Benefit of MPI-IO in Cfoam-CABARET

The addition of MPI-IO to Cfoam-CABARET will be essential for scalable IO on HPC systems when using large numbers of MPI processes. For less than 1000 MPI processes with a $2.5 \times 10^7$ cell test case, the MPI-IO proved 20% slower than the original method. However, if many post-processing files are required then MPI-IO is required in order to prevent too many files being accessed at run-time. Furthermore, for more than 1000 processes MPI-IO performance is at least as good as the original approach and more importantly, it eliminates the problems with having many thousands of files present.