Parallel processing and non-uniform grids in global air quality modeling

A large-scale global air quality model, running efficiently on a single vector processor, is enhanced to make more realistic and more long-term simulations feasible. Two strategies are combined: non-uniform grids and parallel processing. The communication through the hierarchy of non-uniform grids interferes with the inter-processor communication. We discuss load balance in the decomposition of the domain, I/O, and inter-processor communication. A model shows that the communication overhead for both techniques is very low, whence non-uniform grids allow for large speed-ups and high speed-up can be expected from parallelization. The implementation is in progress, and results of experiments will be reported elsewhere.