RLS: A Parallel Runtime System Library for Atmospheric Models with Nesting

Wednesday, March 12, 1997 - 2:30pm - 3:00pm
John Michalakes (Argonne National Laboratory)
Separator Separator Separator Separator
HOME » PROGRAMS/ACTIVITIES » Annual Thematic Program
Annual Thematic Program »Postdoctoral Fellowships »Hot Topics and Special »Public Lectures »New Directions »PI Programs »Math Modeling »Seminars »Be an Organizer »Annual »Hot Topics »PI Summer »PI Conference »Applying to Participate »

Talk Abstract
RSL: A Parallel Runtime System Library for Atmospheric Models with Nesting

John Michalakes, Argonne National Laboratory

Limited area models of the atmosphere for weather forcasting and climate prediction are computationally intensive applications that can benefit the high performance and large memories of scalable distributed memory parallel computers. The models typically employ explicit finite-differencing methods for advection and diffusion and are readily amenable to stencil-library approaches for parallelization. Mesh refinement proveides the ability to focus higher resolution computaion over a subdomain of interest to resolve, for example, the effects of complicated terrain. Parallelization of multiple nested domains requires addtional services form a parallelization library for the conveyance of forcing and feedback data between a parent and nested domain, both decomposed over processors. Finally, the efficient implementaion of these models requires attention to load imbalances.

RSL[1] is a parallel library developed at Argonne National Laboratory that is tailored to regular-grid atmohpheric meodels with nesting. It is similar to effors of a number or othr groups[2][3][5]. It provides high-level stencil and inter-domain communication, irregualr domain decomposition, automatic local/global index translation, distributed I/O, and dynamic load balancing. RSL was used with Fortran90 to parallelize a well-known and widley used regional weither model, the Penn State/NCAR Mesoscale Model (MM5)[1]. Scaling and performance data on the IBM SP2 is presented, along eith data showing the effectiveness of using RSL to implement dynamic load balancin in the model (Figure 1).


[1] G. A. Grell, J. Dudhia, and D.R. Stauffer, A Description of the Fifth-Generation Penn State/NCAR Mesoscale Model (MM5), Tech. Rep. NCAR/TN-398+STR, National Center for Atmospheric Research, Boulder, Colorado, June 1994.

[2] R. Hempel and H. Ritzdorf, The GMD communications library for grid -oriented problems, Tech. Rep. GMD-0589, German National Research Center for Information Technology, 1991.

[3] S. R. Kohn and S. B. Baden, A Parallel Software Infrastructure for Structured Adaptive Mesh Methods, in Proceedings of Supercomputing `95, IEEE Computer Society Press, 1996.

[4] J. Michalakes, RSL: A Parallel Runtime System Library for Regular Grid Finite Difference Models Using Multiple Nests, Tech. Rep. ANL/MCS-TM-197, Mathematics and Computer Science Division, Argonne National Laboratory, Argonne, Illinois, December 1994.

[5] B. Rodriguez, L. Hart, and T. Henderson, A Library for the Portable Parallelization of Operational Weather Forcast Models, in Coming of Age: Proceeding of the Sixth ECMWF Workshop on the Use of Parallel Processors in Meteorology, World Scientific, River Edge, NJ, 1995. pp. 148-161.