Parallel Computing Issues in Mesoscale Air Quality Modeling
Saturday, March 18, 2000 - 2:00pm - 2:55pm
Giulio Giunta (Istituto Universitario Navale)
In our still ongoing project aimed at developing an operational air quality parallel software system for regulatory applications in southern Italy we dealt with basic numerical, computational and implementation issues in order to introduce parallelism in an efficient way. Parallelization is a driving constraint in each step needed to transform the mathematical formulation of the problem into a real production code. We briefly outline some strategies for designing an efficient parallel mesoscale model as well as suitable techniques and tools for balancing the workload, performing I/O, implementing the code and assessing its performance on distributed-memory parallel computers, including low- cost parallel systems of beowulf-class.