The Urban and Regional Multiscale (URM) Airshed Model is a state-of-the art Air
Quality Model developed at Carnegie Mellon University. It is currently being
used to model ozone formation, acid deposition and other forms of air pollution
in regions the size of the Northeastern United States. Air quality models are
essential tools in the study of atmospheric chemistry and in the formulation of
air pollution control policies and regulations. Because of the economic and
environmental importance of these policies, and because of the enormous amount
of computation required to perform such large-scale simulations, the National
Science Foundation has provided an HPCC Grand Challenge grant to fund the
parallelization of this model.

The first version of the Parallel Multiscale Airshed Model was developed under
this grant in 1993 on the Pittsburgh Supercomputing Center's 8-processor
Decstation Supercluster, a dedicated network of Decstation 5000 workstations
that are connected by an FDDI fiber-optic network. Since then, it has been
ported to PSC's 12-processor Alpha Supercluster, 16-processor Cray Y-MP C90, and
512-processor Cray T3D Massively Parallel Processor.  A single run that models
air quality in the Northeastern U.S. over a 24-hour time period takes 16 hours
on a single Decstation 5000, but only 45 minutes on the Alpha Supercluster. A
24-hour run for the Los Angeles region takes only 8 minutes on the Alpha
cluster.

The most recent version of the program can run on the C90 and the T3D at the
same time. The structure of the program is such that it can take advantage of
both architectures, using the C90's powerful vector processors to model the
transport phase of the model (in which chemicals are transported by the motion
of the atmosphere), and the T3D's numerous Alpha microprocessors to model the
chemical reactions above each of thousands of points the earth's surface.

Measurements suggest that this version of the program will be significantly
faster than previous versions. In addition to making multi-day and multi-week
simulations more practical, this development will also make possible new models
that incorporate much more sophisticated chemical interactions. These reactions
are so computationally demanding that modeling them is currently infeasable.
Thus, the parallelization of this code by CMU researchers and its execution on
PSC's powerful supercomputing resources are enabling significant advances to the
speed and accuracy of air quality modeling.

Current development efforts are centered on optimizing performance of the
heterogeneous C90/T3D version, and on supporting an even wider range of
computing platforms in the future.