Use of High Performance Computing in Meteorology: Proceedings of the ECMWF Workshop on the Use of High Performance Computing in Meteorology, Reading, UK, 25-29 October 2004

CONTENTS......Page 8
Preface......Page 6
1.1. High performance Computer Systems for Numerical Weather prediction at ECMWF......Page 10
1.2. Sustained performance measurements using the RAPS benchmark......Page 11
2. Overview of IFS......Page 12
2.1. Parallelisation of IFS......Page 13
2.2. Scalability of the IFS forecast model......Page 14
3.1. Scalability of 4D-Var......Page 16
4. Dr.Hook......Page 18
References......Page 21
1. The “Divergence Problem”......Page 22
1.1. Applications Have Diverse Algorithmic Needs......Page 24
2. The Science Driven System Architecture Approach......Page 25
3. Goals of the Science Driven System Architecture Process......Page 27
3.1. An Example – Sparse Matrix Calculations......Page 28
4. Early Results for Science Driven Architecture......Page 31
Acknowledgments......Page 32
References......Page 33
1. Introduction......Page 34
2. Languages for Scientific Parallel Programming......Page 36
3. New Challenges and Requirements......Page 37
4.1. Overview......Page 38
4.2. Domains, Indez Sets, and Arrays......Page 39
4.3.1. User-Defined Distributions: An Example......Page 40
4.3.2. User-Defined Distributions: The General Approach......Page 41
5 . Conclusion......Page 42
References......Page 43
1. Introduction......Page 45
2. Overview of the Earth Simulator System......Page 46
3. MDPS and NQSII......Page 47
3.1. Mass Data Processing System......Page 48
3.2. Network Queuing System II......Page 49
(2) Stage-in......Page 51
4.1. The number and site of requests......Page 52
4.2 Stage-in/Stage-out......Page 54
4.3. Performance results......Page 55
References......Page 57
Non-hydrostatic Atmospheric GCM Development and Its Computational Performance Keiko Takahashi, Xindong Peng, Kenji Komine, Mitsuru Ohdaira, Koji Goto, Masayuki Yamada, Fuchigami Hiromitsu, Takeshi Sugimura......Page 59
1. Introduction......Page 60
2.1. Governing Equations......Page 61
2.3. Discretization and Schemes......Page 62
3.1. Mass conservation......Page 63
3.2. Williamson's test for shallow water equations......Page 64
3.3. 3-dimensional mountain waves and Held-Suarez experiments......Page 66
4.1. Global cloud water distribution with the developed coupled atmosphere-ocean-land simulation code......Page 67
4.2. Typhoon tracking forecast of ETAU......Page 68
5. High Performance Computation on the Earth Simulator......Page 69
6. Near Future Work......Page 70
References......Page 71
1. Introduction......Page 72
2. Error Subspace Kalman Filters......Page 75
3. The Parallel Data Assimilation Framework PDAF......Page 77
3.1. General Consideration......Page 78
3.2. Interface Structure......Page 82
3.3. Parallelization Aspects......Page 83
3.4. Filter algorithms......Page 85
4. Parallel Performance of PDAF and ESKF Algorithms......Page 86
5. Summary......Page 89
References......Page 91
1. Data assimilation in operational weather forecasting......Page 93
2. General assimilation problem......Page 94
3.1. The cost function......Page 95
4. The Basic Linear Kalman Filter......Page 96
5 . The Extended Kalman Filter......Page 97
6. Computationally feasible approximations to Kalman filtering......Page 98
7. Variational Kalman Filter......Page 100
7.1. Kalman fonnulation of VKF......Page 101
7.2. Cost function formulation of VKF during the local 4D-Var step......Page 102
8. Computational results......Page 103
References......Page 106
1. Introduction......Page 109
1.1. Productivity and Price/Performance......Page 110
1.2. HCP System Architectures......Page 111
1.3. HPC Solutions Building Blocks......Page 113
1.4. Software Tools......Page 114
1.6. Summary......Page 115
References......Page 116
1. Introduction......Page 117
2.1. Rationale for a data-centrk approach......Page 118
2.2.1 Homogenous Computer System......Page 119
2.2.3. Distributed HSM Components......Page 121
2.2.4. LAN-Coupling......Page 122
3. System Architecture of the HLRE at DKRZ......Page 123
4. Final HLRE Configuration......Page 125
5.1. The Global File system GFS......Page 127
5.2. The HSM DiskXtenderunitree......Page 128
5.3. The DBMS Oracle......Page 132
6. Observations from HLRE Operations......Page 134
7. Conclusions......Page 137
Acknowledgments......Page 138
References......Page 139
1 Introduction......Page 140
3 Systems Operational Suite of Numerical Weather Analysis and Prediction......Page 141
4 BMRC Benchmarks Performance on SX-6......Page 142
5.1 Pros and Cons of GFS......Page 143
6 Stages in SX-6 Installation......Page 144
7.1 Performance Tuning Results for GenSI ASSIM......Page 145
7.2 Memory Contention in GASP EPS......Page 146
7.3 Intra-node versus Inter-node Scalability......Page 147
8 SX-6 Usage......Page 148
11 Conclusions......Page 150
12 Acknowledgements......Page 151
1. Introduction......Page 152
2. 4D-Var......Page 153
4. Porting Data Assimilation Codes to the SX-6......Page 154
4.2. Functionality Changes......Page 155
5. Optimisation Techniques and Impacts......Page 156
6. Effect of PE Domain Decomposition......Page 160
7. Code Scalability......Page 161
Appendix......Page 163
References......Page 164
1. Introduction......Page 165
2. WRF Advanced Software Framework......Page 167
3. Nesting and Moving Nests......Page 170
4. I/O and Model Coupling......Page 171
5. Performance......Page 174
6. Conclusion......Page 175
References......Page 176
1. Introduction......Page 178
2. Standardizing the Running Interface......Page 179
3. Visualizing the Running Processes by SMS......Page 182
4. Project “National Meteorological Data Access and Retrieval System”......Page 184
Reference......Page 186
1. Introduction......Page 187
2.2. Supercomputer procurement in 2004......Page 188
2.3. Next-generation supercomputer: HITACHI SR11000 model JI and "model Jl followed-on''......Page 189
3.1. Operational suite on the current supercomputer......Page 190
3.2. Operational suite on the next-generation supercomputer......Page 192
3.3, Parallelization......Page 193
Reference......Page 195
1. Introduction......Page 196
2. Grid Computing......Page 198
2.1. The Globus Toolkit......Page 200
2.1.1. Security......Page 201
2.1.4. Data Access......Page 202
2.2.1. Compute Grids......Page 203
2.2.3. Service Grids......Page 204
2.3. Active Grids......Page 205
2.3.3. LEAD......Page 206
3. Deploying Grids at NOAA......Page 207
3.1.1. The Current Network......Page 208
3.2. Compute Grids......Page 210
3.2.2. Verification......Page 212
3.2.3. Compute Grids to Support VOs......Page 213
3.3. Data Grids......Page 214
3.3.1. Leveraging Existing NOAA Programs......Page 217
3.3.3. Potential Data Grids at NOAA......Page 218
3.4. Service Grids......Page 219
4. Conclusion......Page 220
Acknowledgements......Page 221
References......Page 222
1. Introduction......Page 224
2. The NDG architecture......Page 227
2.1. The NDG Enterprise Viewpoint......Page 228
2.2. The NDG Information viewpoint......Page 229
3. Metadata......Page 231
3.1. 'B' metadata......Page 232
3.2. 'D' metadata......Page 234
4. Datamodel......Page 235
5. Security......Page 237
References......Page 240
Task Geometry for Commodity Linux Clusters and Grids: A Solution for Topology-aware Load Balancing of Synchronously Coupled, Asymmetric Atmospheric Models I. Lumb, B. McMillan, M. Page, G. Carr......Page 243
1. Introduction......Page 244
2.1. Description......Page 246
2.2. Performance Enhancement......Page 250
3.1. Description......Page 254
3.2. Performance Enhancement......Page 256
3.3. Related Usage......Page 258
4.1. Job Forwarding......Page 259
4.2. Resource Leasing......Page 260
4.3. A Grid of Grids......Page 261
5 . Discussion......Page 263
References......Page 264
1. Introduction: The Community Climate System Model......Page 268
1.2 CCSM Configurations......Page 269
3.1 Porting Introduction......Page 270
3.2 CAM PERGRO......Page 271
4.1 CCSM Performance......Page 272
4.2 Processor Load Imbalance......Page 273
4.3 CCSM Production......Page 275
5.3 June, 2004......Page 276
6. Remaining Work......Page 277
About the Authors......Page 278
References......Page 279
1. Introduction......Page 281
2.1. Persistence Property......Page 284
2.2. Unison Property......Page 286
3. The Uniform Memory Model......Page 288
4. Memory race conditions in parallel programs......Page 290
5. Distributed arrays......Page 296
5.1. Memory allocation......Page 297
5.2. Memory access......Page 298
6. Summary......Page 300
Acknowledgements......Page 301
References......Page 302
New Languages and New Hardware Features......Page 304
Linux Clusters......Page 305
Frameworks......Page 306
List of Participants......Page 308