[an error occurred while processing this directive] [an error occurred while processing this directive]
The working group met for approximately three hours on June 25, 2002 in Breckenridge. Ricky Rood opened the meeting with a brief introduction. The next hour was spent in a topical discussion of "Trends in Computer Hardware". Updates were then presented on CCSM software engineering group activities, CCSM Scidac, and ESMF. An opportunity for discussion closed the meeting.
John Dennis, from NCAR/SCD, gave a presentation to the working group titled, "Trends in Computer Hardware". A number of topics were discussed including the current state of supercomputers, the future of supercomputers, and whether codes can be written flexibly enough to take advantage of different hardware. Some of the important points were that there will continue to be 3 flavors of supercomputers; custom (ie. NEC), commodity (ie. IBM), and cluster (ie. Linux Clusters). Vector processors will continue to exist, at least at NEC, for several years. The custom machines will continue to outperform the other styles on a processor versus processor basis, but will likely continue to perform similarly on a cost basis. Commodity hardware is currently plagued problems in system balance where the processor performance is relatively greater that the memory access or interconnect performance. Processor improvement continues to follow Amdahls law while the memory access and interconnect are improving slower, so unbalanced systems will continue to be a problem on non-custom hardware. Finally, a consensus developed that it was very difficult to write software that was flexible enough to take advantage of both vector and cache based systems.
Tony Craig presented a short update on software engineering at NCAR. The main focus in that talk was a summary of the CCSM2.0 release, and plans for a CCSM2.0.1 and CCSM2.1 release. A brief update on the status of the software engineering coordination plan was presented. In addition, there is a request from the SSC to put together an update of a CCSM five year plan. A preliminary outline of some of the software engineering issues was passed out and discussions will continue.
John Drake presented an update on the CCSM Scidac project. Progress is being made on a number of fronts including component model development, the addition of chemistry and biogeochemistry, and performance. CCSM is now using chunking in the atmosphere model and this results in a 20% performance improvement in CAM. Other model developments that are well underway include the reimplementation of pop resulting in a 25% speedup, an incremental remapping scheme in the ice model which speeds up the advection portion of the model by a factor of 3 and speeds up the total model by 30%, parallelization of the runoff model in clm2, and the development of cpl6 which has a 2 to 3 times improvement in both communication and mapping compared to cpl5.
Cecelia Deluca presented an update on ESMF. The project is now fully staffed, and they are completing their first milestone. The requirements process is nearly complete. They had an excellent community meeting in Washington D.C. in May. They are finalizing the EVA suite which is their internal benchmark suite, and the first software delivery milestone is Spring, 2003.
A brief opportunity to have an open discussion was available at the end of the session. No topics were brought up.