Vectorization and parallelization of a numerical scheme for 3D global atmospheric transport-chemistry problems

Atmospheric air quality modeling relies in part on numerical simulation. Required numerical simulations are often hampered by lack of computer capacity and computational speed. This problem is most severe in the field of global modeling where transport and exchange of trace constituents are studied in the whole of the global troposphere/stratosphere. Studies in this field easily lead to computations with millions of unknowns over long time spans. In such cases use of the most advanced computer systems is a prerequisite for making real progress. This report is devoted to a vectorization/parallelization study, on a Cray C90, of an efficient numerical scheme in development for global atmospheric transport-chemistry problems. The scheme employs a particular type of operator splitting. Performance results are presented with respect to (grid) vectorization and parallelization based on autotasking and on a division in subdomains of the globe. For this purpose we use a constructed, three-space dimensional model problem containing advection, vertical turbulent diffusion and chemical reactions. We also study the numerical accuracy/efficiency of the operator splitting scheme for this model problem. The model problem is presented as a benchmark on which other schemes and implementations can be tested. The benchmark problem is available through World Wide Web