Grantee Research Project Results
1998 Progress Report: Parallelization Agent for Legacy Codes in Environmental Modeling
EPA Grant Number: R825206Title: Parallelization Agent for Legacy Codes in Environmental Modeling
Investigators: Kothari, Suraj C. , Gustafson, John L. , Takle, Eugene S.
Institution: Iowa State University
EPA Project Officer: Aja, Hayley
Project Period: January 1, 1997 through December 31, 1999
Project Period Covered by this Report: January 1, 1997 through December 31, 1998
Project Amount: $484,382
RFA: High Performance Computing (1996) RFA Text | Recipients Lists
Research Category: Human Health , Aquatic Ecosystems , Environmental Statistics
Objective:
To create a problem solving environment that will provide parallel computing support to address complex ecological problems and demonstrate its capability by using it to parallelize the MM5, RADM, and ISTAM modelsProgress Summary:
Completion of the basic framework for the Parallelization Agent Completion of the first version of the Parallelization Agent (PA) prototype Demonstration of parallelizing MM5 and RADM codes using the PA prototype The key and novel feature of the PA is the class-specific automatic parallelization where high-level knowledge about the specific class is used to arrive at efficient parallel codes. It combines two technologies: the expert system technology and the compiler technology. Expert programmers arrive at efficient parallel codes by taking advantage of the high-level knowledge about the serial code. The PA follows this expert approach, but at the same time it automates many tedious and time-consuming aspects of hand coding. The current system (PA-1) has the capability to parallelize 3-dimensional time marching explicit finite difference codes. The parallelization process has two major phases. First, an interactive diagnostic phase to check if the given serial code conforms to the class identified by the user. Second, a parallelization phase to detect and optimize communication and to do the serial-to-parallel codes transformation. The capabilities of PA fall into three categories: diagnostic, automatic parallelization, and facilitating high-level understanding of serial and parallel codes. The PA system runs on Unix workstations and PCs running under Linux. The input is FORTRAN 77 code and the output is SPMD code for machines such as IBM SP and CRAY T3 series. The PA is installed on a laptop for the demonstration purposes. Significant Accomplishments
Parallelization of MM5 code was completed in two weeks by one person as compared to 3 2 years for the manual parallelization done by a group at Argonne National Laboratory.
Parallelization of RADM was completed in three weeks as compared to 10 months of manual effort. Moreover, the PA system achieved a two-dimensional as opposed to one-dimensional parallelization by manual effort. The two-dimensional parallelization is more scalable.
Demonstrations of the PA system
The PA system has generated considerable interest in the scientific community and we have been invited to give its demonstration at several places. The demonstrations include:
Demonstration to staff of the EPA/NOAA Atmospheric Sciences Modeling Division in Research Triangle Park, NC. The PA was demonstrated to personnel from the Modeling Systems Analysis Branch (Joan Novak, Chief) and the Atmospheric Model Development Branch (Jason Ching, Chief), and the presentation was delivered by live video to EPA's HPCC group in Michigan. (September 9, 1997)
Demonstrated at the Second Workshop of the Project to Intercompare Regional Climate Simulations, a workshop and project on regional climate simulation supported by UNESCO, Iowa State University and the Electric Power Research Institute. (June 6, 1997)
Demonstration at the Center for Development of Advanced Computing (CDAC), Pune, India. (July 2, 1997)
Demonstration at the International Conference on Parallel Processing for Meteorological Models, at the University of Athens, Greece. ( October 16, 1997)
Demonstrated to Jeff Kuehn of the Scientific Computing Division of the National Center for Atmospheric Research. (November 7, 1997)
Future Activities:
Work has been initiated into the interfacing of models. Atmospheric models that link to chemistry, surface hydrology, biospheric processes, and ocean and ice dynamical processes require interchange of state variables and fluxes that are evaluated on different time and spatial scales in the two media at the interface. Mismatches for model initial and boundary conditions can present significant challenges for coupling models and maintaining efficiency of parallelized models.
Applications of modeling interfaces are being pursued in the general area of modeling heterogeneous ecosystems by use of first principles. Both natural and managed (e.g., agricultural and forestry) ecosystems have complex interactions among components that must be understood to maximize plant production.
Enhancement of the PA for supporting cluster-based computing is being pursued. The objective in this case is to provide an affordable parallel solution by making use of a cluster of powerful PCs. The urgency of pursuing sustainable growth policy in developing countries has created intense interest among China and other developing countries in acquiring routine US EPA modeling capabilities. Cluster-based parallel implementations of environmental models offer the possibility that the use of such models can become affordable worldwide for plant siting and pollution forecasts.
Complementary research activity is currently underway by members of the research team under support of Electrical Power Research Institute (EPRI) and UNESCO.
Journal Articles:
No journal articles submitted with this report: View all 6 publications for this projectSupplemental Keywords:
supercomuting, modeling, problem solving environments, legacy codes, parallel algorithms, RFA, Ecosystem Protection/Environmental Exposure & Risk, computing technology, environmental decision making, legacy codes, supercomputing, computer science, parallel numerical solvers, pollution forecasts, information technologyProgress and Final Reports:
Original AbstractThe perspectives, information and conclusions conveyed in research project abstracts, progress reports, final reports, journal abstracts and journal publications convey the viewpoints of the principal investigator and may not represent the views and policies of ORD and EPA. Conclusions drawn by the principal investigators have not been reviewed by the Agency.