[an error occurred while processing this directive] [an error occurred while processing this directive]

Minutes of the CSM Advisory Board Meeting

30 November 1999

NSF, Room 390


Attendees: Ed Sarachik (Chair), Dick Berk, Francis Bretherton, Michael Ghil, Isaac Held, Bert Semtner, Max Suarez, Kevin Trenberth, Steve Zebiak, Rick Anthes, Bob Serafin, Jay Fein, Maurice Blackmon, Al Kellie, Anjuli Bamzai, Joe Friday, Alexandra Isern, Carter Ford, Steve Colman, Jewel Prendeville, Pam Stephens, Anne-Marie Schmoltner, Cliff Jacobs, Bernard Grant, Eric Saltzman, Mike Ledbetter, Steve Meacham, Dick Greenfield, Mark Eakin, Dave Bader, Keith Crank, and Lydia Harper

GEO Advisory Committee Members Attending: Susan Avery, Chuck Koelbel, et al.

1. Sarachik welcomed everyone to the CSM Advisory Board (CAB) meeting.

2. Serafin discussed the NCAR Code Assessment Panel Report. Earlier this year, NSF convened a panel to assess NCAR's numerical model codes in the context of their efficiencies for running on parallel architecture computers. The Panel visited NCAR and was briefed about its modeling codes, e.g., CSM, MM5, WRF, etc. The Panel concluded that many of NCAR's codes have not been designed to take advantage of emerging highly parallel computer architectures. The Panel recommended that NCAR take a leadership position in large-scale simulation, so that the atmospheric science community will be able to make effective use of the very powerful, highly parallel, tera-scale architectures that will be available. Serafin reported that NCAR has committed to producing a Strategic Plan for Computing by spring of 2000 and is making good progress. In parallel, a computer hardware team will continue to evaluate hardware options. Serafin also reported that NCAR has signed an MOU with other NSF supercomputing centers to collaborate on joint evaluation of hardware, outreach and training, and visualization issues.

During the discussion of the Code Assessment Report, it was noted that this was a community-wide issue and that NCAR should view this as an opportunity to advance our capacity to conduct atmospheric sciences modeling. Serafin strongly agreed with both points.

3. Kellie updated the Board on the status of the NCAR computing systems from a hardware install perspective, software conversions, and lessons learned. The GEO Advisory Committee also joined the CAB during Kellie's presentation, so that they would have up-to-date information on NCAR's computing systems.

Kellie reported that NCAR is underway to position itself as a tera-scale supercomputing center with emphasis on model development, production, and portability. Other dimensions of the facilities will be of the scale that the NCAR Mass Store System will be of order peta-scale; that NCAR will support data analysis and visualization supercomputing; that NCAR will collaborate with other NSF centers; and NCAR will have world class support professionals on staff.

The target for sustained performance in 1999 is 50 gigaflops; the out-years target is 1000 gigaflops sustained in 2004. NCAR's strategy will be to (a) significantly increase computing capacity and capability to meet immediate needs; (b) begin aggressive transition to clustered SMPs; (c) retain maximum flexibility in the evolution of platforms; (d) select platforms that minimize the risk to success for NCAR science and leverage the most mature effort in the industry today for conversion and enhanced support; (e) train NCAR and university scientists in the use of new technologies; (f) use funding in a way that all the above can be accomplished; and (g) work collaboratively with other NSF centers.

Goals for code conversion are (a) to work with programmers to modify their codes and teach the user community how to develop future codes; (b) transition knowledge from the experts of the Computational Science Section (CSS) to the user community; and (c) manage the transition, so that the scientists that utilize the facilities at NCAR can continue to do their science and have the opportunity to expeditiously develop implementations of their codes for the IBM.

4. Blackmon reported on a software engineering meeting that was held at NCAR in November, 1999, to discuss collaborative efforts for a common software engineering framework for distributed development for all participants. This will reduce duplication of development efforts at different locations. A CSM Software Engineering Working Group was recommended to be formed, and this issue will be discussed at the next CSM Scientific Steering Committee (SSC) meeting in January, 2000.

5. Zebiak reported on the history and next steps for a Common Model Infrastructure Initiative. At a NSF/NCEP Workshop, it was decided that the "individual" approach was limiting the community and that a facilitating framework, modeling standards, governing principles, and standard experiments and benchmarks were needed. It was decided that the focus should be on a few core level efforts for weather prediction, seasonal-to-interannual climate, and multi-decadal and global change climate. An ad hoc group was formed as a result of this workshop.

The First Infrastructure Workshop was held in October, 1998, in Tucson. The discussion at the workshop focused on the need for column physics coding, common support tools, dynamical core compatibility, model/data repository, flux coupler design, and a data format survey.

The Second Infrastructure Workshop was held in January, 1999, in Dallas. At this workshop the coding rules were revised for column physics; infrastructure needs were decided for a code repository; a convection parameterization repository experiment was designed; a dynamical core experiment was designed; a discussion was held on standard support tool needs; it was decided the focus would be on netCDF (HDF, GRIB) as the data format standard; GDT and NCAR/CDM were examined for metadata standards; a software repository for common tools was considered; and access to boundary condition and evaluation standards were discussed.

The Third Infrastructure Workshop was held in April, 1999, at NASA's Goddard Space Flight Center. A demonstration of the use of the software repository was successful; a website on standards was available at http://janus.gsfc.nasa.gov/~suarez/ infra; a strawman on data standards would be written and distributed widely in the community for comments; GRIB2 would be examined for meta standards; and a possible NASA/HPCC initiative was discussed that would support software development toward a common framework.

The latest Infrastructure Workshop was held in August, 1999, in Monterey. At that workshop the repository was demonstrated; an expanded repository (all major components of earth system and climate models) was explored; it was decided that the single column model would be used; several approaches to the flux coupler problems were discussed; data standards would continue to be developed; a BAMS article would be written on the Common Model Infrastructure Working Group activities; and to continue these efforts, a joint proposal to U.S. agencies would need to be written for developing a comprehensive repository. The joint proposal would also address the need for funding of 3 to 5 new people, a management structure, documentation, outreach, code maintenance, test designs, and a single column model tool. The next meeting will be at NCAR on 22 and 23 February 2000.

6. Blackmon reported that the Parallel Climate Model (PCM) and the CSM would converge into one model. A new CSM Climate Change Assessment Working Group will be recommended to be established at the next CSM SSC meeting in January, 2000. The co-chairs of that working group will be Warren Washington, Jerry Meehl, and Ben Santer.

7. Blackmon discussed the updated Draft CSM Plan. The CAB recommended that scientific goals be added showing what specific questions will be answered by the experiments proposed in the plan; what the scientific priorities are for each model component regarding development (how model development will make the CSM a better model); tie the model development to the scientific questions being answered; and what the CSM will be able to do at the end of 5 years. They also recommended an Executive Summary be written that would be more general and could be used for other purposes, such as briefing OMB, OSTP, other U.S. agencies, etc. Also, it was felt that an Eos and/or a BAMS article should be written so that the correct information on CSM is available.

Fein said that the updated CSM Plan would be good for internal working group scientists to see their roles in the context of the total plan and for the community-at-large to see what the CSM is doing and planning.

It was noted that seasonal-to-interannual (SI) issues were not discussed in the updated CSM Plan. The SI Working Group (SIWG) previously pointed out that CSM is inadequate for their needs. The changes suggested by the SIWG have not been addressed, so this group still cannot use the CSM. Blackmon reported that the SIWG would probably be incorporated into the Decadal-to-Centennial Variability Working Group after discussion at the next CSM SSC meeting in January, 2000.

8. The CSL Proposal for CSM was distributed. The CAB recommended that coupled runs and anthropogenic climate change experiments be added. Blackmon stated that they would be added. The CSL proposal deadline is 18 January 2000. Missing CSM working group input to the CSL proposal will be added, and the entire proposal will be discussed at the next CSM SSC meeting in January, 2000.

9. Blackmon reminded the CAB that the Fifth Annual CSM Workshop will be held from 27 to 29 June 2000, and the next CAB meeting will be on 30 June 2000 in Breckenridge.