1. Angèle Simard Canadian Meteorological Center Meteorological Service of Canada MSC Computing Update

2. 2 Canadian Meteorological Centre  National Meteorological Centre for Canada  Provide services to National Defence and Emergency organizations  International WMO commitments ( emergency response, Telecomm, data, etc…)  Supercomputer used for NWP, climate, air quality, etc… ; operations and R&D

3. 3 Outline  Current Status  Supercomputer Procurement  Conversion Experience  Scientific Direction

4. Current Status

5. 5 MSC’s Supercomputing History

6. 6

7. 7 Supercomputers: NEC – 10 node – 80 PE – 640 GB Memory – 2 TB disks – 640 Gflops (peak)

8. 8 Archive Statistics

9. 9 Automated Archiving System  22,000 tape mounts/month  11 Terabytes growth/month  maximum of 214 TB

10. Supercomputer Procurement

11. 11 RFP/Contract Process  Competitive process  Formal RFP released in mid-2002  Bidders need to meet all mandatory requirements to make it to next phase  Bidders were rejected if bids above the funding envelope  Bidders were rated on performance (90%) and price (10%)  IBM came first  Live preliminary benchmark at IBM facility  Contract awarded to IBM in November 2002

12. 12 IBM Committed Sustained Performance in MSC TFlops

13. 13 On-going Support  The systems must achieve a minimum of 99 % availability.  Remedial maintenance 24/7: 30 minute response time between 8 A.M. and 5 P.M. on weekdays. One hour response outside above periods.  Preventive maintenance or engineering changes: maximum of eight (8) hours a month, in blocks of time not exceeding two (2) or four (4) hours per day (subject to certain conditions).  Software support: Provision of emergency and non- emergency assistance.

14. 14

15. 15 Supercomputers: Now and Then… NEC – 10 node – 80 PE – 640 GB Memory – 2 TB disks –640 Gflops (peak) IBM – 112 nodes – 928 PE – 2.19 TB Memory – 15 TB Disks – 4.825 Tflops (peak)

16. Conversion Experience

17. 17 Transition to IBM Prior to conversion:  Developed and tested code on multiple platforms (MPP & vector)  Where possible, avoided proprietary tools  When required, hid proprietary routines under a local API to centralize changes  Following contract award, obtained early access to the IBM (thanks to NCAR)

18. 18 Transition to IBM Application Conversion Plan:  Plan was made to comply to tight schedule  Plan’s key element included a rapid implementation of an operational “core” (to test mechanics)  Phased approach for optimized version of models  End to End testing scheduled to begin mid- September

19. 19 Scalability Rule of Thumb  To obtain required wall-clock timings:  Require 4 times more IBM processors to obtain same performance with NEC SX-6

20. 20 Timing (min) Configuration NECIBM NECIBM GEMDM regional 38 348 MPI 8 MPI x4 OpenMP (48 hour) (32 total) GEMDM Global35 364 MPI 4 MPI x4 OpenMP (240 hour)(16 total) 3D-Var(R1)11 09 4 6 CPU OpenMP (without AMSU-B) 3D-Var (G2)16 12 4 6 CPU OpenMP (with AMSU-B) Preliminary Results

21. SCIENTIFIC DIRECTIONS

22. 22 Global System Now:  Uniform resolution of 100 km (400 X 200 X 28)  3D-Var at T108 on model levels, 6-hr cycle  Use of raw radiances from AMSUA, AMSUB and GOES 2004:  Resolution to 35 km (800 X 600 X 80)  Top at 0.1 hPa (instead of 10 hPa) with additional AMSUA and AMSUB channels  4D-Var assimilation, 6-hr time window with 3 outer loops at full model resolution and inner loops at T108 (cpu equivalent of a 5- day forecast of full resolution model)  new datasets: profilers, MODIS winds, QuikScat 2005+:  Additional datasets (AIRS, MSG, MTSAT, IASI, GIFTS, COSMIC)  Improved data assimilation

23. 23 Regional System Now:  Variable resolution, uniform region at 24 km (354 X 415 X 28)  3D-Var assimilation on model levels at T108, 12-hr spin-up 2004:  Resolution to 15 km in uniform region (576 X 641 X 58)  Inclusion of AMSU-B and GOES data in assimilation cycle  Four model runs a day (instead of two)  New datasets: profilers, MODIS winds, QuikScat 2006:  Limited area model at 10 km resolution (800 X 800 X 60)  LAM 4D-Var data assimilation  Assimilation of Radar data

24. 24 Ensemble Prediction System Now:  16 members global system (300 X 150 X 28)  Forecasts up to 10 days once a day at 00Z  Optimal Interpolation assimilation system, 6-hr cycle,use of derived radiance data (Satems) 2004:  Ensemble Kalman Filter assimilation system, 6-hr cycle, use of raw radiances from AMSUA, AMSUB and GOES  Two forecast runs per day (12Z run added)  Forecasts extended to 15 days (instead of 10)

25. 25...Ensemble Prediction System 2005:  Increased resolution to 100 km (400 X 200 X 58)  Increased members to 32  Additional datasets such as in global deterministic system 2007:  Prototype regional ensemble  10 members LAM (500 X 500 X 58)  No distinct data assimilation; initial and boundary conditions from global EPS

26. 26 Mesoscale System Now:  Variable resolution, uniform region at 10 km (290 X 371 X 35) Two windows; no data assimilation 2004:  Prototype Limited area model at 2.5 km (500 X 500 X 45) over one area 2005:  Five Limited area model windows at 2.5 km (500 X 500 X 45) 2008:  4D data assimilation

27. 27 Coupled Models Today  In R&D: coupled atmosphere, ocean, ice, wave, hydrology, biology & chemistry  In Production: storm surge, wave Future  Global coupled model for both prediction & data assimilation

28. 28 Estuary & Ice Circulation

29. 29 Merchant Ship Routing Puissance Temps …2 jours... Route sud Route nord

30. 30 Challenges  IT Security  Accelerated Storage Needs  Replacement of Front-end & Storage Infrastructure