Parallelisation of CFD methods for multi-physics problems

The focus of this paper is on the parallel computing aspects of multi-physics CFD methods and the implications for the design of a CFD framework encompassing mesh-based and particle-based approaches. In the first part of the paper, the shared-memory parallelisation of a multi-block, structured CFD is described and the parallel performance of this method is assessed for multi-block meshes for realistic complex configurations. It was found that for the coarse-grained approach taken, a limited parallel scaling was obtained, indicating that a hybrid shared-memory/message-passing approach will be of particular interest. The second part of the paper describes the design and parallelisation of an object-oriented molecular dynamics method, which will form a particle-based component of the multi-physics CFD framework. It is shown that for the object-oriented implementation of this method, the shared-memory paradigm leads to an excellent parallel performance