Date: Wed, 20 Nov 1996 19:25:33 GMT Server: NCSA/1.4 Content-type: text/html Last-modified: Tue, 17 Sep 1996 13:47:47 GMT Content-length: 5698 GATOR

GATOR

This work is part of an HPCC project funded by NASA to develop a state-of-the-art Earth System Model (ESM) that will be comprised of a coupled atmosphere and ocean system including chemical tracers that are found in, and may be exchanged between, the atmosphere and the oceans. The starting point of the ESM model is the UCLA AGCM, which has been parallelized on different platforms and will be coupled with GATOR, a Gas, Aerosol, Transport, and Radiation Chemistry model, developed by M. Jacobson, O. Toon, R. Turco, and R. Lu. The UCLA ESM model will archive and retrieve model output via the Sequoia database. Coupled systems are an important tool for helping scientists to understand complex phenomena such as El Nino and stratospheric ozone depletion.

A parallel version of GATOR is being developed at Berkeley by Jim Demmel and Sharon Smith . GATOR models atmospheric chemistry in the Los Angeles Basin, and has been used for detailed air pollution studies. Our task is to parallelize GATOR and scale it to the globe.

GATOR includes both gases and aerosols, modeling

gas-phase chemistry;
aqueous-phase chemistry;
radiation transfer;
aerosol nucleation, coagulation, growth and evaporation;
horizontal advection and vertical convection;
dry and wet deposition;
visibility and emissions.

The Parallel Version of GATOR

A primary motivation for this research is to provide atmospheric scientists with the best computational means to further their studies of the earth's future climate. Since the UCLA ESM is still evolving, it is important that the codes also be portable, so our efforts are thus focused on developing portable, parallel code that can be performance tuned for different parallel architectures.

The main challenges in parallelizing GATOR are overcoming problems due to load imbalance and minimizing the communication costs. The figure below shows the differences in computation that can result in load imbalance:

  

In the picture, the z-axis shows time, while the x and y axis show the latitudinal and longitudinal partitioning of the atmosphere into block columns. Each processor owns one set of block columns. The difference in time among the processors illustrates the amount of computation due to solving ODEs that arise from chemical kinetic equations. The largest amount of computation occurs in parts of the globe that are in the summer and daylight, whereas the processors holding atmospheric cells in winter and night require the least amount of computation. Click here to see the ODE solver computation costs for the world.

Our solution to the load balancing problem is to use a block-cyclic layout which will collect different atmospheric cells from the different parts of the globe together in a single processor. The hope is that each processor will have cells requiring different amounts of computation, but not all of the computationally intensive cells at once. The pictures below illustrate a block layout, which does not prevent against load imbalance, and a block-cyclic layout.

  

Performance Modeling and the ESM

Throughout the design of the parallel implementation of GATOR, we used a timing model to help choose between different design alternatives that would benefit GATOR both as a stand-alone system and as part of the UCLA AGCM. The results of this work are reported in:

J. Demmel and S. L. Smith. Parallelizing a global atmospheric chemical tracer model. In Proceedings of the Scalable High Performance Computing Conf., pages 718--725, Knoxville, TN., May 1994.

J. Demmel and S. L. Smith. Performance of a parallel global atmospheric chemical tracer model. Submitted to Supercomputing 95 .

EOSDIS

We are extending our modeling effort to help explore an architecture for the Earth Observing System (EOS). This effort is part of "End-to-End Problems in EOSDIS", a NASA-sponsored multi-year project to investigate alternative data management strategies for the EOS. The project involves researchers at the Berkeley, Los Angeles, San Diego, and Santa Barbara campuses of the University of California.

For some sample views of environmental and climate data, courtesy of Jeff Dozier at UC Santa Barbara, click here.


last updated on December 12, 1995.
ssmith@CS.Berkeley.EDU