Parallelization of Air Quality Models on Structured Grids Using MPI

The technological development during the past two centuries has had a significant impact on our environment. One of the greatest concerns is air pollution. Air pollutants affect human health, the biosphere and damage property. Because of these effects air pollution is now monitored and regulated worldwide, for example by the Clean Air Act in the United States. In order to understand air pollution and design effective control strategies, comprehensive computer models are necessary. Air quality models simulate the chemical reactions and regional movement of pollutants in our atmosphere. The complexity of the computations and the resulting high time-to-solution call for the application of parallelization strategies.

We review some approaches taken by other researchers to efficiently parallelize air quality models. Further we implement a MPI-based communication library to parallelize air quality models using uniform structured grids. The routines serve for distribution, gathering and reshuffling of typical arrays used in these models. Portability of the communication library is proved by testing it on two different architectures, a Beowulf cluster using distributed memory and a Sun Enterprise using shared memory. Our domain decomposition approaches accommodate the most general transportation schemes. We implement different data decomposition strategies and compare them with respect to performance. Finally we use the library to parallelize STEM-III, a state-of-the-science air quality model developed at the University of Iowa. Results of the parallelization on a Beowulf cluster are presented in this work.

Click here for a ps version of the report.