High Performance Computing on Vector Systems 2011

1 - High Performance Computing on Vector Systems 2011 [Seite High Performance Computing on Vector Systems 2011]
- 3 [Seite 3]
1.1 - Preface [Seite 5]
1.2 - Contents [Seite 7]
1.3 - Part I: Techniques and Tools for High Performance Systems [Seite 9]
1.3.1 - Performance and Scalability Analysisof a Chip Multi Vector Processor [Seite 10]
1.3.1.1 - 1 Introduction [Seite 11]
1.3.1.2 - 2 Chip Multi Vector Processor [Seite 12]
1.3.1.2.1 - 2.1 Structure of a Chip Multi Vector Processor [Seite 12]
1.3.1.2.2 - 2.2 Performance Model of a Chip Multi Vector Processor [Seite 13]
1.3.1.3 - 3 Performance Tuning for a Chip Multi Vector Processor [Seite 15]
1.3.1.3.1 - 3.1 Performance Analysis Using the Roofline Model [Seite 15]
1.3.1.3.2 - 3.2 Program Optimization [Seite 16]
1.3.1.3.2.1 - 3.2.1 Loop Unrolling [Seite 16]
1.3.1.3.2.2 - 3.2.2 Cache Blocking [Seite 17]
1.3.1.3.2.3 - 3.2.3 Performance Tuning Strategy Based on the Roofline Model [Seite 17]
1.3.1.4 - 4 Performance and Scalability Analysis [Seite 18]
1.3.1.4.1 - 4.1 Methodology [Seite 18]
1.3.1.4.2 - 4.2 Benchmarks [Seite 19]
1.3.1.4.3 - 4.3 Performance Evaluation of CMVP [Seite 20]
1.3.1.4.4 - 4.4 Performance Evaluation of CMVP with Performance Tuning [Seite 22]
1.3.1.5 - 5 Conclusions [Seite 25]
1.3.1.6 - References [Seite 26]
1.3.2 - I/O Forwarding for Quiet Clusters [Seite 28]
1.3.2.1 - 1 Introduction [Seite 29]
1.3.2.2 - 2 Operating System Noise [Seite 30]
1.3.2.2.1 - 2.1 So .Who's the Noisy Neighbour? [Seite 31]
1.3.2.2.2 - 2.2 Impact on Applications [Seite 31]
1.3.2.2.3 - 2.3 Mitigation [Seite 32]
1.3.2.2.3.1 - 2.3.1 Silence Your System [Seite 32]
1.3.2.2.3.2 - 2.3.2 Embrace Noise [Seite 33]
1.3.2.2.3.3 - 2.3.3 Synchronize Noise [Seite 33]
1.3.2.2.3.4 - 2.3.4 Prioritize [Seite 33]
1.3.2.2.3.5 - 2.3.5 Travel Light [Seite 33]
1.3.2.3 - 3 Measuring Noise [Seite 34]
1.3.2.3.1 - 3.1 Test System [Seite 34]
1.3.2.3.2 - 3.2 Fixed Work Quanta Benchmark [Seite 35]
1.3.2.3.3 - 3.3 Fixed Time Quanta Benchmark [Seite 36]
1.3.2.4 - 4 I/O Induced Noise [Seite 36]
1.3.2.5 - 5 I/O Forwarding [Seite 38]
1.3.2.5.1 - 5.1 I/O Forwarding Architecture [Seite 39]
1.3.2.5.2 - 5.2 System I/O Interceptors: Libsysio [Seite 40]
1.3.2.5.3 - 5.3 I/O Forwarding Protocol: IOD Driver and Server [Seite 41]
1.3.2.5.4 - 5.4 Communication Framework: Portals [Seite 41]
1.3.2.5.5 - 5.5 Using the I/O Forwarding Framework [Seite 42]
1.3.2.5.6 - 5.6 Noise [Seite 42]
1.3.2.5.7 - 5.7 FUSE Driver [Seite 44]
1.3.2.6 - 6 Conclusion [Seite 44]
1.3.2.7 - References [Seite 45]
1.3.3 - A Prototype Implementation of OpenCL for SX Vector Systems [Seite 47]
1.3.3.1 - 1 Introduction [Seite 48]
1.3.3.2 - 2 OpenCL [Seite 48]
1.3.3.3 - 3 OpenCL for SX [Seite 49]
1.3.3.4 - 4 Early Evaluation and Discussions [Seite 51]
1.3.3.5 - 5 Conclusions [Seite 53]
1.3.3.6 - References [Seite 55]
1.3.4 - Distributed Parallelization of Semantic Web Java Applications by Means of the Message-Passing Interface [Seite 57]
1.3.4.1 - 1 Introduction [Seite 57]
1.3.4.2 - 2 Use Case Description: Random Indexing [Seite 59]
1.3.4.3 - 3 Parallelization Strategy [Seite 60]
1.3.4.4 - 4 Realization by Means of MPI [Seite 61]
1.3.4.5 - 5 Implementation [Seite 63]
1.3.4.6 - 6 Application Performance Evaluation [Seite 64]
1.3.4.7 - 7 Performance Tailoring: Hybrid MPI-Java Threads Communication Pattern [Seite 66]
1.3.4.8 - 8 Final Discussion and Conclusion [Seite 68]
1.3.4.9 - References [Seite 69]
1.3.5 - HPC Systems at JAIST and Development of Dynamic Loop Monitoring Tools Toward Runtime Parallelization [Seite 71]
1.3.5.1 - 1 Introduction [Seite 71]
1.3.5.2 - 2 Information Environment and HPC Systems at JAIST [Seite 72]
1.3.5.3 - 3 Development of Dynamic Loop Monitoring Tools Toward Runtime Parallelization [Seite 74]
1.3.5.3.1 - 3.1 Background and Objectives of Dynamic Loop Monitoring Tools [Seite 75]
1.3.5.3.2 - 3.2 Parallelism and Loop Nest Structures [Seite 75]
1.3.5.3.3 - 3.3 Loop Nest Detection and Loop-Call Context Tree Generation [Seite 76]
1.3.5.3.4 - 3.4 Evaluation of Our L-CCT Generation [Seite 78]
1.3.5.3.4.1 - 3.4.1 Experiment [Seite 78]
1.3.5.3.4.2 - 3.4.2 Results [Seite 78]
1.3.5.3.5 - 3.5 Run-Time Data Dependence Analysis [Seite 80]
1.3.5.3.5.1 - 3.5.1 Motivations and Strategies [Seite 81]
1.3.5.3.5.2 - 3.5.2 Details of Our Runtime Data Dependence Analysis [Seite 81]
1.3.5.3.5.3 - 3.5.3 Preliminary Evaluation of Runtime Data Dependence Analysis [Seite 82]
1.3.5.4 - 4 Conclusions [Seite 83]
1.3.5.5 - References [Seite 83]
1.4 - Part II: Methods and Technologies for Large-Scale Systems [Seite 85]
1.4.1 - Tree Based Voxelization of STL Data [Seite 86]
1.4.1.1 - 1 Introduction [Seite 86]
1.4.1.2 - 2 Octree Overview [Seite 88]
1.4.1.3 - 3 Mesh Generation [Seite 89]
1.4.1.3.1 - 3.1 Intersection Algorithm and Tree Generation [Seite 90]
1.4.1.3.2 - 3.2 Flooding [Seite 92]
1.4.1.3.3 - 3.3 Boundary Conditions [Seite 92]
1.4.1.3.4 - 3.4 The File Format [Seite 94]
1.4.1.4 - 4 Sample Mesh [Seite 95]
1.4.1.5 - 5 Outlook [Seite 96]
1.4.1.6 - References [Seite 96]
1.4.2 - An Adaptable Simulation Framework Based on a Linearized Octree [Seite 98]
1.4.2.1 - 1 Introduction and Overall Layout of the Apes Framework [Seite 98]
1.4.2.1.1 - 1.1 Used Technologies [Seite 99]
1.4.2.1.2 - 1.2 Components of the Apes Suite [Seite 99]
1.4.2.1.3 - 1.3 Distributed Computing [Seite 101]
1.4.2.2 - 2 Related Work [Seite 101]
1.4.2.3 - 3 The Distributed Linearized Octree [Seite 102]
1.4.2.3.1 - 3.1 Implementation of the Element Description [Seite 102]
1.4.2.3.2 - 3.2 Element Properties [Seite 104]
1.4.2.3.3 - 3.3 Acting on the Tree [Seite 106]
1.4.2.4 - 4 Configuration of Simulation Runs [Seite 107]
1.4.2.5 - 5 Usage in Solvers [Seite 107]
1.4.2.5.1 - 5.1 Ateles [Seite 108]
1.4.2.5.2 - 5.2 Musubi [Seite 109]
1.4.2.6 - 6 Outlook [Seite 110]
1.4.2.7 - References [Seite 110]
1.4.3 - High Performance Computing for Analyzing PB-Scale Data in Nuclear Experiments and Simulations [Seite 111]
1.4.3.1 - 1 Introduction [Seite 111]
1.4.3.2 - 2 Large-Scale Data Integrated Analysis System [Seite 112]
1.4.3.3 - 3 Heterogeneous Processors for Acceleration Large-Data Analyses [Seite 113]
1.4.3.4 - 4 Distributed Parallel Computing Framework with Fault-Tolerance [Seite 116]
1.4.3.5 - 5 Summary [Seite 120]
1.4.3.6 - References [Seite 121]
1.5 - Part III: Computational Fluid Dynamics, Physical Simulation and Engineering Application [Seite 122]
1.5.1 - TASCOM3D: A Scientific Code for Compressible Reactive Flows [Seite 123]
1.5.1.1 - 1 Introduction [Seite 123]
1.5.1.2 - 2 Governing Equations and Numerical Schemes [Seite 124]
1.5.1.3 - 3 Numerical Investigations of NOx-Formation in Scramjet Combustors Using Wall and Strut Injectors [Seite 125]
1.5.1.3.1 - 3.1 Configuration and Numerical Setup [Seite 126]
1.5.1.3.2 - 3.2 Results [Seite 128]
1.5.1.3.3 - 3.3 Conclusion and Further Reading [Seite 131]
1.5.1.4 - 4 Steady and Unsteady RANS Simulations of a Cryogenic Rocket Combustor [Seite 131]
1.5.1.4.1 - 4.1 Configuration and Numerical Setup [Seite 131]
1.5.1.4.2 - 4.2 Results [Seite 133]
1.5.1.4.3 - 4.3 Conclusion and Further Reading [Seite 136]
1.5.1.5 - 5 Performance Analysis [Seite 136]
1.5.1.5.1 - 5.1 Single CPU Performance [Seite 137]
1.5.1.5.2 - 5.2 Scaling Performance [Seite 138]
1.5.1.5.3 - 5.3 Conclusion and Further Reading [Seite 139]
1.5.1.6 - 6 Conclusion [Seite 140]
1.5.1.7 - References [Seite 141]
1.5.2 - Investigations of Human Nasal Cavity Flows Based on a Lattice-Boltzmann Method [Seite 144]
1.5.2.1 - 1 Introduction [Seite 144]
1.5.2.2 - 2 Numerical Methods [Seite 146]
1.5.2.2.1 - 2.1 The Lattice-Boltzmann Method with Local Grid Refinement [Seite 146]
1.5.2.2.2 - 2.2 Computational Grid [Seite 149]
1.5.2.3 - 3 Scalability and Performance Analysis [Seite 150]
1.5.2.4 - 4 Nasal Cavity Flows [Seite 152]
1.5.2.5 - 5 Discussion [Seite 156]
1.5.2.6 - 6 Conflict of Interest [Seite 157]
1.5.2.7 - References [Seite 158]
1.5.3 - Influence of Adatoms on the Quantum Conductance and Metal-Insulator Transition of Atomic-Scale Nanowires [Seite 160]
1.5.3.1 - 1 Introduction [Seite 160]
1.5.3.2 - 2 Computational Method [Seite 161]
1.5.3.3 - 3 Results [Seite 162]
1.5.3.4 - References [Seite 170]
1.5.4 - Current Status and Future Direction of Full-Scale Vibration Simulator for Entire Nuclear Power Plants [Seite 172]
1.5.4.1 - 1 Introduction [Seite 172]
1.5.4.2 - 2 Three-Dimensional Vibration Simulator for an Entire Nuclear Power Plant [Seite 174]
1.5.4.2.1 - 2.1 Methodology of Assembled-Structure Analysis [Seite 174]
1.5.4.2.2 - 2.2 Computation Platform for Large-Scale Simulation of an Entire Nuclear Plant [Seite 175]
1.5.4.3 - 3 Current Status of Vibration Simulator [Seite 175]
1.5.4.3.1 - 3.1 Development of Elastic Analysis of High Temperature Engineering Test Reactor [Seite 175]
1.5.4.3.2 - 3.2 Development of a Feasible Design for the New Concept Tubesheet Structure in Fast Breeder Reactors [Seite 177]
1.5.4.4 - 4 Future Direction of Vibration Simulator [Seite 177]
1.5.4.4.1 - 4.1 Development of Algorithm in Numerical Calculation [Seite 177]
1.5.4.4.2 - 4.2 Response Estimation Method for Elasto-Plastic Analysis [Seite 179]
1.5.4.4.3 - 4.3 Analysis Capability for Seismic Fluid Phenomena [Seite 179]
1.5.4.4.3.1 - 4.3.1 Installation of Open Source CFD Software on BX900 [Seite 179]
1.5.4.4.3.2 - 4.3.2 Development of Characteristic Simulation of Two-Phase Flow Turbulence [Seite 180]
1.5.4.5 - 5 Conclusion [Seite 183]
1.5.4.6 - References [Seite 184]