High-Performance Computing on Complex Environments
Published on 10. April 2014
512 pages
978-1-118-71207-8 (ISBN)
Available for download
Description
With recent changes in multicore and general-purpose computingon graphics processing units, the way parallel computers are usedand programmed has drastically changed. It is important to providea comprehensive study on how to use such machines written byspecialists of the domain. The book provides recent researchresults in high-performance computing on complex environments,information on how to efficiently exploit heterogeneous andhierarchical architectures and distributed systems, detailedstudies on the impact of applying heterogeneous computing practicesto real problems, and applications varying from remote sensing totomography. The content spans topics such as Numerical Analysis forHeterogeneous and Multicore Systems; Optimization of Communicationfor High Performance Heterogeneous and Hierarchical Platforms;Efficient Exploitation of Heterogeneous Architectures, HybridCPU+GPU, and Distributed Systems; Energy Awareness inHigh-Performance Computing; and Applications of HeterogeneousHigh-Performance Computing.
* Covers cutting-edge research in HPC on complexenvironments, following an international collaboration of membersof the ComplexHPC
* Explains how to efficiently exploit heterogeneous andhierarchical architectures and distributed systems
* Twenty-three chapters and over 100 illustrations coverdomains such as numerical analysis, communication and storage,applications, GPUs and accelerators, and energy efficiency
More details
Other editions
Additional editions
Emmanuel Jeannot | Julius Zilinskas
High-Performance Computing on Complex Environments
Book
07/2014
Wiley
€123.50
Shipment within 15-20 days
Persons
Emmanuel Jeannot is a Senior Research Scientist at INRIA.He received his PhD in computer science from Ecole NormaleSuperieur de Lyon. His main research interests are processesplacement, scheduling for heterogeneous environments and grids,data redistribution, algorithms and models for parallelmachines.
Julius ?ilinskas is a Principal Researcher and a Head ofDepartment at Vilnius University in Vilnius, Lithuania. Hisresearch interests include parallel computing, optimization, dataanalysis and visualization.
Content
1 - Cover [Seite 1]
2 - Title Page [Seite 5]
3 - Contents [Seite 9]
4 - Contributors [Seite 25]
5 - Preface [Seite 29]
6 - European Science Foundation [Seite 31]
7 - Part I Introduction [Seite 33]
7.1 - Chapter 1 Summary of the Open European Network for High-Performance Computing in Complex Environments [Seite 35]
7.1.1 - 1.1 Introduction and Vision [Seite 36]
7.1.2 - 1.2 Scientific Organization [Seite 38]
7.1.2.1 - 1.2.1 Scientific Focus [Seite 38]
7.1.2.2 - 1.2.2 Working Groups [Seite 38]
7.1.3 - 1.3 Activities of the Project [Seite 38]
7.1.3.1 - 1.3.1 Spring Schools [Seite 38]
7.1.3.2 - 1.3.2 International Workshops [Seite 39]
7.1.3.3 - 1.3.3 Working Groups Meetings [Seite 39]
7.1.3.4 - 1.3.4 Management Committee Meetings [Seite 39]
7.1.3.5 - 1.3.5 Short-Term Scientific Missions [Seite 39]
7.1.4 - 1.4 Main Outcomes of the Action [Seite 39]
7.1.5 - 1.5 Contents of the Book [Seite 40]
7.1.6 - Acknowledgment [Seite 42]
8 - Part II Numerical Analysis for Heterogeneous and Multicore Systems [Seite 43]
8.1 - Chapter 2 On the Impact of the Heterogeneous Multicore and Many-Core Platforms on Iterative Solution Methods and Preconditioning Techniques [Seite 45]
8.1.1 - 2.1 Introduction [Seite 46]
8.1.2 - 2.2 General Description of Iterative Methods and Preconditioning [Seite 48]
8.1.2.1 - 2.2.1 Basic Iterative Methods [Seite 48]
8.1.2.2 - 2.2.2 Projection Methods: CG and GMRES [Seite 50]
8.1.3 - 2.3 Preconditioning Techniques [Seite 52]
8.1.4 - 2.4 Defect-Correction Technique [Seite 53]
8.1.5 - 2.5 Multigrid Method [Seite 54]
8.1.6 - 2.6 Parallelization of Iterative Methods [Seite 54]
8.1.7 - 2.7 Heterogeneous Systems [Seite 55]
8.1.7.1 - 2.7.1 Heterogeneous Computing [Seite 56]
8.1.7.2 - 2.7.2 Algorithm Characteristics and Resource Utilization [Seite 57]
8.1.7.3 - 2.7.3 Exposing Parallelism [Seite 58]
8.1.7.4 - 2.7.4 Heterogeneity in Matrix Computation [Seite 58]
8.1.7.5 - 2.7.5 Setup of Heterogeneous Iterative Solvers [Seite 59]
8.1.8 - 2.8 Maintenance and Portability [Seite 61]
8.1.9 - 2.9 Conclusion [Seite 62]
8.1.10 - Acknowledgments [Seite 63]
8.1.11 - References [Seite 63]
8.2 - Chapter 3 Efficient Numerical Solution of 2D Diffusion Equation on Multicore Computers [Seite 65]
8.2.1 - 3.1 Introduction [Seite 66]
8.2.2 - 3.2 Test Case [Seite 67]
8.2.2.1 - 3.2.1 Governing Equations [Seite 67]
8.2.2.2 - 3.2.2 Solution Procedure [Seite 68]
8.2.3 - 3.3 Parallel Implementation [Seite 71]
8.2.3.1 - 3.3.1 Intel PCM Library [Seite 71]
8.2.3.2 - 3.3.2 OpenMP [Seite 72]
8.2.4 - 3.4 Results [Seite 73]
8.2.4.1 - 3.4.1 Results of Numerical Integration [Seite 73]
8.2.4.2 - 3.4.2 Parallel Efficiency [Seite 74]
8.2.5 - 3.5 Discussion [Seite 77]
8.2.6 - 3.6 Conclusion [Seite 79]
8.2.7 - Acknowledgment [Seite 79]
8.2.8 - References [Seite 79]
8.3 - Chapter 4 Parallel Algorithms for Parabolic Problems on Graphs in Neuroscience [Seite 83]
8.3.1 - 4.1 Introduction [Seite 83]
8.3.2 - 4.2 Formulation of the Discrete Model [Seite 85]
8.3.2.1 - 4.2.1 The theta-Implicit Discrete Scheme [Seite 87]
8.3.2.2 - 4.2.2 The Predictor--Corrector Algorithm I [Seite 89]
8.3.2.3 - 4.2.3 The Predictor--Corrector Algorithm II [Seite 90]
8.3.3 - 4.3 Parallel Algorithms [Seite 91]
8.3.3.1 - 4.3.1 Parallel theta-Implicit Algorithm [Seite 91]
8.3.3.2 - 4.3.2 Parallel Predictor--Corrector Algorithm I [Seite 94]
8.3.3.3 - 4.3.3 Parallel Predictor--Corrector Algorithm II [Seite 95]
8.3.4 - 4.4 Computational Results [Seite 95]
8.3.4.1 - 4.4.1 Experimental Comparison of Predictor--Corrector Algorithms [Seite 98]
8.3.4.2 - 4.4.2 Numerical Experiment of Neuron Excitation [Seite 100]
8.3.5 - 4.5 Conclusions [Seite 101]
8.3.6 - Acknowledgments [Seite 102]
8.3.7 - References [Seite 102]
9 - Part III Communication and Storage Considerations in High-Performance Computing [Seite 105]
9.1 - Chapter 5 An Overview of Topology Mapping Algorithms and Techniques in High-Performance Computing [Seite 107]
9.1.1 - 5.1 Introduction [Seite 108]
9.1.2 - 5.2 General Overview [Seite 108]
9.1.2.1 - 5.2.1 A Key to Scalability: Data Locality [Seite 109]
9.1.2.2 - 5.2.2 Data Locality Management in Parallel Programming Models [Seite 109]
9.1.2.3 - 5.2.3 Virtual Topology: Definition and Characteristics [Seite 110]
9.1.2.4 - 5.2.4 Understanding the Hardware [Seite 111]
9.1.3 - 5.3 Formalization of the Problem [Seite 111]
9.1.4 - 5.4 Algorithmic Strategies for Topology Mapping [Seite 113]
9.1.4.1 - 5.4.1 Greedy Algorithm Variants [Seite 113]
9.1.4.2 - 5.4.2 Graph Partitioning [Seite 114]
9.1.4.3 - 5.4.3 Schemes Based on Graph Similarity [Seite 114]
9.1.4.4 - 5.4.4 Schemes Based on Subgraph Isomorphism [Seite 114]
9.1.5 - 5.5 Mapping Enforcement Techniques [Seite 114]
9.1.5.1 - 5.5.1 Resource Binding [Seite 115]
9.1.5.2 - 5.5.2 Rank Reordering [Seite 115]
9.1.5.3 - 5.5.3 Other Techniques [Seite 116]
9.1.6 - 5.6 Survey of Solutions [Seite 117]
9.1.6.1 - 5.6.1 Algorithmic Solutions [Seite 117]
9.1.6.2 - 5.6.2 Existing Implementations [Seite 117]
9.1.7 - 5.7 Conclusion and Open Problems [Seite 121]
9.1.8 - Acknowledgment [Seite 122]
9.1.9 - References [Seite 122]
9.2 - Chapter 6 Optimization of Collective Communication for Heterogeneous HPC Platforms [Seite 127]
9.2.1 - 6.1 Introduction [Seite 127]
9.2.2 - 6.2 Overview of Optimized Collectives and Topology-Aware Collectives [Seite 129]
9.2.3 - 6.3 Optimizations of Collectives on Homogeneous Clusters [Seite 130]
9.2.4 - 6.4 Heterogeneous Networks [Seite 131]
9.2.4.1 - 6.4.1 Comparison to Homogeneous Clusters [Seite 131]
9.2.5 - 6.5 Topology- and Performance-Aware Collectives [Seite 132]
9.2.6 - 6.6 Topology as Input [Seite 133]
9.2.7 - 6.7 Performance as Input [Seite 134]
9.2.7.1 - 6.7.1 Homogeneous Performance Models [Seite 135]
9.2.7.2 - 6.7.2 Heterogeneous Performance Models [Seite 137]
9.2.7.3 - 6.7.3 Estimation of Parameters of Heterogeneous Performance Models [Seite 138]
9.2.7.4 - 6.7.4 Other Performance Models [Seite 138]
9.2.8 - 6.8 Non-MPI Collective Algorithms for Heterogeneous Networks [Seite 138]
9.2.8.1 - 6.8.1 Optimal Solutions with Multiple Spanning Trees [Seite 139]
9.2.8.2 - 6.8.2 Adaptive Algorithms for Efficient Large-Message Transfer [Seite 139]
9.2.8.3 - 6.8.3 Network Models Inspired by BitTorrent [Seite 140]
9.2.9 - 6.9 Conclusion [Seite 143]
9.2.10 - Acknowledgments [Seite 143]
9.2.11 - References [Seite 143]
9.3 - Chapter 7 Effective Data Access Patterns on Massively Parallel Processors [Seite 147]
9.3.1 - 7.1 Introduction [Seite 147]
9.3.2 - 7.2 Architectural Details [Seite 148]
9.3.3 - 7.3 K-Model [Seite 149]
9.3.3.1 - 7.3.1 The Architecture [Seite 149]
9.3.3.2 - 7.3.2 Cost and Complexity Evaluation [Seite 150]
9.3.3.3 - 7.3.3 Efficiency Evaluation [Seite 151]
9.3.4 - 7.4 Parallel Prefix Sum [Seite 152]
9.3.4.1 - 7.4.1 Experiments [Seite 157]
9.3.5 - 7.5 Bitonic Sorting Networks [Seite 158]
9.3.5.1 - 7.5.1 Experiments [Seite 163]
9.3.6 - 7.6 Final Remarks [Seite 164]
9.3.7 - Acknowledgments [Seite 165]
9.3.8 - References [Seite 165]
9.4 - Chapter 8 Scalable Storage I/O Software for Blue Gene Architectures [Seite 167]
9.4.1 - 8.1 Introduction [Seite 167]
9.4.2 - 8.2 Blue Gene System Overview [Seite 168]
9.4.2.1 - 8.2.1 Blue Gene Architecture [Seite 168]
9.4.2.2 - 8.2.2 Operating System Architecture [Seite 168]
9.4.3 - 8.3 Design and Implementation [Seite 170]
9.4.3.1 - 8.3.1 The Client Module [Seite 171]
9.4.3.2 - 8.3.2 The I/O Module [Seite 173]
9.4.4 - 8.4 Conclusions and Future Work [Seite 174]
9.4.5 - Acknowledgments [Seite 174]
9.4.6 - References [Seite 174]
10 - Part IV Efficient Exploitation of Heterogeneous Architectures [Seite 177]
10.1 - Chapter 9 Fair Resource Sharing for Dynamic Scheduling of Workflows on Heterogeneous Systems [Seite 179]
10.1.1 - 9.1 Introduction [Seite 180]
10.1.1.1 - 9.1.1 Application Model [Seite 180]
10.1.1.2 - 9.1.2 System Model [Seite 183]
10.1.1.3 - 9.1.3 Performance Metrics [Seite 184]
10.1.2 - 9.2 Concurrent Workflow Scheduling [Seite 185]
10.1.2.1 - 9.2.1 Offline Scheduling of Concurrent Workflows [Seite 186]
10.1.2.2 - 9.2.2 Online Scheduling of Concurrent Workflows [Seite 187]
10.1.3 - 9.3 Experimental Results and Discussion [Seite 192]
10.1.3.1 - 9.3.1 DAG Structure [Seite 192]
10.1.3.2 - 9.3.2 Simulated Platforms [Seite 192]
10.1.3.3 - 9.3.3 Results and Discussion [Seite 194]
10.1.4 - 9.4 Conclusions [Seite 197]
10.1.5 - Acknowledgments [Seite 198]
10.1.6 - References [Seite 198]
10.2 - Chapter 10 Systematic Mapping of Reed--Solomon Erasure Codes on Heterogeneous Multicore Architectures [Seite 201]
10.2.1 - 10.1 Introduction [Seite 201]
10.2.2 - 10.2 Related Works [Seite 203]
10.2.3 - 10.3 Reed--Solomon Codes and Linear Algebra Algorithms [Seite 204]
10.2.4 - 10.4 Mapping Reed--Solomon Codes on Cell/B.E. Architecture [Seite 205]
10.2.4.1 - 10.4.1 Cell/B.E. Architecture [Seite 205]
10.2.4.2 - 10.4.2 Basic Assumptions for Mapping [Seite 206]
10.2.4.3 - 10.4.3 Vectorization Algorithm and Increasing its Efficiency [Seite 207]
10.2.4.4 - 10.4.4 Performance Results [Seite 209]
10.2.5 - 10.5 Mapping Reed--Solomon Codes on Multicore GPU Architectures [Seite 210]
10.2.5.1 - 10.5.1 Parallelization of Reed--Solomon Codes on GPU Architectures [Seite 210]
10.2.5.2 - 10.5.2 Organization of GPU Threads [Seite 212]
10.2.6 - 10.6 Methods of Increasing the Algorithm Performance on GPUs [Seite 213]
10.2.6.1 - 10.6.1 Basic Modifications [Seite 213]
10.2.6.2 - 10.6.2 Stream Processing [Seite 214]
10.2.6.3 - 10.6.3 Using Shared Memory [Seite 216]
10.2.7 - 10.7 GPU Performance Evaluation [Seite 217]
10.2.7.1 - 10.7.1 Experimental Results [Seite 217]
10.2.7.2 - 10.7.2 Performance Analysis using the Roofline Model [Seite 219]
10.2.8 - 10.8 Conclusions and Future Works [Seite 222]
10.2.9 - Acknowledgments [Seite 223]
10.2.10 - References [Seite 223]
10.3 - Chapter 11 Heterogeneous Parallel Computing Platforms and Tools for Compute-Intensive Algorithms: A Case Study [Seite 225]
10.3.1 - 11.1 Introduction [Seite 226]
10.3.2 - 11.2 A Low-Cost Heterogeneous Computing Environment [Seite 228]
10.3.2.1 - 11.2.1 Adopted Computing Environment [Seite 231]
10.3.3 - 11.3 First Case Study: The N-Body Problem [Seite 232]
10.3.3.1 - 11.3.1 The Sequential N-Body Algorithm [Seite 233]
10.3.3.2 - 11.3.2 The Parallel N-Body Algorithm for Multicore Architectures [Seite 235]
10.3.3.3 - 11.3.3 The Parallel N-Body Algorithm for CUDA Architectures [Seite 236]
10.3.4 - 11.4 Second Case Study: The Convolution Algorithm [Seite 238]
10.3.4.1 - 11.4.1 The Sequential Convolver Algorithm [Seite 238]
10.3.4.2 - 11.4.2 The Parallel Convolver Algorithm for Multicore Architectures [Seite 239]
10.3.4.3 - 11.4.3 The Parallel Convolver Algorithm for GPU Architectures [Seite 240]
10.3.5 - 11.5 Conclusions [Seite 243]
10.3.6 - Acknowledgments [Seite 244]
10.3.7 - References [Seite 244]
10.4 - Chapter 12 Efficient Application of Hybrid Parallelism in Electromagnetism Problems [Seite 247]
10.4.1 - 12.1 Introduction [Seite 247]
10.4.2 - 12.2 Computation of Green's functions in Hybrid Systems [Seite 248]
10.4.2.1 - 12.2.1 Computation in a Heterogeneous Cluster [Seite 249]
10.4.2.2 - 12.2.2 Experiments [Seite 250]
10.4.3 - 12.3 Parallelization in Numa Systems of a Volume Integral Equation Technique [Seite 254]
10.4.3.1 - 12.3.1 Experiments [Seite 254]
10.4.4 - 12.4 Autotuning Parallel Codes [Seite 258]
10.4.4.1 - 12.4.1 Empirical Autotuning [Seite 259]
10.4.4.2 - 12.4.2 Modeling the Linear Algebra Routines [Seite 261]
10.4.5 - 12.5 Conclusions and Future Research [Seite 262]
10.4.6 - Acknowledgments [Seite 263]
10.4.7 - References [Seite 264]
11 - Part V CPU + GPU Coprocessing [Seite 267]
11.1 - Chapter 13 Design and Optimization of Scientific Applications for Highly Heterogeneous and Hierarchical HPC Platforms Using Functional Computation Performance Models [Seite 269]
11.1.1 - 13.1 Introduction [Seite 270]
11.1.2 - 13.2 Related Work [Seite 273]
11.1.3 - 13.3 Data Partitioning Based on Functional Performance Model [Seite 275]
11.1.4 - 13.4 Example Application: Heterogeneous Parallel Matrix Multiplication [Seite 277]
11.1.5 - 13.5 Performance Measurement on CPUs/GPUs System [Seite 279]
11.1.6 - 13.6 Functional Performance Models of Multiple Cores and GPUs [Seite 280]
11.1.7 - 13.7 FPM-Based Data Partitioning on CPUs/GPUs System [Seite 282]
11.1.8 - 13.8 Efficient Building of Functional Performance Models [Seite 283]
11.1.9 - 13.9 FPM-Based Data Partitioning on Hierarchical Platforms [Seite 285]
11.1.10 - 13.10 Conclusion [Seite 289]
11.1.11 - Acknowledgments [Seite 291]
11.1.12 - References [Seite 291]
11.2 - Chapter 14 Efficient Multilevel Load Balancing on Heterogeneous CPU + GPU Systems [Seite 293]
11.2.1 - 14.1 Introduction: Heterogeneous CPU + GPU Systems [Seite 294]
11.2.1.1 - 14.1.1 Open Problems and Specific Contributions [Seite 295]
11.2.2 - 14.2 Background and Related Work [Seite 297]
11.2.2.1 - 14.2.1 Divisible Load Scheduling in Distributed CPU-Only Systems [Seite 297]
11.2.2.2 - 14.2.2 Scheduling in Multicore CPU and Multi-GPU Environments [Seite 300]
11.2.3 - 14.3 Load Balancing Algorithms for Heterogeneous CPU + GPU Systems [Seite 301]
11.2.3.1 - 14.3.1 Multilevel Simultaneous Load Balancing Algorithm [Seite 302]
11.2.3.2 - 14.3.2 Algorithm for Multi-Installment Processing with Multidistributions [Seite 305]
11.2.4 - 14.4 Experimental Results [Seite 307]
11.2.4.1 - 14.4.1 MSLBA Evaluation: Dense Matrix Multiplication Case Study [Seite 307]
11.2.4.2 - 14.4.2 AMPMD Evaluation: 2D FFT Case Study [Seite 309]
11.2.5 - 14.5 Conclusions [Seite 311]
11.2.6 - Acknowledgments [Seite 312]
11.2.7 - References [Seite 312]
11.3 - Chapter 15 The All-Pair Shortest-Path Problem in Shared-Memory Heterogeneous Systems [Seite 315]
11.3.1 - 15.1 Introduction [Seite 315]
11.3.2 - 15.2 Algorithmic Overview [Seite 317]
11.3.2.1 - 15.2.1 Graph Theory Notation [Seite 317]
11.3.2.2 - 15.2.2 Dijkstra's Algorithm [Seite 318]
11.3.2.3 - 15.2.3 Parallel Version of Dijkstra's Algorithm [Seite 319]
11.3.3 - 15.3 CUDA Overview [Seite 319]
11.3.4 - 15.4 Heterogeneous Systems and Load Balancing [Seite 320]
11.3.5 - 15.5 Parallel Solutions to The APSP [Seite 321]
11.3.5.1 - 15.5.1 GPU Implementation [Seite 321]
11.3.5.2 - 15.5.2 Heterogeneous Implementation [Seite 322]
11.3.6 - 15.6 Experimental Setup [Seite 323]
11.3.6.1 - 15.6.1 Methodology [Seite 323]
11.3.6.2 - 15.6.2 Target Architectures [Seite 324]
11.3.6.3 - 15.6.3 Input Set Characteristics [Seite 324]
11.3.6.4 - 15.6.4 Load-Balancing Techniques Evaluated [Seite 324]
11.3.7 - 15.7 Experimental Results [Seite 325]
11.3.7.1 - 15.7.1 Complete APSP [Seite 325]
11.3.7.2 - 15.7.2 512-Source-Node-to-All Shortest Path [Seite 327]
11.3.7.3 - 15.7.3 Experimental Conclusions [Seite 328]
11.3.8 - 15.8 Conclusions [Seite 329]
11.3.9 - Acknowledgments [Seite 329]
11.3.10 - References [Seite 329]
12 - Part VI Efficient Exploitation of Distributed Systems [Seite 333]
12.1 - Chapter 16 Resource Management for HPC on the Cloud [Seite 335]
12.1.1 - 16.1 Introduction [Seite 335]
12.1.2 - 16.2 On the Type of Applications for HPC and HPC2 [Seite 337]
12.1.3 - 16.3 HPC on the Cloud [Seite 338]
12.1.3.1 - 16.3.1 General PaaS Solutions [Seite 338]
12.1.3.2 - 16.3.2 On-Demand Platforms for HPC [Seite 342]
12.1.4 - 16.4 Scheduling Algorithms for HPC2 [Seite 343]
12.1.5 - 16.5 Toward an Autonomous Scheduling Framework [Seite 344]
12.1.5.1 - 16.5.1 Autonomous Framework for RMS [Seite 345]
12.1.5.2 - 16.5.2 Self-Management [Seite 347]
12.1.5.3 - 16.5.3 Use Cases [Seite 349]
12.1.6 - 16.6 Conclusions [Seite 351]
12.1.7 - Acknowledgment [Seite 352]
12.1.8 - References [Seite 352]
12.2 - Chapter 17 Resource Discovery in Large-Scale Grid Systems [Seite 355]
12.2.1 - 17.1 Introduction and Background [Seite 355]
12.2.1.1 - 17.1.1 Introduction [Seite 355]
12.2.1.2 - 17.1.2 Resource Discovery in Grids [Seite 356]
12.2.1.3 - 17.1.3 Background [Seite 357]
12.2.2 - 17.2 The Semantic Communities Approach [Seite 357]
12.2.2.1 - 17.2.1 Grid Resource Discovery Using Semantic Communities [Seite 357]
12.2.2.2 - 17.2.2 Grid Resource Discovery Based on Semantically Linked Virtual Organizations [Seite 359]
12.2.3 - 17.3 The P2P Approach [Seite 361]
12.2.3.1 - 17.3.1 On Fully Decentralized Resource Discovery in Grid Environments Using a P2P Architecture [Seite 361]
12.2.3.2 - 17.3.2 P2P Protocols for Resource Discovery in the Grid [Seite 362]
12.2.4 - 17.4 The Grid-Routing Transferring Approach [Seite 365]
12.2.4.1 - 17.4.1 Resource Discovery Based on Matchmaking Routers [Seite 365]
12.2.4.2 - 17.4.2 Acquiring Knowledge in a Large-Scale Grid System [Seite 367]
12.2.5 - 17.5 Conclusions [Seite 369]
12.2.6 - Acknowledgment [Seite 370]
12.2.7 - References [Seite 370]
13 - Part VII Energy Awareness in High-Performance Computing [Seite 373]
13.1 - Chapter 18 Energy-Aware Approaches for HPC Systems [Seite 375]
13.1.1 - 18.1 Introduction [Seite 376]
13.1.2 - 18.2 Power Consumption of Servers [Seite 377]
13.1.2.1 - 18.2.1 Server Modeling [Seite 378]
13.1.2.2 - 18.2.2 Power Prediction Models [Seite 379]
13.1.3 - 18.3 Classification and Energy Profiles of HPC Applications [Seite 386]
13.1.3.1 - 18.3.1 Phase Detection [Seite 388]
13.1.3.2 - 18.3.2 Phase Identification [Seite 390]
13.1.4 - 18.4 Policies and Leverages [Seite 391]
13.1.5 - 18.5 Conclusion [Seite 392]
13.1.6 - Acknowledgements [Seite 393]
13.1.7 - References [Seite 393]
13.2 - Chapter 19 Strategies for Increased Energy Awareness in Cloud Federations [Seite 397]
13.2.1 - 19.1 Introduction [Seite 397]
13.2.2 - 19.2 Related Work [Seite 399]
13.2.3 - 19.3 Scenarios [Seite 401]
13.2.3.1 - 19.3.1 Increased Energy Awareness Across Multiple Data Centers within a Single Administrative Domain [Seite 401]
13.2.3.2 - 19.3.2 Energy Considerations in Commercial Cloud Federations [Seite 404]
13.2.3.3 - 19.3.3 Reduced Energy Footprint of Academic Cloud Federations [Seite 406]
13.2.4 - 19.4 Energy-Aware Cloud Federations [Seite 406]
13.2.4.1 - 19.4.1 Availability of Energy-Consumption-Related Information [Seite 407]
13.2.4.2 - 19.4.2 Service Call Scheduling at the Meta-Brokering Level of FCM [Seite 408]
13.2.4.3 - 19.4.3 Service Call Scheduling and VM Management at the Cloud-Brokering Level of FCM [Seite 409]
13.2.5 - 19.5 Conclusions [Seite 411]
13.2.6 - Acknowledgments [Seite 412]
13.2.7 - References [Seite 412]
13.3 - Chapter 20 Enabling Network Security in HPC Systems Using Heterogeneous CMPs [Seite 415]
13.3.1 - 20.1 Introduction [Seite 416]
13.3.2 - 20.2 Related Work [Seite 418]
13.3.3 - 20.3 Overview of Our Approach [Seite 419]
13.3.3.1 - 20.3.1 Heterogeneous CMP Architecture [Seite 419]
13.3.3.2 - 20.3.2 Network Security Application Behavior [Seite 420]
13.3.3.3 - 20.3.3 High-Level View [Seite 421]
13.3.4 - 20.4 Heterogeneous CMP Design for Network Security Processors [Seite 422]
13.3.4.1 - 20.4.1 Task Assignment [Seite 422]
13.3.4.2 - 20.4.2 ILP Formulation [Seite 423]
13.3.4.3 - 20.4.3 Discussion [Seite 425]
13.3.5 - 20.5 Experimental Evaluation [Seite 426]
13.3.5.1 - 20.5.1 Setup [Seite 426]
13.3.5.2 - 20.5.2 Results [Seite 427]
13.3.6 - 20.6 Concluding Remarks [Seite 429]
13.3.7 - Acknowledgments [Seite 429]
13.3.8 - References [Seite 429]
14 - Part VIII Applications of Heterogeneous High-Performance Computing [Seite 433]
14.1 - Chapter 21 Toward a High-Performance Distributed CBIR System for Hyperspectral Remote Sensing Data: A Case Study in Jungle Computing [Seite 435]
14.1.1 - 21.1 Introduction [Seite 436]
14.1.2 - 21.2 CBIR For Hyperspectral Imaging Data [Seite 439]
14.1.2.1 - 21.2.1 Spectral Unmixing [Seite 439]
14.1.2.2 - 21.2.2 Proposed CBIR System [Seite 441]
14.1.3 - 21.3 Jungle Computing [Seite 442]
14.1.3.1 - 21.3.1 Jungle Computing: Requirements [Seite 443]
14.1.4 - 21.4 IBIS and Constellation [Seite 444]
14.1.5 - 21.5 System Design and Implementation [Seite 447]
14.1.5.1 - 21.5.1 Endmember Extraction [Seite 450]
14.1.5.2 - 21.5.2 Query Execution [Seite 450]
14.1.5.3 - 21.5.3 Equi-Kernels [Seite 451]
14.1.5.4 - 21.5.4 Matchmaking [Seite 452]
14.1.6 - 21.6 Evaluation [Seite 452]
14.1.6.1 - 21.6.1 Performance Evaluation [Seite 453]
14.1.7 - 21.7 Conclusions [Seite 458]
14.1.8 - Acknowledgments [Seite 458]
14.1.9 - References [Seite 458]
14.2 - Chapter 22 Taking Advantage of Heterogeneous Platforms in Image and Video Processing [Seite 461]
14.2.1 - 22.1 Introduction [Seite 462]
14.2.2 - 22.2 Related Work [Seite 463]
14.2.2.1 - 22.2.1 Image Processing on GPU [Seite 463]
14.2.2.2 - 22.2.2 Video Processing on GPU [Seite 464]
14.2.2.3 - 22.2.3 Contribution [Seite 465]
14.2.3 - 22.3 Parallel Image Processing on GPU [Seite 465]
14.2.3.1 - 22.3.1 Development Scheme for Image Processing on GPU [Seite 465]
14.2.3.2 - 22.3.2 GPU Optimization [Seite 466]
14.2.3.3 - 22.3.3 GPU Implementation of Edge and Corner Detection [Seite 466]
14.2.3.4 - 22.3.4 Performance Analysis and Evaluation [Seite 466]
14.2.4 - 22.4 Image Processing on Heterogeneous Architectures [Seite 469]
14.2.4.1 - 22.4.1 Development Scheme for Multiple Image Processing [Seite 469]
14.2.4.2 - 22.4.2 Task Scheduling within Heterogeneous Architectures [Seite 470]
14.2.4.3 - 22.4.3 Optimization Within Heterogeneous Architectures [Seite 470]
14.2.5 - 22.5 Video Processing on GPU [Seite 470]
14.2.5.1 - 22.5.1 Development Scheme for Video Processing on GPU [Seite 471]
14.2.5.2 - 22.5.2 GPU Optimizations [Seite 472]
14.2.5.3 - 22.5.3 GPU Implementations [Seite 472]
14.2.5.4 - 22.5.4 GPU-Based Silhouette Extraction [Seite 472]
14.2.5.5 - 22.5.5 GPU-Based Optical Flow Estimation [Seite 472]
14.2.5.6 - 22.5.6 Result Analysis [Seite 475]
14.2.6 - 22.6 Experimental Results [Seite 476]
14.2.6.1 - 22.6.1 Heterogeneous Computing for Vertebra Segmentation [Seite 476]
14.2.6.2 - 22.6.2 GPU Computing for Motion Detection Using a Moving Camera [Seite 477]
14.2.7 - 22.7 Conclusion [Seite 479]
14.2.8 - Acknowledgment [Seite 480]
14.2.9 - References [Seite 480]
14.3 - Chapter 23 Real-Time Tomographic Reconstruction Through CPU + GPU Coprocessing [Seite 483]
14.3.1 - 23.1 Introduction [Seite 484]
14.3.2 - 23.2 Tomographic Reconstruction [Seite 485]
14.3.3 - 23.3 Optimization of Tomographic Reconstruction for CPUs and for GPUs [Seite 487]
14.3.4 - 23.4 Hybrid CPU + GPU Tomographic Reconstruction [Seite 489]
14.3.5 - 23.5 Results [Seite 491]
14.3.6 - 23.6 Discussion and Conclusion [Seite 493]
14.3.7 - Acknowledgments [Seite 495]
14.3.8 - References [Seite 495]
15 - Index [Seite 499]
16 - Series Page [Seite 501]
2 - Title Page [Seite 5]
3 - Contents [Seite 9]
4 - Contributors [Seite 25]
5 - Preface [Seite 29]
6 - European Science Foundation [Seite 31]
7 - Part I Introduction [Seite 33]
7.1 - Chapter 1 Summary of the Open European Network for High-Performance Computing in Complex Environments [Seite 35]
7.1.1 - 1.1 Introduction and Vision [Seite 36]
7.1.2 - 1.2 Scientific Organization [Seite 38]
7.1.2.1 - 1.2.1 Scientific Focus [Seite 38]
7.1.2.2 - 1.2.2 Working Groups [Seite 38]
7.1.3 - 1.3 Activities of the Project [Seite 38]
7.1.3.1 - 1.3.1 Spring Schools [Seite 38]
7.1.3.2 - 1.3.2 International Workshops [Seite 39]
7.1.3.3 - 1.3.3 Working Groups Meetings [Seite 39]
7.1.3.4 - 1.3.4 Management Committee Meetings [Seite 39]
7.1.3.5 - 1.3.5 Short-Term Scientific Missions [Seite 39]
7.1.4 - 1.4 Main Outcomes of the Action [Seite 39]
7.1.5 - 1.5 Contents of the Book [Seite 40]
7.1.6 - Acknowledgment [Seite 42]
8 - Part II Numerical Analysis for Heterogeneous and Multicore Systems [Seite 43]
8.1 - Chapter 2 On the Impact of the Heterogeneous Multicore and Many-Core Platforms on Iterative Solution Methods and Preconditioning Techniques [Seite 45]
8.1.1 - 2.1 Introduction [Seite 46]
8.1.2 - 2.2 General Description of Iterative Methods and Preconditioning [Seite 48]
8.1.2.1 - 2.2.1 Basic Iterative Methods [Seite 48]
8.1.2.2 - 2.2.2 Projection Methods: CG and GMRES [Seite 50]
8.1.3 - 2.3 Preconditioning Techniques [Seite 52]
8.1.4 - 2.4 Defect-Correction Technique [Seite 53]
8.1.5 - 2.5 Multigrid Method [Seite 54]
8.1.6 - 2.6 Parallelization of Iterative Methods [Seite 54]
8.1.7 - 2.7 Heterogeneous Systems [Seite 55]
8.1.7.1 - 2.7.1 Heterogeneous Computing [Seite 56]
8.1.7.2 - 2.7.2 Algorithm Characteristics and Resource Utilization [Seite 57]
8.1.7.3 - 2.7.3 Exposing Parallelism [Seite 58]
8.1.7.4 - 2.7.4 Heterogeneity in Matrix Computation [Seite 58]
8.1.7.5 - 2.7.5 Setup of Heterogeneous Iterative Solvers [Seite 59]
8.1.8 - 2.8 Maintenance and Portability [Seite 61]
8.1.9 - 2.9 Conclusion [Seite 62]
8.1.10 - Acknowledgments [Seite 63]
8.1.11 - References [Seite 63]
8.2 - Chapter 3 Efficient Numerical Solution of 2D Diffusion Equation on Multicore Computers [Seite 65]
8.2.1 - 3.1 Introduction [Seite 66]
8.2.2 - 3.2 Test Case [Seite 67]
8.2.2.1 - 3.2.1 Governing Equations [Seite 67]
8.2.2.2 - 3.2.2 Solution Procedure [Seite 68]
8.2.3 - 3.3 Parallel Implementation [Seite 71]
8.2.3.1 - 3.3.1 Intel PCM Library [Seite 71]
8.2.3.2 - 3.3.2 OpenMP [Seite 72]
8.2.4 - 3.4 Results [Seite 73]
8.2.4.1 - 3.4.1 Results of Numerical Integration [Seite 73]
8.2.4.2 - 3.4.2 Parallel Efficiency [Seite 74]
8.2.5 - 3.5 Discussion [Seite 77]
8.2.6 - 3.6 Conclusion [Seite 79]
8.2.7 - Acknowledgment [Seite 79]
8.2.8 - References [Seite 79]
8.3 - Chapter 4 Parallel Algorithms for Parabolic Problems on Graphs in Neuroscience [Seite 83]
8.3.1 - 4.1 Introduction [Seite 83]
8.3.2 - 4.2 Formulation of the Discrete Model [Seite 85]
8.3.2.1 - 4.2.1 The theta-Implicit Discrete Scheme [Seite 87]
8.3.2.2 - 4.2.2 The Predictor--Corrector Algorithm I [Seite 89]
8.3.2.3 - 4.2.3 The Predictor--Corrector Algorithm II [Seite 90]
8.3.3 - 4.3 Parallel Algorithms [Seite 91]
8.3.3.1 - 4.3.1 Parallel theta-Implicit Algorithm [Seite 91]
8.3.3.2 - 4.3.2 Parallel Predictor--Corrector Algorithm I [Seite 94]
8.3.3.3 - 4.3.3 Parallel Predictor--Corrector Algorithm II [Seite 95]
8.3.4 - 4.4 Computational Results [Seite 95]
8.3.4.1 - 4.4.1 Experimental Comparison of Predictor--Corrector Algorithms [Seite 98]
8.3.4.2 - 4.4.2 Numerical Experiment of Neuron Excitation [Seite 100]
8.3.5 - 4.5 Conclusions [Seite 101]
8.3.6 - Acknowledgments [Seite 102]
8.3.7 - References [Seite 102]
9 - Part III Communication and Storage Considerations in High-Performance Computing [Seite 105]
9.1 - Chapter 5 An Overview of Topology Mapping Algorithms and Techniques in High-Performance Computing [Seite 107]
9.1.1 - 5.1 Introduction [Seite 108]
9.1.2 - 5.2 General Overview [Seite 108]
9.1.2.1 - 5.2.1 A Key to Scalability: Data Locality [Seite 109]
9.1.2.2 - 5.2.2 Data Locality Management in Parallel Programming Models [Seite 109]
9.1.2.3 - 5.2.3 Virtual Topology: Definition and Characteristics [Seite 110]
9.1.2.4 - 5.2.4 Understanding the Hardware [Seite 111]
9.1.3 - 5.3 Formalization of the Problem [Seite 111]
9.1.4 - 5.4 Algorithmic Strategies for Topology Mapping [Seite 113]
9.1.4.1 - 5.4.1 Greedy Algorithm Variants [Seite 113]
9.1.4.2 - 5.4.2 Graph Partitioning [Seite 114]
9.1.4.3 - 5.4.3 Schemes Based on Graph Similarity [Seite 114]
9.1.4.4 - 5.4.4 Schemes Based on Subgraph Isomorphism [Seite 114]
9.1.5 - 5.5 Mapping Enforcement Techniques [Seite 114]
9.1.5.1 - 5.5.1 Resource Binding [Seite 115]
9.1.5.2 - 5.5.2 Rank Reordering [Seite 115]
9.1.5.3 - 5.5.3 Other Techniques [Seite 116]
9.1.6 - 5.6 Survey of Solutions [Seite 117]
9.1.6.1 - 5.6.1 Algorithmic Solutions [Seite 117]
9.1.6.2 - 5.6.2 Existing Implementations [Seite 117]
9.1.7 - 5.7 Conclusion and Open Problems [Seite 121]
9.1.8 - Acknowledgment [Seite 122]
9.1.9 - References [Seite 122]
9.2 - Chapter 6 Optimization of Collective Communication for Heterogeneous HPC Platforms [Seite 127]
9.2.1 - 6.1 Introduction [Seite 127]
9.2.2 - 6.2 Overview of Optimized Collectives and Topology-Aware Collectives [Seite 129]
9.2.3 - 6.3 Optimizations of Collectives on Homogeneous Clusters [Seite 130]
9.2.4 - 6.4 Heterogeneous Networks [Seite 131]
9.2.4.1 - 6.4.1 Comparison to Homogeneous Clusters [Seite 131]
9.2.5 - 6.5 Topology- and Performance-Aware Collectives [Seite 132]
9.2.6 - 6.6 Topology as Input [Seite 133]
9.2.7 - 6.7 Performance as Input [Seite 134]
9.2.7.1 - 6.7.1 Homogeneous Performance Models [Seite 135]
9.2.7.2 - 6.7.2 Heterogeneous Performance Models [Seite 137]
9.2.7.3 - 6.7.3 Estimation of Parameters of Heterogeneous Performance Models [Seite 138]
9.2.7.4 - 6.7.4 Other Performance Models [Seite 138]
9.2.8 - 6.8 Non-MPI Collective Algorithms for Heterogeneous Networks [Seite 138]
9.2.8.1 - 6.8.1 Optimal Solutions with Multiple Spanning Trees [Seite 139]
9.2.8.2 - 6.8.2 Adaptive Algorithms for Efficient Large-Message Transfer [Seite 139]
9.2.8.3 - 6.8.3 Network Models Inspired by BitTorrent [Seite 140]
9.2.9 - 6.9 Conclusion [Seite 143]
9.2.10 - Acknowledgments [Seite 143]
9.2.11 - References [Seite 143]
9.3 - Chapter 7 Effective Data Access Patterns on Massively Parallel Processors [Seite 147]
9.3.1 - 7.1 Introduction [Seite 147]
9.3.2 - 7.2 Architectural Details [Seite 148]
9.3.3 - 7.3 K-Model [Seite 149]
9.3.3.1 - 7.3.1 The Architecture [Seite 149]
9.3.3.2 - 7.3.2 Cost and Complexity Evaluation [Seite 150]
9.3.3.3 - 7.3.3 Efficiency Evaluation [Seite 151]
9.3.4 - 7.4 Parallel Prefix Sum [Seite 152]
9.3.4.1 - 7.4.1 Experiments [Seite 157]
9.3.5 - 7.5 Bitonic Sorting Networks [Seite 158]
9.3.5.1 - 7.5.1 Experiments [Seite 163]
9.3.6 - 7.6 Final Remarks [Seite 164]
9.3.7 - Acknowledgments [Seite 165]
9.3.8 - References [Seite 165]
9.4 - Chapter 8 Scalable Storage I/O Software for Blue Gene Architectures [Seite 167]
9.4.1 - 8.1 Introduction [Seite 167]
9.4.2 - 8.2 Blue Gene System Overview [Seite 168]
9.4.2.1 - 8.2.1 Blue Gene Architecture [Seite 168]
9.4.2.2 - 8.2.2 Operating System Architecture [Seite 168]
9.4.3 - 8.3 Design and Implementation [Seite 170]
9.4.3.1 - 8.3.1 The Client Module [Seite 171]
9.4.3.2 - 8.3.2 The I/O Module [Seite 173]
9.4.4 - 8.4 Conclusions and Future Work [Seite 174]
9.4.5 - Acknowledgments [Seite 174]
9.4.6 - References [Seite 174]
10 - Part IV Efficient Exploitation of Heterogeneous Architectures [Seite 177]
10.1 - Chapter 9 Fair Resource Sharing for Dynamic Scheduling of Workflows on Heterogeneous Systems [Seite 179]
10.1.1 - 9.1 Introduction [Seite 180]
10.1.1.1 - 9.1.1 Application Model [Seite 180]
10.1.1.2 - 9.1.2 System Model [Seite 183]
10.1.1.3 - 9.1.3 Performance Metrics [Seite 184]
10.1.2 - 9.2 Concurrent Workflow Scheduling [Seite 185]
10.1.2.1 - 9.2.1 Offline Scheduling of Concurrent Workflows [Seite 186]
10.1.2.2 - 9.2.2 Online Scheduling of Concurrent Workflows [Seite 187]
10.1.3 - 9.3 Experimental Results and Discussion [Seite 192]
10.1.3.1 - 9.3.1 DAG Structure [Seite 192]
10.1.3.2 - 9.3.2 Simulated Platforms [Seite 192]
10.1.3.3 - 9.3.3 Results and Discussion [Seite 194]
10.1.4 - 9.4 Conclusions [Seite 197]
10.1.5 - Acknowledgments [Seite 198]
10.1.6 - References [Seite 198]
10.2 - Chapter 10 Systematic Mapping of Reed--Solomon Erasure Codes on Heterogeneous Multicore Architectures [Seite 201]
10.2.1 - 10.1 Introduction [Seite 201]
10.2.2 - 10.2 Related Works [Seite 203]
10.2.3 - 10.3 Reed--Solomon Codes and Linear Algebra Algorithms [Seite 204]
10.2.4 - 10.4 Mapping Reed--Solomon Codes on Cell/B.E. Architecture [Seite 205]
10.2.4.1 - 10.4.1 Cell/B.E. Architecture [Seite 205]
10.2.4.2 - 10.4.2 Basic Assumptions for Mapping [Seite 206]
10.2.4.3 - 10.4.3 Vectorization Algorithm and Increasing its Efficiency [Seite 207]
10.2.4.4 - 10.4.4 Performance Results [Seite 209]
10.2.5 - 10.5 Mapping Reed--Solomon Codes on Multicore GPU Architectures [Seite 210]
10.2.5.1 - 10.5.1 Parallelization of Reed--Solomon Codes on GPU Architectures [Seite 210]
10.2.5.2 - 10.5.2 Organization of GPU Threads [Seite 212]
10.2.6 - 10.6 Methods of Increasing the Algorithm Performance on GPUs [Seite 213]
10.2.6.1 - 10.6.1 Basic Modifications [Seite 213]
10.2.6.2 - 10.6.2 Stream Processing [Seite 214]
10.2.6.3 - 10.6.3 Using Shared Memory [Seite 216]
10.2.7 - 10.7 GPU Performance Evaluation [Seite 217]
10.2.7.1 - 10.7.1 Experimental Results [Seite 217]
10.2.7.2 - 10.7.2 Performance Analysis using the Roofline Model [Seite 219]
10.2.8 - 10.8 Conclusions and Future Works [Seite 222]
10.2.9 - Acknowledgments [Seite 223]
10.2.10 - References [Seite 223]
10.3 - Chapter 11 Heterogeneous Parallel Computing Platforms and Tools for Compute-Intensive Algorithms: A Case Study [Seite 225]
10.3.1 - 11.1 Introduction [Seite 226]
10.3.2 - 11.2 A Low-Cost Heterogeneous Computing Environment [Seite 228]
10.3.2.1 - 11.2.1 Adopted Computing Environment [Seite 231]
10.3.3 - 11.3 First Case Study: The N-Body Problem [Seite 232]
10.3.3.1 - 11.3.1 The Sequential N-Body Algorithm [Seite 233]
10.3.3.2 - 11.3.2 The Parallel N-Body Algorithm for Multicore Architectures [Seite 235]
10.3.3.3 - 11.3.3 The Parallel N-Body Algorithm for CUDA Architectures [Seite 236]
10.3.4 - 11.4 Second Case Study: The Convolution Algorithm [Seite 238]
10.3.4.1 - 11.4.1 The Sequential Convolver Algorithm [Seite 238]
10.3.4.2 - 11.4.2 The Parallel Convolver Algorithm for Multicore Architectures [Seite 239]
10.3.4.3 - 11.4.3 The Parallel Convolver Algorithm for GPU Architectures [Seite 240]
10.3.5 - 11.5 Conclusions [Seite 243]
10.3.6 - Acknowledgments [Seite 244]
10.3.7 - References [Seite 244]
10.4 - Chapter 12 Efficient Application of Hybrid Parallelism in Electromagnetism Problems [Seite 247]
10.4.1 - 12.1 Introduction [Seite 247]
10.4.2 - 12.2 Computation of Green's functions in Hybrid Systems [Seite 248]
10.4.2.1 - 12.2.1 Computation in a Heterogeneous Cluster [Seite 249]
10.4.2.2 - 12.2.2 Experiments [Seite 250]
10.4.3 - 12.3 Parallelization in Numa Systems of a Volume Integral Equation Technique [Seite 254]
10.4.3.1 - 12.3.1 Experiments [Seite 254]
10.4.4 - 12.4 Autotuning Parallel Codes [Seite 258]
10.4.4.1 - 12.4.1 Empirical Autotuning [Seite 259]
10.4.4.2 - 12.4.2 Modeling the Linear Algebra Routines [Seite 261]
10.4.5 - 12.5 Conclusions and Future Research [Seite 262]
10.4.6 - Acknowledgments [Seite 263]
10.4.7 - References [Seite 264]
11 - Part V CPU + GPU Coprocessing [Seite 267]
11.1 - Chapter 13 Design and Optimization of Scientific Applications for Highly Heterogeneous and Hierarchical HPC Platforms Using Functional Computation Performance Models [Seite 269]
11.1.1 - 13.1 Introduction [Seite 270]
11.1.2 - 13.2 Related Work [Seite 273]
11.1.3 - 13.3 Data Partitioning Based on Functional Performance Model [Seite 275]
11.1.4 - 13.4 Example Application: Heterogeneous Parallel Matrix Multiplication [Seite 277]
11.1.5 - 13.5 Performance Measurement on CPUs/GPUs System [Seite 279]
11.1.6 - 13.6 Functional Performance Models of Multiple Cores and GPUs [Seite 280]
11.1.7 - 13.7 FPM-Based Data Partitioning on CPUs/GPUs System [Seite 282]
11.1.8 - 13.8 Efficient Building of Functional Performance Models [Seite 283]
11.1.9 - 13.9 FPM-Based Data Partitioning on Hierarchical Platforms [Seite 285]
11.1.10 - 13.10 Conclusion [Seite 289]
11.1.11 - Acknowledgments [Seite 291]
11.1.12 - References [Seite 291]
11.2 - Chapter 14 Efficient Multilevel Load Balancing on Heterogeneous CPU + GPU Systems [Seite 293]
11.2.1 - 14.1 Introduction: Heterogeneous CPU + GPU Systems [Seite 294]
11.2.1.1 - 14.1.1 Open Problems and Specific Contributions [Seite 295]
11.2.2 - 14.2 Background and Related Work [Seite 297]
11.2.2.1 - 14.2.1 Divisible Load Scheduling in Distributed CPU-Only Systems [Seite 297]
11.2.2.2 - 14.2.2 Scheduling in Multicore CPU and Multi-GPU Environments [Seite 300]
11.2.3 - 14.3 Load Balancing Algorithms for Heterogeneous CPU + GPU Systems [Seite 301]
11.2.3.1 - 14.3.1 Multilevel Simultaneous Load Balancing Algorithm [Seite 302]
11.2.3.2 - 14.3.2 Algorithm for Multi-Installment Processing with Multidistributions [Seite 305]
11.2.4 - 14.4 Experimental Results [Seite 307]
11.2.4.1 - 14.4.1 MSLBA Evaluation: Dense Matrix Multiplication Case Study [Seite 307]
11.2.4.2 - 14.4.2 AMPMD Evaluation: 2D FFT Case Study [Seite 309]
11.2.5 - 14.5 Conclusions [Seite 311]
11.2.6 - Acknowledgments [Seite 312]
11.2.7 - References [Seite 312]
11.3 - Chapter 15 The All-Pair Shortest-Path Problem in Shared-Memory Heterogeneous Systems [Seite 315]
11.3.1 - 15.1 Introduction [Seite 315]
11.3.2 - 15.2 Algorithmic Overview [Seite 317]
11.3.2.1 - 15.2.1 Graph Theory Notation [Seite 317]
11.3.2.2 - 15.2.2 Dijkstra's Algorithm [Seite 318]
11.3.2.3 - 15.2.3 Parallel Version of Dijkstra's Algorithm [Seite 319]
11.3.3 - 15.3 CUDA Overview [Seite 319]
11.3.4 - 15.4 Heterogeneous Systems and Load Balancing [Seite 320]
11.3.5 - 15.5 Parallel Solutions to The APSP [Seite 321]
11.3.5.1 - 15.5.1 GPU Implementation [Seite 321]
11.3.5.2 - 15.5.2 Heterogeneous Implementation [Seite 322]
11.3.6 - 15.6 Experimental Setup [Seite 323]
11.3.6.1 - 15.6.1 Methodology [Seite 323]
11.3.6.2 - 15.6.2 Target Architectures [Seite 324]
11.3.6.3 - 15.6.3 Input Set Characteristics [Seite 324]
11.3.6.4 - 15.6.4 Load-Balancing Techniques Evaluated [Seite 324]
11.3.7 - 15.7 Experimental Results [Seite 325]
11.3.7.1 - 15.7.1 Complete APSP [Seite 325]
11.3.7.2 - 15.7.2 512-Source-Node-to-All Shortest Path [Seite 327]
11.3.7.3 - 15.7.3 Experimental Conclusions [Seite 328]
11.3.8 - 15.8 Conclusions [Seite 329]
11.3.9 - Acknowledgments [Seite 329]
11.3.10 - References [Seite 329]
12 - Part VI Efficient Exploitation of Distributed Systems [Seite 333]
12.1 - Chapter 16 Resource Management for HPC on the Cloud [Seite 335]
12.1.1 - 16.1 Introduction [Seite 335]
12.1.2 - 16.2 On the Type of Applications for HPC and HPC2 [Seite 337]
12.1.3 - 16.3 HPC on the Cloud [Seite 338]
12.1.3.1 - 16.3.1 General PaaS Solutions [Seite 338]
12.1.3.2 - 16.3.2 On-Demand Platforms for HPC [Seite 342]
12.1.4 - 16.4 Scheduling Algorithms for HPC2 [Seite 343]
12.1.5 - 16.5 Toward an Autonomous Scheduling Framework [Seite 344]
12.1.5.1 - 16.5.1 Autonomous Framework for RMS [Seite 345]
12.1.5.2 - 16.5.2 Self-Management [Seite 347]
12.1.5.3 - 16.5.3 Use Cases [Seite 349]
12.1.6 - 16.6 Conclusions [Seite 351]
12.1.7 - Acknowledgment [Seite 352]
12.1.8 - References [Seite 352]
12.2 - Chapter 17 Resource Discovery in Large-Scale Grid Systems [Seite 355]
12.2.1 - 17.1 Introduction and Background [Seite 355]
12.2.1.1 - 17.1.1 Introduction [Seite 355]
12.2.1.2 - 17.1.2 Resource Discovery in Grids [Seite 356]
12.2.1.3 - 17.1.3 Background [Seite 357]
12.2.2 - 17.2 The Semantic Communities Approach [Seite 357]
12.2.2.1 - 17.2.1 Grid Resource Discovery Using Semantic Communities [Seite 357]
12.2.2.2 - 17.2.2 Grid Resource Discovery Based on Semantically Linked Virtual Organizations [Seite 359]
12.2.3 - 17.3 The P2P Approach [Seite 361]
12.2.3.1 - 17.3.1 On Fully Decentralized Resource Discovery in Grid Environments Using a P2P Architecture [Seite 361]
12.2.3.2 - 17.3.2 P2P Protocols for Resource Discovery in the Grid [Seite 362]
12.2.4 - 17.4 The Grid-Routing Transferring Approach [Seite 365]
12.2.4.1 - 17.4.1 Resource Discovery Based on Matchmaking Routers [Seite 365]
12.2.4.2 - 17.4.2 Acquiring Knowledge in a Large-Scale Grid System [Seite 367]
12.2.5 - 17.5 Conclusions [Seite 369]
12.2.6 - Acknowledgment [Seite 370]
12.2.7 - References [Seite 370]
13 - Part VII Energy Awareness in High-Performance Computing [Seite 373]
13.1 - Chapter 18 Energy-Aware Approaches for HPC Systems [Seite 375]
13.1.1 - 18.1 Introduction [Seite 376]
13.1.2 - 18.2 Power Consumption of Servers [Seite 377]
13.1.2.1 - 18.2.1 Server Modeling [Seite 378]
13.1.2.2 - 18.2.2 Power Prediction Models [Seite 379]
13.1.3 - 18.3 Classification and Energy Profiles of HPC Applications [Seite 386]
13.1.3.1 - 18.3.1 Phase Detection [Seite 388]
13.1.3.2 - 18.3.2 Phase Identification [Seite 390]
13.1.4 - 18.4 Policies and Leverages [Seite 391]
13.1.5 - 18.5 Conclusion [Seite 392]
13.1.6 - Acknowledgements [Seite 393]
13.1.7 - References [Seite 393]
13.2 - Chapter 19 Strategies for Increased Energy Awareness in Cloud Federations [Seite 397]
13.2.1 - 19.1 Introduction [Seite 397]
13.2.2 - 19.2 Related Work [Seite 399]
13.2.3 - 19.3 Scenarios [Seite 401]
13.2.3.1 - 19.3.1 Increased Energy Awareness Across Multiple Data Centers within a Single Administrative Domain [Seite 401]
13.2.3.2 - 19.3.2 Energy Considerations in Commercial Cloud Federations [Seite 404]
13.2.3.3 - 19.3.3 Reduced Energy Footprint of Academic Cloud Federations [Seite 406]
13.2.4 - 19.4 Energy-Aware Cloud Federations [Seite 406]
13.2.4.1 - 19.4.1 Availability of Energy-Consumption-Related Information [Seite 407]
13.2.4.2 - 19.4.2 Service Call Scheduling at the Meta-Brokering Level of FCM [Seite 408]
13.2.4.3 - 19.4.3 Service Call Scheduling and VM Management at the Cloud-Brokering Level of FCM [Seite 409]
13.2.5 - 19.5 Conclusions [Seite 411]
13.2.6 - Acknowledgments [Seite 412]
13.2.7 - References [Seite 412]
13.3 - Chapter 20 Enabling Network Security in HPC Systems Using Heterogeneous CMPs [Seite 415]
13.3.1 - 20.1 Introduction [Seite 416]
13.3.2 - 20.2 Related Work [Seite 418]
13.3.3 - 20.3 Overview of Our Approach [Seite 419]
13.3.3.1 - 20.3.1 Heterogeneous CMP Architecture [Seite 419]
13.3.3.2 - 20.3.2 Network Security Application Behavior [Seite 420]
13.3.3.3 - 20.3.3 High-Level View [Seite 421]
13.3.4 - 20.4 Heterogeneous CMP Design for Network Security Processors [Seite 422]
13.3.4.1 - 20.4.1 Task Assignment [Seite 422]
13.3.4.2 - 20.4.2 ILP Formulation [Seite 423]
13.3.4.3 - 20.4.3 Discussion [Seite 425]
13.3.5 - 20.5 Experimental Evaluation [Seite 426]
13.3.5.1 - 20.5.1 Setup [Seite 426]
13.3.5.2 - 20.5.2 Results [Seite 427]
13.3.6 - 20.6 Concluding Remarks [Seite 429]
13.3.7 - Acknowledgments [Seite 429]
13.3.8 - References [Seite 429]
14 - Part VIII Applications of Heterogeneous High-Performance Computing [Seite 433]
14.1 - Chapter 21 Toward a High-Performance Distributed CBIR System for Hyperspectral Remote Sensing Data: A Case Study in Jungle Computing [Seite 435]
14.1.1 - 21.1 Introduction [Seite 436]
14.1.2 - 21.2 CBIR For Hyperspectral Imaging Data [Seite 439]
14.1.2.1 - 21.2.1 Spectral Unmixing [Seite 439]
14.1.2.2 - 21.2.2 Proposed CBIR System [Seite 441]
14.1.3 - 21.3 Jungle Computing [Seite 442]
14.1.3.1 - 21.3.1 Jungle Computing: Requirements [Seite 443]
14.1.4 - 21.4 IBIS and Constellation [Seite 444]
14.1.5 - 21.5 System Design and Implementation [Seite 447]
14.1.5.1 - 21.5.1 Endmember Extraction [Seite 450]
14.1.5.2 - 21.5.2 Query Execution [Seite 450]
14.1.5.3 - 21.5.3 Equi-Kernels [Seite 451]
14.1.5.4 - 21.5.4 Matchmaking [Seite 452]
14.1.6 - 21.6 Evaluation [Seite 452]
14.1.6.1 - 21.6.1 Performance Evaluation [Seite 453]
14.1.7 - 21.7 Conclusions [Seite 458]
14.1.8 - Acknowledgments [Seite 458]
14.1.9 - References [Seite 458]
14.2 - Chapter 22 Taking Advantage of Heterogeneous Platforms in Image and Video Processing [Seite 461]
14.2.1 - 22.1 Introduction [Seite 462]
14.2.2 - 22.2 Related Work [Seite 463]
14.2.2.1 - 22.2.1 Image Processing on GPU [Seite 463]
14.2.2.2 - 22.2.2 Video Processing on GPU [Seite 464]
14.2.2.3 - 22.2.3 Contribution [Seite 465]
14.2.3 - 22.3 Parallel Image Processing on GPU [Seite 465]
14.2.3.1 - 22.3.1 Development Scheme for Image Processing on GPU [Seite 465]
14.2.3.2 - 22.3.2 GPU Optimization [Seite 466]
14.2.3.3 - 22.3.3 GPU Implementation of Edge and Corner Detection [Seite 466]
14.2.3.4 - 22.3.4 Performance Analysis and Evaluation [Seite 466]
14.2.4 - 22.4 Image Processing on Heterogeneous Architectures [Seite 469]
14.2.4.1 - 22.4.1 Development Scheme for Multiple Image Processing [Seite 469]
14.2.4.2 - 22.4.2 Task Scheduling within Heterogeneous Architectures [Seite 470]
14.2.4.3 - 22.4.3 Optimization Within Heterogeneous Architectures [Seite 470]
14.2.5 - 22.5 Video Processing on GPU [Seite 470]
14.2.5.1 - 22.5.1 Development Scheme for Video Processing on GPU [Seite 471]
14.2.5.2 - 22.5.2 GPU Optimizations [Seite 472]
14.2.5.3 - 22.5.3 GPU Implementations [Seite 472]
14.2.5.4 - 22.5.4 GPU-Based Silhouette Extraction [Seite 472]
14.2.5.5 - 22.5.5 GPU-Based Optical Flow Estimation [Seite 472]
14.2.5.6 - 22.5.6 Result Analysis [Seite 475]
14.2.6 - 22.6 Experimental Results [Seite 476]
14.2.6.1 - 22.6.1 Heterogeneous Computing for Vertebra Segmentation [Seite 476]
14.2.6.2 - 22.6.2 GPU Computing for Motion Detection Using a Moving Camera [Seite 477]
14.2.7 - 22.7 Conclusion [Seite 479]
14.2.8 - Acknowledgment [Seite 480]
14.2.9 - References [Seite 480]
14.3 - Chapter 23 Real-Time Tomographic Reconstruction Through CPU + GPU Coprocessing [Seite 483]
14.3.1 - 23.1 Introduction [Seite 484]
14.3.2 - 23.2 Tomographic Reconstruction [Seite 485]
14.3.3 - 23.3 Optimization of Tomographic Reconstruction for CPUs and for GPUs [Seite 487]
14.3.4 - 23.4 Hybrid CPU + GPU Tomographic Reconstruction [Seite 489]
14.3.5 - 23.5 Results [Seite 491]
14.3.6 - 23.6 Discussion and Conclusion [Seite 493]
14.3.7 - Acknowledgments [Seite 495]
14.3.8 - References [Seite 495]
15 - Index [Seite 499]
16 - Series Page [Seite 501]
