A benchmark for the parallel code used in FMS and MOM-4

The numerical performance of the parallel code implemented in the Modular Ocean Model MOM-4 is investigated by numerical experiments. Several model configurations with different horizontal grid size are integrated in parallel with a varying number of MPI-tasks. The measured computation rates can be understood in terms of a generalized Amdahl’s law. Advantages and limits of parallel integration of MOM-4 are discussed. An optimum task number and a maximum task number are given for each model configuration. MOM-4 is found to be suitable for parallel processing on high performance hardware. Distributing the numerical work over up to 512 tasks also projects with a high horizontal grid size of 1024 times 1024 grid points can be integrated with a sufficient computation rate.