High Performance Computing: ISC High Performance Digital 2021 International Workshops, Frankfurt am Main, Germany, June 24 – July 2, 2021, Revised ... (Lecture Notes in Computer Science, 12761)

✍ Scribed by Heike Jagode (editor), Hartwig Anzt (editor), Hatem Ltaief (editor), Piotr Luszczek (editor)

• Publisher: Springer
• Year: 2021
• Tongue: English
• Leaves: 519
• Category: Library

✦ Synopsis

This book constitutes the refereed post-conference proceedings of 9 workshops held at the 35th International ISC High Performance 2021 Conference, in Frankfurt, Germany, in June-July 2021:

Second International Workshop on the Application of Machine Learning Techniques to Computational Fluid Dynamics and Solid Mechanics Simulations and Analysis; HPC-IODC: HPC I/O in the Data Center Workshop; Compiler-assisted Correctness Checking and Performance Optimization for HPC; Machine Learning on HPC Systems;4th International Workshop on Interoperability of Supercomputing and Cloud Technologies;2nd International Workshop on Monitoring and Operational Data Analytics;16th Workshop on Virtualization in High­-Performance Cloud Computing; Deep Learning on Supercomputers; 5th International Workshop on In Situ Visualization.

The 35 papers included in this volume were carefully reviewed and selected. They cover all aspects of research, development, and application of large-scale,high performance experimental and commercial systems. Topics include high-performance computing (HPC), computer architecture and hardware, programming models, system software, performance analysis and modeling, compiler analysis and optimization techniques, software sustainability, scientific applications, deep learning.

Chapter “Machine-Learning-Based Control of Perturbed and Heated Channel Flows” is available open access under a Creative Commons Attribution 4.0 International License via link.springer.com.

✦ Table of Contents

Preface
Organization
Contents
Second International Workshop on the Application of Machine Learning Techniques to Computational Fluid Dynamics and Solid Mechanics Simulations and Analysis
Second International Workshop on the Application of Machine Learning Techniques to Computational Fluid Dynamics and Solid Mechanics Simulations and Analysis
1 Background and Description
2 Workshop Summary
2.1 Keynote
2.2 Research Papers
Outline placeholder
2.2.1 Session 1: Fluid Mechanics with Turbulence, Reduced Models, and Machine Learning
2.2.2 Session 2: Novel Methods Development in Machine Learning and Fluid Simulation
2.2.3 Session 3: Confluence of Machine Learning and Fluid Simulation Applications
Organizing Committee
Machine-Learning-Based Control of Perturbed and Heated Channel Flows
1 Introduction
2 Numerical Methods
2.1 Computational Domain and Shape Generator
2.2 Thermal Lattice-Boltzmann Method
2.3 Proximal Policy Algorithm
3 Results
3.1 Validation of the Simulation Method
3.2 Comparison of Simulative and RL-Based Results
4 Summary, Conclusion, and Outlook
References
Novel DNNs for Stiff ODEs with Applications to Chemically Reacting Flows
1 Introduction
2 Deep Residual Neural Networks
2.1 Problem Formulation
2.2 Proposed Approach and Applications
3 Implementation
4 Results
4.1 Training with a Single Data Set
4.2 Training Data Sets with Varying Equivalence Ratio And fixed Initial Temperature
4.3 Grouping Training Data Based on Equivalence Ratio
4.4 Training Data Sets with Varying Initial Temperatures, but Fixed Equivalence Ratio
5 Conclusions and Future Directions
A Description of the Training Data Sets
References
Lettuce: PyTorch-Based Lattice Boltzmann Framework
1 Introduction
2 Software Description
2.1 Software Functionalities
2.2 Code Example
3 Advanced Functionalities
3.1 Machine Learning
3.2 Flow Control Through Automatic Differentiation
3.3 Benchmark
4 Conclusion
References
Reservoir Computing in Reduced Order Modeling for Chaotic Dynamical Systems
1 Introduction
2 Dimensionality Reduction with POD
3 Reservoir Computing - Echo State Neural Network
4 Network Training Parallelism
5 Numerical Results
5.1 Lorenz 63
5.2 Lorenz 96
5.3 Atmospheric Dispersion of a Pollutant Passive Scalar
6 Discussion and Conclusions
A Non Linear Transformation in RC-ESN Hidden Space
B Single Shot RC-ESN Loss Function Minimization
References
Film Cooling Prediction and Optimization Based on Deconvolution Neural Network
1 Introduction
2 Methodology and Data Acquisition
2.1 Test Case Configuration
2.2 Computational Setups
3 Deconvolution Modeling and Validation
3.1 Generation of Data Samples
3.2 Selection on Network Structure
3.3 Training and Validation of the Deconv NN
4 Results and Discussion
4.1 Comparison with Other Methods
4.2 Comparison with Different Data Size
5 Optimization
6 Conclusion
References
Turbomachinery Blade Surrogate Modeling Using Deep Learning
1 Introduction
2 Proposed Framework
3 Data Generation
4 Results and Discussion
5 Conclusion
References
A Data-Driven Wall-Shear Stress Model for LES Using Gradient Boosted Decision Trees
1 Introduction
2 Methodology
2.1 Datasets
2.2 Model Inputs and Output
2.3 Model Training and Testing
2.4 Model Integration
3 Results and Discussion
3.1 Turbulent Channel Flow
3.2 Wall-Mounted Hump
4 Conclusions
References
Nonlinear Mode Decomposition and Reduced-Order Modeling for Three-Dimensional Cylinder Flow by Distributed Learning on Fugaku
1 Introduction
1.1 Mode Decomposition Method
1.2 Mode Decomposition with Convolutional Neural Networks (CNNs)
1.3 Supercomputer Fugaku
2 Methods
2.1 Flow Field Simulation
2.2 Mode Decomposition
2.3 Reduced-Order Model with LSTM
3 Results
3.1 Computational Performance
3.2 Results of ROM Simulation
4 Conclusions
References
Using Physics-Informed Enhanced Super-Resolution Generative Adversarial Networks to Reconstruct Mixture Fraction Statistics of Turbulent Jet Flows
1 Introduction
2 Problem Definition and Data-Set Description
3 Modeling
3.1 Turbulent Mixing Models
3.2 Data-Driven Approach
4 Application to Turbulent Jet Flows
5 Conclusion
References
HPC I/O in the Data Center
1 ƒIntroduction
2 ƒOrganization of the Workshop
2.1 ‚Programm Committee
3 ƒWorkshop Summary
References
Toward a Workflow for Identifying Jobs with Similar I/O Behavior Utilizing Time Series Analysis
1 Introduction
2 Related Work
3 Methodology
3.1 Job Data
3.2 Algorithms for Computing Similarity
3.3 Methodology
4 Reference Job
5 Evaluation
5.1 Performance
5.2 Quantitative Analysis
6 Assessing Timelines for Similar Jobs
7 Conclusion
References
H3: An Application-Level, Low-Overhead Object Store
1 Introduction
2 Overview
3 Design and Implementation
3.1 Data and Metadata Separation
3.2 Metadata Organization
3.3 Data Organization
3.4 Example
4 The H3 Ecosystem
5 Evaluation
5.1 Single-Node Setup
5.2 Distributed Setup
6 Related Work
7 Future Extensions
8 Conclusion
References
Compiler-assisted Correctness Checking and Performance Optimization for HPC
Automatic Partitioning of MPI Operations in MPI+OpenMP Applications
1 Introduction
2 Analysis Approach
3 Demonstration of Feasibility
4 Conclusion
References
heimdallr: Improving Compile Time Correctness Checking for Message Passing with Rust
1 Introduction
2 Motivation
3 Correctness Problems with MPI
3.1 Type Safety Errors in MPI
3.2 Memory Safety Concerns with Non-blocking Communication
4 heimdallr: A Proof of Concept Message Passing Library Implementation in Rust
4.1 Type Safety Through Generics and Message Serialization
4.2 Ensuring Buffer Safety for Non-blocking Communication
5 Related Work
6 Evaluation
6.1 Performance Comparison on a Realistic Application
6.2 Micro Benchmarks for Message Passing Operations
6.3 Limitations of Compile Time Correctness Checks for Message Passing Applications
7 Conclusion and Future Work
References
Potential of Interpreter Specialization for Data Analysis
1 Introduction
2 Motivating Example
3 Characteristics of Data Analysis Workloads
3.1 Methodology
3.2 Hypothesis of Stable Data Types
3.3 Hypothesis About Full Script Coverage
3.4 Impact of the Use of Libraries
4 Semi-Automatic Source-Level Specialization via Compiler Optimizations
4.1 Identify the Interpretation Loop and Invariants
4.2 Specialization via Code Templates
4.3 Protect Against Misspeculation on Data Types
5 Evaluation of Potential Performance Improvement
6 Conclusion and Future Work
References
Refactoring for Performance with Semantic Patching: Case Study with Recipes
1 Introduction
2 Prospective Factorization Steps and Tool Choice
3 Thinking Out Coccinelle Transformation Rules
3.1 Identify AoS Variables for Reuse in SoA
3.2 Clone Structures and Make Them SoA
3.3 Helper Functions for SoA Array Memory Management
3.4 Transform Expressions from AoS to SoA, Globally
4 Current Status and Conclusions
References
Negative Perceptions About the Applicability of Source-to-Source Compilers in HPC: A Literature Review
1 Introduction
2 Methodology
2.1 Search Process
2.2 Inclusion and Exclusion Criteria
2.3 Qualitative Analysis
3 Results
4 Discussion
5 Threats to Validity
6 Related Work
7 Conclusion
References
Machine Learning on HPC Systems
Automatic Tuning of Tensorflow's CPU Backend Using Gradient-Free Optimization Algorithms
1 Introduction
1.1 Contributions
2 Background
2.1 TensorFlow
2.2 Black-Box Optimization Algorithms
3 Optimization Framework and Methodology
4 Evaluation
4.1 Experimental Setup
4.2 Results
4.3 Comparison of the Optimization Algorithms
5 Related Work
6 Conclusion
References
MSM: Multi-stage Multicuts for Scalable Image Clustering
1 Introduction
2 Related Work
3 Notation and Methods
3.1 Minimum Cost Mulitcut Problem
3.2 Image Clustering with Multicuts
3.3 MSM: Multi-stage Multicuts
3.4 Theoretical Proof
4 Experiments
4.1 Two-Stage Approach on CIFAR10 Dataset
4.2 Sparsity is Influencing the Runtime on Merge
4.3 MSM with 3-Stages on CelebA
4.4 MSM with 4-Stages on CIFAR100
5 Conclusion and Future Work
References
OmniOpt – A Tool for Hyperparameter Optimization on HPC
1 Introduction
1.1 Outline
2 Methods and Tools
2.1 The HPC System
2.2 Features of OmniOpt
2.3 Architecture of OmniOpt
2.4 Application to a Problem from Material Science
3 Results
3.1 Initial Tests
3.2 Optimization Results
3.3 Visualization in the HP Space
4 Conclusions
References
Parallel/Distributed Intelligent Hyperparameters Search for Generative Artificial Neural Networks
1 Introduction
2 Automatic Parameter Configuration of GANs
2.1 Parameter Configutation of GANs
2.2 Software Libraries Used in the Proposed System
2.3 Related Work
3 The Proposed Parallel/Distributed Approach for Hyperparameters Search
3.1 Overall Description
4 Experimental Evaluation
4.1 Evaluation Methodology
4.2 Numerical Results
5 Conclusions and Future Work
References
Machine Learning for Generic Energy Models of High Performance Computing Resources
1 Introduction
2 Generalization of HPC Energy Models
2.1 Energy Consumption Models
2.2 Related Work
3 Generalization Capabilities of HPC Power Consumption Models Models
3.1 Overview
3.2 Description of the Evaluated Techniques
3.3 Training Data for Building the Models
4 Experimental Evaluation
4.1 Design of Experiments
4.2 Metrics Considered in the Evaluation
4.3 Results
5 Conclusions and Future Work
References
Fourth International Workshop on Interoperability of Supercomputing and Cloud Technologies
Automation for Data-Driven Research with the NERSC Superfacility API
1 Introduction and Motivation
2 Related and Prior Work
2.1 Similar HPC APIs
2.2 Comparison to the Superfacility API
3 Summary of API Functions
4 General Use-Case Drivers for API Functionality
4.1 Example: Integrating the Superfacility API into an LCLS-II Pipeline
5 API Design and Architecture
5.1 Deployment Details and Integration with Underlying Systems
5.2 Authentication and Authorization Model
5.3 API Internal Message Queue and Call Handling Sequence
6 Remarks on Portability
7 Future Capabilities Envisioned
References
A Middleware Supporting Data Movement in Complex and Software-Defined Storage and Memory Architectures
1 Introduction
2 Architecture
3 Core Middleware API
4 Performance Evaluation
5 Dynamic Provisioning and Workflow Support
6 Related Work
7 Conclusion
References
Second International Workshop on Monitoring and Operational Data Analytics
1 Introduction
2 Workshop Organisation
Organising Committee
Program Committee
Technical Program
3 Conclusion
An Operational Data Collecting and Monitoring Platform for Fugaku: System Overviews and Case Studies in the Prelaunch Service Period
1 Introduction
2 Related Works
3 Operational Data Collection and Monitoring Platform in the Fugaku Environment
3.1 Overview of Data Sources
3.2 System Architecture
4 Results
4.1 Case 1: The Integrated Monitoring Platform for the HPC System and the Data Center Infrastructure
4.2 Case 2: Monitoring Overloading Jobs on the Lustre-Based Parallel File System
5 Summary and Future Work
References
An Explainable Model for Fault Detection in HPC Systems
1 Introduction
2 Related Works
3 Methodology
4 Case Study: Marconi HPC System
4.1 Dataset
4.2 Raw Data Pre-processing
4.3 Data Description
5 Experimental Results
5.1 Fault Detection Robustness
5.2 Visualization of Learned Parameters for Bayesian Classifier TrueExplain
6 Discussion
7 Conclusions
8 Code Access
References
Sixteenth Workshop on Virtualization in High–Performance Cloud Computing
A Scalable Cloud Deployment Architecture for High-Performance Real-Time Online Applications
1 Introduction
2 Our Scalable Cloud Deployment Architecture for High-Performance ROIAs
3 Use Case and Evaluation
4 Conclusion and Future Work
References
Leveraging HW Approximation for Exploiting Performance-Energy Trade-offs Within the Edge-Cloud Computing Continuum
1 Introduction
2 Background and Related Work
2.1 HLS and RTL Approaches for FPGA Design
2.2 Approximation Techniques
3 Proposed Framework Extension
3.1 Common FPGA Interface
3.2 Building a Source-to-Source HLS Compiler
3.3 Kernels and Algorithms Tested
4 Evaluation
4.1 Resource Utilization and Algorithmic Performance
4.2 Approximate and Default Algorithm Comparative Analysis
5 Conclusion
References
Datashim and Its Applications in Bioinformatics
1 Introduction
2 Method
2.1 Dataset Custom Resource Definition
2.2 Dataset Operator
2.3 Pods Admission Controller
3 Result
3.1 ML: Data Volumes for Notebook Servers
3.2 ML: Pod Labels for TensorBoard
3.3 Non-ML: Dynamic Dataset with Kubeflow Pipelines APIs
3.4 Pipeline for 1000 Genomes Project Simplified
4 Discussion
5 Conclusion
References
FaaS and Curious: Performance Implications of Serverless Functions on Edge Computing Platforms
1 Introduction
2 Background and Related Work
3 Systematic Analysis of Serverless Infrastructures
3.1 Proposed Methodology
3.2 Target Serverless Frameworks
3.3 Target Cluster Infrastructure
4 Evaluation
5 Conclusion and Future Work
References
Differentiated Performance in NoSQL Database Access for Hybrid Cloud-HPC Workloads
1 Introduction
1.1 Contributions
2 Related Work
3 Proposed Approach
4 Experimental Evaluation
5 Conclusions
References
Deep Learning on Supercomputers
JUWELS Booster – A Supercomputer for Large-Scale AI Research
1 Introduction
2 JUWELS Booster System
2.1 JSC Supercomputer Ecosystem
2.2 JUWELS Booster
2.3 Distributed Model Training on JUWELS Booster
2.4 Benchmark Results
3 Large Scale AI Research at JSC
3.1 Large-Scale Deep Learning for Efficient Cross-Domain Transfer
3.2 Deep Learning-Driven Weather Forecast
3.3 Multispectral Remote Sensing Image Classification
3.4 RNA Structure with ML
4 Summary and Outlook
References
Fifth International Workshop on in Situ Visualization
1 Background and Description
2 Workshop Summary
2.1 Keynote
2.2 Abstracts
2.3 Full Papers
3 Organizing Committee
Workshop Chairs
Workshop Co-organizers
Programm Committee
In Situ Visualization of WRF Data Using Universal Data Junction
1 Introduction
2 Materials and Methods
2.1 Universal Data Junction
2.2 Inshimtu
2.3 Inshimtu+UDJ
2.4 Test Case
2.5 Simulation Environment and Setup
3 Results and Discussion
4 Related Work
5 Conclusion and Future Work
References
Catalyst Revised: Rethinking the ParaView in Situ Analysis and Visualization API
1 Introduction
2 Design
2.1 Simplifying the Adaptor
2.2 Simplifying Build and Deployment
3 Implementation
4 Evaluation
4.1 Debugging and Regression Testing
5 Conclusion and Future Work
References
Fides: A General Purpose Data Model Library for Streaming Data
1 Introduction
2 Previous Work
3 Design and Implementation of Fides
3.1 Enabling Technologies
3.2 Data Description Schema
3.3 Mesh Support
3.4 Code-Specific Support
3.5 Data Set Writer
4 Results
4.1 Use Cases
4.2 Metric Evaluation
5 Conclusion and Future Work
References
Including in Situ Visualization and Analysis in PDI
1 Motivation
2 pdi2sensei
3 Example
4 Conclusion
References
Correction to: An Explainable Model for Fault Detection in HPC Systems
Correction to: Chapter “An Explainable Model for Fault Detection in HPC Systems” in: H. Jagode et al. (Eds.): High Performance Computing, LNCS 12761, https://doi.org/10.1007/978-3-030-90539-2_25
Correction to: Machine-Learning-Based Control of Perturbed and Heated Channel Flows
Correction to: Chapter “Machine-Learning-Based Control of Perturbed and Heated Channel Flows” in: H. Jagode et al. (Eds.): High Performance Computing, LNCS 12761, https://doi.org/10.1007/978-3-030-90539-2_1
Author Index