EMC/HPC/MPC synergy meeting highlights 10/02/2000
This meeting was led by Stephen Jascourt and followed an agenda denoted by the subtitles below. Attendees included Jim Hoke, Steve Tracton, Carl Staton, Geoff DiMego, John Ward, Bill Bua, Hua-Lu Pan, Kevin McCarthy, Ed Danaher, Dave Reynolds, Pete Caplan, Brett McDonald, Zoltan Toth, Steve Schotz and Ed O'Lenic.
1. IBM SP (lengthy discussion about using computing power for earlier product delivery schedules versus long-term improvements in product quality)
a. Current Schedule (presently everything on schedule):
Eta runs 3 times faster and operational suite overall around 2 times faster on same number of processors on Phase II tests. The speed-up comes from running on a faster processor (375 MHz) and with faster bandwidth.
b. "Best" use of computing resources: better models, more model development, earlier delivery times. What are priorities, how decide?
A variety of issues were raised:
It appeared that a consensus was developing, along the lines of previous synergy meeting discussions, of maintaining a constant delivery schedule instead of speeding up delivery early in the lifetime of the computer. This avoids fighting to get the time back to the original schedule as the models "grow into" the computing resources and maintains consistency for the operational forecasters. Failure to win that fight would default to requiring more computing resources on the operational side, making less available for development and testing, or would reduce the ability of the models (and their data assimilation systems) to improve resolution and complexity.
It was agreed that further discussion requires seeing how Louie wants to weigh in on this and to look for him as our (NCEP's) go-between to Headquarters.
Action plan was set to brief Louie on this at the next "projects meeting", scheduled for Friday Oct. 6, and to report back how he wants to handle this and continue discussing at the next synergy meeting.
2. Notes from EMC
a. Mesoscale Modeling Group:
3D-VAR tuning was found to produce not such nice numbers for summer tunings but across-the-board improvements in Eta 22 compared to the Eta 32 for winter tunings. Thus the September 26, 2000 implementation uses the winter tunings. Revised warm season tunings will be prepared for a major model upgrade anticipated for March 2001.
b. Global Modeling Group:
Topography influence on hurricane relocation/reconstruction was fixed. Also changed, if the tropical storm center is located over land then there will now be no relocation performed and no synthetic data used. Over oceans, no change: relocation will continue to be used unless the storm is very weak in the guess field, in which case synthetic data is used (since there's not much in the model fields to relocate).
3. Discussion of making short-range ensembles available earlier
Postponed. Folds into the continuing discussion of computing resource allocation for products versus for development.
4. Miscellaneous - Global parallels
Global branch is preparing for a minor December implementation, including those items other than convection which had been planned to go into the May 2000 implementation. Therefore, effective October 3, the MRF parallels will be changed:
5. Next Meeting Proposed Monday October 30, 2000 at noon in room 209.
Discussion continues on defining requirements for product delivery times in such a way as to meet operational needs, maintain consistency, and avoid eroding development efforts. Meeting should start with report back from meeting with Louie on how he wants to proceed and with news from NCO on any new experience with Phase II providing a better feel for what the machine will be able to handle.