Researchers benefit from Computational Science work
Section highlights research, software development, and collaboration . . .
![]() CSS
|
by Brian Bevirt SCD's Computational Science Section (CSS) specializes in computational science, applied mathematics, numerical analysis, and software engineering. CSS staff focus their expertise in three areas beneficial to the atmospheric and oceanic sciences research community: research, software development, and direct collaborations with researchers.
Computer technology evolves rapidly, and CSS works to develop accurate and efficient routines that work on new computer architectures, as well as to improve the routines that run on existing systems. CSS shares the results of their work with users through presentations at conferences and papers published in research journals. Users of NCAR computers gain benefits through more accurate simulations, more effective use of computing resources, and the freedom to focus more on their scientific inquiry rather than computing requirements.
The software libraries Spherepack, Mudpack, and Regridpack combine the numerical and computational expertise of CSS staff with the results of their research to improve computer modeling. CSS staff optimize the routines in these libraries for efficient operation on multiple computing platforms. Again, researchers benefit from this work every time they simply make a subroutine call to a library routine rather than having to write all the code themselves. CSS section manager Steve Hammond says, "Scientific progress would be extremely slow if researchers had to develop everything themselves. If you can build on what others have done or use computer codes that others have written, then you can focus on furthering your own work."
CSS contributions to this project include completely reworking the platform-specific message-passing code in the model and revising the intramodel communication mechanisms. In addition, CSS has done significant performance optimization on the code. These changes achieve the two goals of portability (platform independence) and high performance. The PCM model now runs efficiently on the Cray T3E, the SGI Origin 2000, and the HP Exemplar. The PCM has three component models -- one to simulate the atmosphere and land surface, one for oceans, and one for polar ice caps -- plus a flux coupler to handle the communication of intermediate results between the component models. The project goal was to develop an efficient model that can run on any of the production computing platforms likely to be available to NCAR researchers in the near future. This goal was achieved in 1998 with a high-performance parallel/distributed coupled climate model that runs on any computer architecture complying with the de-facto message-passing standard MPI (Message Passing Interface). A summary of CSS work on the PCM project is published in SCD's Annual Scientific Report. |