High Performance Computing on Vector Systems 2009 (eBook)

Preface 5
Contents 7
I Petaflop/s Computing 14
Lessons Learned from 1-Year Experience with SX-9 and Toward the Next Generation Vector Computing 15
Introduction 15
SX-9 System Overview 16
HPC Challenge Benchmark Results 18
Case Study Analysis of Memory-Conscious Tuning for SX-9 25
Multi-Vector Cores Processor Design 30
Summary 33
References 34
BSC-CNS Research and Supercomputing Resources 35
Overview 35
Supercomputing Resources at BSC 36
MareNostrum 36
MareNostrum Performance 2008 38
Shared Memory System 38
Backup and HSM Service 39
Spanish Supercomputing Network 39
PRACE Prototype 40
Research at BSC 42
Challenges and Opportunities of Hybrid Computing Systems 43
Introduction 43
European Context 45
Validation Scenario 46
Initial Results 47
Operational Requirements 49
Conclusions and Future Work 51
References 51
Going Forward with GPU Computing 52
Computing needs at CEA 52
Starting the Process 54
Available Hardware 54
Choosing a Programming Language 57
CUDA 57
OpenCL 58
RapidMind 58
HMPP 58
A Remark on Languages 60
Training Sessions 60
The System Administration Side 60
The Grand Challenges Strategy 61
Foreseen Problems 61
First Results 62
Conclusion 63
Optical Interconnection Technology for the Next Generation Supercomputers 64
Introduction 64
Components and Structure 66
Performance 67
Conclusions 69
References 69
HPC Architecture from Application Perspectives 70
Introduction 70
Trend of CPU Performance 72
Architectural Challenges 74
SIMD-based Approaches 75
Conclusions 77
References 78
II Strategies 79
A Language for Fortran Source to Source Transformation 80
Compiler 80
Self Defined Transformations 81
The Transformation Language 81
Transformation Variables 82
Transformation Constructs 82
Self Defined Procedures in the Transformation Code 83
Intrinsic Procedures 83
Parsing Primitives in Parsing Mode 84
Examples 85
Concluding Remarks 87
The SX-Linux Project: A Progress Report 88
Introduction 89
Project Paths 89
Progress and Status 91
The GNU Toolchain 91
Binutils 91
GCC 92
Current Toolchain Status 93
Future Work 94
User Space and I/O Forwarding 94
Newlib 94
Future of Newlib 95
Virtualization Layer 96
I/O Forwarding 97
I/O Forwarding-Current Implementation 98
I/O Forwarding Library Status 99
Kernel 99
Kitten LWK 100
Implementation and Status 100
Bootstrapping 100
Early Introspection 100
Stack and Memory Layout 101
Interrupts 102
System Calls 102
Context Switch 102
User Space 103
Status 103
Outlook 104
References 105
Development of APIs for Desktop Supercomputing 106
Introduction 106
Client APIs for GDS 108
Client APIs 108
Script Generator API 108
Implementation of Script Generator API in AEGIS 110
Development of GDS Application of Three-dimensional Virtual Plant Vibration Simulator 112
Three-dimensional Virtual Plant Vibration Simulator 112
Development of GDS Application of Three-dimensional Virtual Plant Vibration Simulator 113
Summary 114
References 115
The Grid Middleware on SX and Its Operation for Nation-Wide Service 117
Introduction 117
Structure of NAREGI Grid Middleware 118
Managing Resources by Using Web Services: IS, SS, and GridVM 118
Virtualizing the Computing Resources: GridVM Scheduler and GridVM Engines 119
Features and Issues of NAREGI Grid Middleware 119
Reservation-Type Job Scheduling 120
Virtualization and Overheads 120
Load Concentration on Management Nodes 121
Scheduling of Non-reserved Jobs 121
Maintaining Coherency and Consistency in the Web Services on the Grid 122
Features of NEC's NQS-II/JobManipulator Local Scheduler and Its Use at the Cybermedia Center of Osaka University 122
GridVM for SX 123
Creating a System to Verify the Coherence and Consistency of Web Services 123
Delegation of Reservation Table Administration Authorization by Synchronization of Tables 125
Co-existence with GRAM/MDS Interface 126
Enabling MPI/SX Job Execution 126
Future Issues 126
References 127
III Applications 128
From Static Domains to Graph Decomposition for Heterogeneous Cluster Programming 129
Introduction 129
Epitaxial Surface Growth 130
Introduction to Physical Model 130
Simulation 131
Domain Decomposition 132
Atomic Interaction 134
Results 135
Potts Model Simulations 136
Domain Decomposition 139
Results 141
Graph Domain Decomposition 143
Model 144
Workbalance 147
Domain Decomposition. Grouping Algorithm 148
Programming 149
Results 150
References 152
IV Computational Fluid Dynamics 154
Direct Numerical Simulations of Turbulent Shear Flows 155
Introduction 155
Numerical Method 157
Performance on Distributed Memory Systems 157
Performance on a Vector System 158
The `Virtual Wind Tunnel' 159
Supersonic Axisymmetric Wakes 159
Turbulent Flow over Airfoil Trailing Edges 161
Compressible Mixing Layer 162
Jet Noise 163
Turbulent Spots in Supersonic Boundary Layers 164
Turbulent Breakdown of Vortex Rings 165
Wing Tip Vortex Breakdown and Far Wakes 166
Summary 167
References 168
Large-Scale Flow Computation of Complex Geometries by Building-Cube Method 170
Introduction 170
Building-Cube Method 172
Overview 172
Flow Solver 172
Code Optimization (Vectorization and Parallelization) 173
Vectorization 173
Parallelization 174
Large-Scale Flow Computation 175
Conclusion 180
References 181
A New Parallel SPH Method for 3D Free Surface Flows 182
The SPH Approach 182
The MPI Parallelization with Dynamic Load-Balancing 183
3D Dam Break and Impact Test Problem 185
Mesh-Convergence Test 187
Application to a Realistic Mudflow 187
References 191
V Climate Modeling 192
The Agulhas System as a Prime Example for the Use of Nesting Capabilities in Ocean Modelling 193
Motivation 193
Modelling Environment 195
Scientific Achievements 197
Conclusion 199
References 199
Seamless Simulations in Climate Variability and HPC 201
Introduction 201
Model Description 203
The Atmosphere Component: MSSG-A 203
The Ocean Component: MSSG-O 204
Grid Configuration of MSSG 205
Differencing Schemes 205
Algebraic Multigrid Method in a Poisson Solver 206
Coupling Between MSSG-A and MSSG-O 206
Implementation of MSSG on the Earth Simulator 207
Coding Style 207
Distribution Architecture and Communications 207
Inter-/Intra-node Parallel Architectures and Vector Processing 208
Memory and Cost Reductions for Land Area in MSSG-O 209
Overlapped Computations in the Ocean Component 210
Coupling Scheme with High Computational Performance in MSSG 210
Computational Performance on MSSG on the Earth Simulator 211
Performance and Scalability 211
Cost Balance and Communication Cost 213
Efficiency of Overlapped Computation in the Oceanic Component 214
Simulation Results 215
Global Simulation with MSSG-A 215
Stand Alone Oceanic Component 215
Prediction of Typhoon Tracking with MSSG 216
Conclusions and Perspectives 220
References 220
VI Computational Physics 222
Construction of Vibration Table in an Extended World for Safety Assessment of Nuclear Power Plants 223
Introduction 223
Overview of Seismic Simulation 224
Seismic Simulation of Mechanical Components 226
Governing Equations 226
Balancing Domain Decomposition Method 227
Optimization of Number of Subdomains in Balancing Domain Decomposition Method 229
Computation Cost for Each Iteration 229
Prediction Curve of Total Computation Cost 229
Numerical Validation on a Parallel Computer 230
Concluding Remarks 232
References 232
Understanding Electron Transport in Atomic Nanowires from Large-Scale Numerical Calculations 233
Introduction 233
Computational Method 234
Results 236
Summary 241
References 241
Multi-scale Simulations for Laser Plasma Physics 243
Introduction 244
Numerical Methods 245
Radiation Hydrodynamics Code (PINOCO) 246
Collective PIC Code (FISCOF1D and 2D) 247
Relativistic Fokker-Planck Hydrodynamic Code (FIBMET) 247
Distributed Computing Collaboration Protocol (DCCP) 248
Fully Integrated Simulation of Fast Ignition 249
Summary 250
References 250