at broadband perturbation is shown on the left. This computation used 3 billion points, and ran on 1024 cores for a week.
Massively Parallel Simulations
All USyd CFD codes are MPI parallelised and run efficiently on the local cluster ‘Borg’, the Univerity cluster ‘Artemis’ and the national facilities ‘Orange’, ‘Raijin’ and ‘Magnus’. The group has a number of projects with substantial CPU hours granted on the last three mentioned Australian national computing facilities. This enable researchers within our group to routines run computations on 512-1000s cores at once.
The scaling of the USyd algorithm FLAMENCO up to 4096 cores on the EU HPC Hermit is shown in the image on the right