overview of the scidac project: collaborative design and development of the ccsm for terascale...
TRANSCRIPT
Overview of the SciDAC Project: Collaborative Design and
Development of the CCSM for Terascale Computers
PI: Malone(LANL), Drake(ORNL)Co-I (DOE): Ding(LBL), Duffy(LLNL), Erickson(ORNL), Foster(ANL), Ghan(PNL), Jacob(ANL), Jones(LANL), Larson(ANL), Mirin(LLNL), Rotman(LLNL), Taylor(ANL), Worley(ORNL)Co-I (NCAR&NASA): Bettge(NCAR), Kiehl(NCAR), Craig(NCAR), Deluca(NCAR), Lin(DAO), Washington(NCAR), Williamson(NCAR)
Goals for SciDAC CCSM Collaborative Development
• Comprehensive treatment of physical and chemical processes– High resolution ocean and atmosphere support– Hybrid vertical coordinate in ocean code – Tropospheric chemistry package– Biogeochemistry
• Modular “packages” with well defined interfaces and testing procedures– Atm, Coupler (Avant Garde), Ice, Ocn, Lnd
• Performance optimized yet portable and adaptable for utilization of emerging architectural features of terascale computers
• Ready for DOE, NSF and NASA applications– High resolution historical and climate change scenario studies– Carbon cycle studies and data assimilation for climate
Climate Science Enabled
Science 13April 2001: “Detection of Anthropogenic Climate Change in the Worlds Oceans,” Barnett, Pierce, Schnur
Method: Ensemble simulations of the DOE Parallel Climate Model (PCM)
Results: • Detection of Anthropogenic Climate Change in the Worlds Oceans• Ensembles establish 95% confidence intervals of model predictions• Simulated ocean heat storage matches historical record of rising ocean
temperatures
Enabling Technology: • Parallel Climate Model
developed in collaborative effort lead by Warren Washington (NCAR)
• Terascale computing resources
Firsts: • Ensemble study with US model and computers • Coupled model reproducing ocean response• Establishing new level of US model quality
Software Infrastructure and Applied Math Challenges
Challenge Existing Solutions Enhanced Solutions
Project management, Version control and testing infrastructure for competing, distributed development teams
CVS with monitoring procedures BitKeeper? SourceForge? Variety of systems and institutions.
Evolving Three layer architecture: Library(eg. FFT), Utility(eg. orbital calendar, data transpose) and Model ( eg. radiation physics)
F90 modules, vendor math libraries, customized data movement, PILGRM, MCT, …Open design process
Optimized math libraries, performance monitors, CCA, NASA Earth System Modeling Framework(ESMF)
Accurate, fast dynamical methods: atm, ocn, ice
Spectral horizontal, finite volume, conservative semi-Lagrangian advection, hybrid vertical coordinates, fast Helmholtz solvers, two-time level monotone advection
Scalable algorithms, grid technologies, nonlinear solvers, new discretization techniques, subcycling explicit barotropic modes, new formulations
Incorporating new model components and improving coupled model climate simulation
Control simulations, component working group activities, scientific steering and review of non-linear interactions
Available cycles, improved analysis capabilities and extensive data handling capabilities, collaborative technologies
Unknown source ISIC and external Internal to project
Software Tasks• Develop comprehensive design documents for each
component of the model• Implement performance-portable, run-time configurable
CCSM on target machines• Incorporate emerging programming paradigms and
software design practices• Develop testing and validation procedures for all
component models• Incorporate new dynamical components for ocean,
atmosphere and sea ice.• Incorporate new modular physical and chemical process
models
Task List Excerpt and Schedule
Date Component Milestone Deliverable Lead Lab
02Q1 Ocean Reviewed requirements document Report LANL
02Q1 Land Complete requirements document Report NCAR
02Q2 Ocean Hybrid programming model completed in POP Released code LANL
02Q2 OceanComplete blending of orthobaric surfaces with z-levels in HYPOP
Progress report LANL
02Q2 Coupler Load balancing in MCT New release of MCT ANL
02Q3 Atmosphere Demonstrate tuned and optimized model at T85 Validated code ORNL
02Q3 Atmosphere Implement 3D block decomposition in all dycores Benchmarked code ORNL
02Q3 Sea ice Complete requirements document Report LANL
02Q3 Atmosphere Subgrid topography scheme applied to dycores Progress report PNNL
02Q4Atmospheric chemistry
Complete development of chemistry solver Benchmarked code LLNL
03Q1 Atmosphere Separate dynamics and transport New release of dynamical core ORNL
03Q1 Atmosphere Demonstrate T31 performance optimization Validated code ANL
03Q1 Land Complete cache-friendly decomposition Benchmarked code ORNL
03Q1 Sea ice Use automatic differentiation to tune parameters in CICEImproved values of model parameters
ANL
03Q2 Ocean MLP and dynamic load balancing in POP Benchmark report, released code LANL
03Q2 Sea iceComplete hybrid programming model in CICE with subblocked decomposition and load balancing
Benchmarked code LANL
03Q4Atmospheric chemistry
Test GHG distributions with tropospheric chemistry forcing Progress report LLNL
041Q OceanFinish tracer validation tests in HYPOP (ready for biogeochemistry)
LANL
042Q Coupler New data types and structures New realease of MCT ANL
04Q3 All Parallel IO Release of IO library LBNL
05Q1 Ocean Complete validation of HYPOP in coupled model tests Validated code LANL
05Q1Atmospheric chemistry
Test GHG distributions with stratosphereic chemistry forcing added
Progress report LLNL
05Q3 CouplerIntegration of coupler with CCSM utility and machine-specific layers
New release of MCT ANL
06Q2Atmospheric chemistry
Complete addition of aerosol physics Validated code NCAR
• Chemical solver technologies
• GHG with transport
• Interactive ozone : sulfur cycle and aerosols
• Upper ocean dimethyl sulfide
• Load balancing
• Off-line chemical simulations (DAM)
• Support C4MIP
• Multiple resolution atm model– T31, T42, T85 and T170 – Sub-grid orography
precip scheme• Scalable dycores
– 2d, 3d blocks, MLP tests
– Optimized and load balancing physics
• Design documents
• Chunking for cache performance
• Load balanced, m to n transfers
• Land Surface Modeling Toolkit as part of utility layer
Next Generation Couplers
• MCT version 2– Higher level abstractions, component model registry– Scalability to thousands of processors– Dynamic load balancing– 3d fields
• Interoperation of climate, weather and data assimilation functions
• MCT as CCA compliant prototype for Earth System Modeling Framework
SciDAC CCSM – ISIC Collaboration Goals
• Improve performance characterizations• Accelerate development of mathematical software
for climate modeling and analysis• Explore novel methods and advance the theory
and simulation of geophysical flows• Improve software practices for the scientific
endeavor of climate change prediction• Enhance analysis and data handling methods
ISIC Projects
• Performance Evaluation(D. Bailey) 02Q1– PCTM, CCSM Instrumentation - P. Worley (ORNL)
• Earth System Grid (Ian Foster) 02Q3– Data archive and analysis grid - Bernholdt(ORNL)
• Data Access (Arie Shoshani) 02Q3– Improved efficiencies and monitoring - Samatova
(ORNL)
ISIC Projects
• Grids (Glimm, Brown) 03Q1– Ocean nested grids – Smith (LANL), Brown(LLNL)
– Smooth grid transformations – Drake(ORNL), Khamayseh(ORNL)
– Spectral element refinement strategies – Fournier (NCAR, U.Md), Fisher(ANL),Taylor&Wingate(LANL)
– Adaptive models – Joyce Penner (U. Mich)
ISIC Projects
• PDE methods (Phil Collela) 03Q1– Adaptive diagnostics (Drake-Williamson)– Static adapted western boundary currents Malone(LANL)
• Solvers (David Keyes) 03Q4– Baroclinic model (a new discretization) – Krylov methods – Balou Nadiga (LANL)
• Common Component Architecture (Rob Armstrong) 04Q1– Model Coupler Toolkit – Larson (ANL)– CCA compliant CCSM
SciDAC Climate ProjectsProject Title Institutions Involved Principal InvesigatorPredictive Understanding of the Oceans' Wind-Driven Circulation on Interdecadal Time Scales Univ of Indiana Michael Ghil, UCLA
Decadal Variability in the Coupled Ocean-Atmoshere Systems Paola Cessi, UCSDA Geodesic Climate Model with Quasi-Lagrangian Vertical Coordinates LANL, Clarkson University David Randall, Colorado State University
Improving the Processes of Land-Atmoshpere Interactions in CCSM 2.0 at High Resolution University of Arizona Robert Dickinson, Georgia Tech Research Corp.Continuous Dynamic Grid Adaptation in a Global Atmoshperic Model William Gutow ski, Iow a State university
Collaborative Design and Development of the Community Climate System Model for Terascale Computers
ORNL. LBNL, PNNL, ANL, LLNL, National center for Atmospheric Research Robert Malone, LANL
Decadal Regional Climate Studies and Applications with Variable-Resoulation GCMs Using Advanced Numerical Techniques Michael Fox-Rabinovitz, University of Maryland
Multi-Resolution Climate Modeling Ferdinand Baer, University of Maryland
Development of an Atmospheric Climate Model with Self-Adapting Grid and Physics Joyce Penner, University of Michigan
Testing a New Hybrid ocean Circulation Model Based on POP LANL Kirk Bryan, Princeton University
Decadal Regional Climate Studies and Applications with Variable-Resoulation GCMs Using Advanced Numerical Techniques Jean Cote, University of Quebec
Modeling Dynamic Vegetation for Decadal to Multi-Century Climate Change Studies Goddard Institute for Space Studies Andrew Friend, Rutgers University
Towards the Prediction of Decadal to Multi-Century Processes in a High-Throughput Climate System Model
Zhengyu Liu, University of WisconsinJohn Kutzbach, University of Wisconsin
Modeling and Analysis of Global and Regional Hydrologic Processes and Appropriate Conservation of Moist Entropy Donald Johnson, University of Wisconsin
Global Climate Research