| The General Curvilinear Coastal Ocean Model is a high-resolution, nonhydrostatic model with a fully 3D curvilinear coordinate system. It is an inherently computationally expensive model that requires processing large amounts of data. As a solution, the PETSc library is chosen to implement an efficient parallel scheme. Through the use of the PETSC DMDA objects, communication management of multi-dimensional arrays on an Arakawa-C staggered grid is simplified. This DMDA domain decomposition strategy allows a 3D mesh to be easily divided among processors for improved performance. This parallel model is tested for correctness and performance on three experiments that model different physical conditions. The most notable is the Seamount-Beam test case which requires a longer period of simulation to begin manifesting physical properties, making it the most time-consuming experiment that this thesis aims to resolve. The parallel model also retains the numerical methods and accuracy as the serial version which gives model developers a consistent way to analyze results in faster time. In addition, the performance gain allows the parallel GCCOM to accommodate larger problem sizes, resulting in more detailed simulations. |
