Practical Parallel Computing Strategies, with Application to Air Quality and Cross-Media Models

EPA Grant Number: R825211
Title: Practical Parallel Computing Strategies, with Application to Air Quality and Cross-Media Models
Investigators: Coats, Carlie J.
Institution: North Carolina Supercomputing Center
EPA Project Officer: Saint, Chris
Project Period: October 1, 1996 through September 30, 1999
Project Amount: $738,231
RFA: High Performance Computing (1996) RFA Text |  Recipients Lists
Research Category: Health , Ecosystems , Environmental Statistics

Description:

The investigators propose to investigate practical parallel modeling of environmental issues using the growing computational power made available by networks of workstations and servers with high performance microprocessors. There are two aspects to this investigation: 1) What programming paradigm and what software tools should be used to deal with the complexity within individual environmental models; and 2) What paradigm and tools should be used to couple individual models the way they need to be coupled for cross-media modeling? Both aspects need solutions accessible to environmental modelers, not just high performance computing specialists.

This research is required because turnaround time has become a major obstacle for the use of computer modeling to assess the nature and severity of environmental problems. The conventional solution to turnaround has been the use of supercomputers--vector supercomputers at first, and more lately massively parallel (MPP) supercomputers--in an attempt to achieve the required performance. The effectiveness of this solution has been greatly diminished by two things: 1) the relative scarcity of these systems and the corresponding backlog of users and their jobs (frequently so great that a dedicated desktop machine with sufficient memory and disk delivers more timely results); and 2) the difficulty of the specialized programming required for effective use of these systems, especially the MPPs.

This difficulty is exacerbated by the complexity of environmental models, which is so much greather than that of models in fluid dynamics, astrophysics, or molecular chemistry. As a practical matter, the conventional solution fails. Cross-media modeling, which couples together several models of the kind used today, makes the problem even more severe.

The investigators will investigate the use of data parallel language extensions and software libraries to find and document accessible solutions to these problems, applying what is learned to three test cases, each of which explores a different aspect of the parallel modeling problem: 1) an air quality chemistry-transport model with modular science processes supporting aerosols and toxics, to test intra-program parallelism; 2) an air quality prediction model, to test one-way coupling among a meteorology model, an emissions model, and a chemistry-transport model; and 3) a cross-media hydrology-PBL model, to test two-way coupling between a soil hydrology model and a planetary boundary layer submodel embedded in a mesoscale meteorology model.

In the coupled-model cases, the individual models will be parallel in their own right, allowing us to study the itneractions between the different levels of parallelism. The investigators will target three classes of readily-available, cost-effective platforms--workstations, networks or clusters of workstations, and symmetric multiprocessing (SMP) servers--and models portable across all three. The investigators expect the results to be protable to MPPs as well, although that is not the focus of our efforts.

Supplemental Keywords:

supercomuting, modeling, parallel algorithms, air quality., RFA, Scientific Discipline, Air, Ecosystem Protection/Environmental Exposure & Risk, particulate matter, Environmental Chemistry, computing technology, air quality modeling, cross media problem solving, fate and transport, HPCC, fluid dynamics, chemical transport model, computer science, high performance microprocessors, data analysis, information technology, parallel computing, specialized programming, cross-media environmental monitoring