Hybrid SWAN for Fast and Efficient Practical Wave Modelling - Part 2

In the Netherlands, for coastal and inland water applications, wave modelling with SWAN on structured computational grids has become a main ingredient. However, computational times are relatively high. Benchmarks showed that the MPI version of SWAN is not that efficient as the OpenMP version within a single node. Therefore, in a previous paper [5] a hybrid version of SWAN was proposed for computations on structured computational grids. It combines the efficiency of the OpenMP version on shared memory with the capability of the MPI version to distribute memory over nodes. In the current paper we extend this approach by an improved implementation, verification of the model performance with a testbed, and extensive benchmarks of its parallel performance. With these benchmarks for important real life applications we show the significance of this hybrid version. We optimize the approach and illustrate the behavior for larger number of nodes. Parallel I/O will be subject of future research.