In the first dCSE project related to the CABARET code, a hybrid OpenMP / MPI parallel version of the unstructured code was developed which was scalable to more than 1000 cores on HECToR for problems of more than $ 5 \times 10^7$ cells. More information about that work is given in [8]. However, this code requires the use of a Gambit (Fluent) generated unstructured grid and initial pre-processing for the decomposition. It also uses a master I-O model which gathers data from each process for check pointing and visualisation and also broadcasts the initial Gambit (Fluent) generated finite element mesh. The code can handle arbitrary data decompositions of the Gambit unstructured mesh, although it is recommended to use structured ones wherever geometrical arrangements allow.

Phil Ridley 2012-10-01