Front Cover

Creativity in Computing and DataFlow SuperComputing

Copyright

Contents

Preface

Chapter One: A Systematic Approach to Generation of New Ideas for PhD Research in Computing

1. Introduction

2. Classification of Innovation Methods

2.1. Mendeleyevization (M)

2.2. Generalization (G)

2.3. Specialization (S)

2.4. Revitalization (R)

2.5. Crossdisciplinarization (C)

2.6. Implantation (I)

2.7. Adaptation (A)

2.8. Hybridization (H)

2.9. Transgranularization (T)

2.10. Extraparameterization (E)

3. Representative Examples From the Authors' PhD Theses

4. Conclusions

Acknowledgments

References

Methodology-Related References

Common Examples

Author´s PhD-Related References

Selected References of Young Researchers on the Faculty of the Department of Computer Engineering and Informatics, School o ...

Chapter Two: Exploring Future Many-Core Architectures: The TERAFLUX Evaluation Framework

1. Introduction

2. Terminology and Related Work

2.1. The Trade-off Between Simulation Accuracy and Speed

2.2. Simulation vs Emulation

2.3. The ``Functional-Directed´´ Simulation Technique

2.4. Using Sampling and FPGAs to Accelerate Simulation of Large Systems

2.5. Other Relevant Simulator Features

3. COTSon Framework Organization

4. Targeting a 1000-Core Simulation

4.1. Comparison Among Approaches to Evaluate Novel 1000-Core Architectures

4.2. Notes on the Evaluations Based on Physical Machines

5. How to Simulate 1000 Cores

5.1. Setup #1: Physical Machines, MPI Programming Model

5.2. Setup #2: Virtual Machines Running on Several Physical Machines, MPI Programming Model

5.3. Setup #3: Virtual Machines Running on a Single Physical Computer, MPI Programming Model

5.4. Setup #4: Virtual Machines Running on a Single Physical Computer, Flexible Programming Model on Top of a Distributed ...

5.5. Setup #5: Virtual Machines Running on a Single Physical Computer, Flexible Programming Model on Top of a Shared-Memo ...

5.6. Setup #6: Single Virtual Machine Running on a Single Physical Computer, Flexible Programming Model on Top of a Share ...

6. The Search for ``Efficient Benchmarks´´

7. Simulation Experiments

7.1. TERAFLUX Basic Node With up to 32 Cores

7.2. TERAFLUX Basic Communication Case With Two Nodes

7.3. TERAFLUX 1024-Core Machine (32 Nodes by 32 Cores)

8. Conclusions

Acknowledgments

References

Chapter Three: Dataflow-Based Parallelization of Control-Flow Algorithms

1. Introduction

2. Problem Statement

3. Dataflow Approaches and the Feynman Paradigm

4. Existing Solutions and Their Criticism

4.1. Methods Inherited From the Theory of Systolic Arrays

4.2. Methods Inherited From the Theory of Dataflow Analysis in Compilers

4.3. Methods Inherited From the Theory of Dataflow Programming Tools

5. Exploring Dataflow Potentials

5.1. The LBM

5.2. A Lattice–Boltzmann Implementation in the C Programming Language

5.3. Analytical Analysis of Potentials

5.4. A Lattice–Boltzmann Implementation for the Maxeler Dataflow Architecture

6. Performance Evaluation

6.1. Case Study: A Control-Flow and a Dataflow Implementation of the LBM

6.2. Dataflow Acceleration for Other Algorithms

6.3. Threat to Validity

7. Conclusions

Acknowledgments

Appendix

References

Chapter Four: Data Flow Computing in Geoscience Applications

1. Introduction

2. Data Flow Computing in HPC

2.1. Brief Summary of Data Flow Computing Model

2.2. Maxeler DFE

3. Geoscience Applications in HPC

3.1. Brief Summary

3.2. Climate Modeling

3.3. Exploration Geophysics

4. Case Study 1: Global Shallow Water Equations

4.1. Problem Description

4.1.1. Equations and Discretization

4.1.2. SWE Algorithm and Challenges

4.2. Hybrid Domain Partition Scheme

4.2.1. Hybrid Domain Decomposition Methodology

4.2.2. Adjustable Task Partition

4.3. Mixed Precision Arithmetic

4.3.1. Range Analysis

4.3.2. Precision Analysis

4.4. Performance and Power Efficiency

5. Case Study 2: Euler Atmospheric Equations

5.1. Problem Description

5.2. Algorithmic Offsetting

5.3. Fast Memory Table and Mixed Precision Arithmetic

5.4. Performance and Power Efficiency

6. Case Study 3: Reverse Time Migration

6.1. Problem Description

6.1.1. The Reverse Time Migration Algorithm

6.1.2. Computational Challenges in Reverse Time Migration

6.2. Random Boundary

6.3. A Customized Window Buffer

6.4. Cascading Multiple Computations

6.5. Number Representations

6.6. Hardware (De)compression Scheme

6.7. Performance and Power Efficiency

7. Summary and Concluding Remarks

Acknowledgments

Appendix

References

Chapter Five: A Streaming Dataflow Implementation of Parallel Cocke-Younger-Kasami Parser

1. Introduction

2. Problem Statement

2.1. Context-Free Languages

2.2. The CYK Algorithm

2.3. Modifications to the CYK Algorithm

2.4. Parallelizing CYK Parsing

3. Existing Solutions and Their Criticism

3.1. Existing Solutions for Shared Memory Multicore Systems

3.2. Existing Solutions for Distributed Memory Systems

3.3. Existing Solutions for Reconfigurable Hardware Systems

3.4. Existing Solutions for Many-Core (GPU) Systems

3.5. Summary of Presented Solutions

4. A Dataflow Implementation of a CYK Parser

5. Performance Analysis

5.1. Modeling Space and Time Requirements

5.2. Experimental Analysis

6. Conclusion

Acknowledgment

Appendix

References

Author Index

Subject Index

Contents of Volumes in this Series

Back Cover