Parallel Processing and Applied Mathematics: 14th International Conference, PPAM 2022, Gdansk, Poland, September 11–14, 2022, Revised Selected Papers, ... (Lecture Notes in Computer Science, 13826)

This two-volume set, LNCS 13826 and LNCS 13827, constitutes the proceedings of the 14th International Conference on Parallel Processing and Applied Mathematics, PPAM 2022, held in Gdansk, Poland, in September 2022. The 77 regular papers presented in these volumes were selected from 132 submissions. For regular tracks of the conference, 33 papers were selected from 62 submissions. The papers were organized in topical sections named as follows: Part I: numerical algorithms and parallel scientific computing; parallel non-numerical algorithms; GPU computing; performance analysis and prediction in HPC systems; scheduling for parallel computing; environments and frameworks for parallel/cloud computing; applications of parallel and distributed computing; soft computing with applications and special session on parallel EVD/SVD and its application in matrix computations. Part II: 9th Workshop on Language-Based Parallel Programming (WLPP 2022); 6th Workshop on Models, Algorithms and Methodologies for Hybrid Parallelism in New HPC Systems (MAMHYP 2022); first workshop on quantum computing and communication; First Workshop on Applications of Machine Learning and Artificial Intelligence in High Performance Computing (WAML 2022); 4th workshop on applied high performance numerical algorithms for PDEs; 5th minisymposium on HPC applications in physical sciences; 8th minisymposium on high performance computing interval methods; 7th workshop on complex collective systems. Preface Organization Contents – Part I Contents – Part II Numerical Algorithms and Parallel Scientific Computing How Accurate Does Newton Have to Be? 1 Introduction 2 Auxiliary Results 3 Main Results 3.1 Stagnation 3.2 The Decay of the Error 3.3 Convergence 3.4 How Accurate Does Newton Have to Be? 4 Numerical Experiments 4.1 Computing Square Roots 4.2 Constrained Molecular Dynamics 5 Related Work 6 Conclusions References General Framework for Deriving Reproducible Krylov Subspace Algorithms: BiCGStab Case 1 Introduction 2 Background 3 General Framework for Reproducible Krylov Solvers 4 BiCGStab 5 Experimental Results 6 Conclusions References A Generalized Parallel Prefix Sums Algorithm for Arbitrary Size Arrays 1 Introduction 2 Parallel Prefix 3 Handling Arbitrary Size Inputs 4 The Right-Sweep Phase 5 Experimental Results 6 Conclusions References Infinite-Precision Inner Product and Sparse Matrix-Vector Multiplication Using Ozaki Scheme with Dot2 on Manycore Processors 1 Introduction 2 Related Work 3 Method 4 Performance Estimation 4.1 Throughput of GEMM and SpMM Using Dot2 4.2 Performance of IP-DOT and IP-SpMV 5 Demonstration on CPU and GPU 5.1 DOT 5.2 Reproducible CG Solvers 6 Conclusion References Advanced Stochastic Approaches for Applied Computing in Environmental Modeling 1 Introduction 2 Sensitivity Analysis - Definitions 3 Stochastic Approaches 4 Sensitivity Studies with Respect to Emission Levels 5 Sensitivity Studies with Respect to Chemical Reactions Rates 6 Conclusion References Parallel Non-numerical Algorithms Parallel Suffix Sorting for Large String Analytics 1 Introduction 2 Problem Description 3 Algorithm Design 3.1 Algorithm Framework 3.2 Parallel Induce Method 4 Complexity Analysis 5 Related Work 6 Conclusion References Parallel Extremely Randomized Decision Forests on Graphics Processors for Text Classification 1 Introduction 2 Trees and Ensembles 2.1 Decision Trees 2.2 Bagging 2.3 Boosting 2.4 Ensembles of Ensembles 3 Parallel Approach 3.1 Sampling 3.2 Class Count 3.3 Min/Max and Candidates 3.4 Find Best and Split 3.5 The Complete Solution 4 Experimental Results 4.1 System 4.2 Datasets 4.3 Analysis 5 Conclusions References RDBMS Speculative Support Improvement by the Use of the Query Hypergraph Representation 1 Introduction 2 Related Work 3 The Speculative Layer 4 A New Strategy for Speculative Query Assignment to Input Queries 5 A Hypergraph for Speculative Query Assignment 6 In Advance Speculative Query Matching Algorithm 7 Experimental Results 8 Conclusions References GPU Computing Mixed Precision Algebraic Multigrid on GPUs*-4pt 1 Introduction 2 Background on AMG and Related Work 3 Design of the Flexible and Platform-Portable AMG 4 Experiments 5 Conclusion References Compact In-Memory Representation of Decision Trees in GPU-Accelerated Evolutionary Induction 1 Introduction 2 Background 2.1 Decision Trees 2.2 Decision Tree Induction 2.3 Global Decision Tree System 3 GPU-Supported Evolution Using Compact In-Memory Representation of Decision Trees 3.1 In-Memory Representation of Decision Trees 3.2 GPU Kernels Implementation 4 Experimental Validation 4.1 Results 5 Conclusion References Neural Nets with a Newton Conjugate Gradient Method on Multiple GPUs 1 Introduction 2 Related Work 3 Methods 3.1 Scientific Computing for Deep Learning 3.2 State-of-the-Art Optimization Approaches 3.3 Proposed 2nd-Order Optimizer 4 Scenarios and Neural Network Architectures 5 Implementation 5.1 Automatic Differentiation Framework 5.2 Data Parallelism 5.3 Software and Hardware Setup 6 Results 6.1 Accuracy Results for Different Scenarios 6.2 Parallel Runs 7 Conclusion and Future Work References Performance Analysis and Prediction in HPC Systems Exploring Techniques for the Analysis of Spontaneous Asynchronicity in MPI-Parallel Applications 1 Introduction and Related Work 2 Case Studies, Testbed and Experimental Methods 2.1 Test Systems and Methodology 2.2 Synthetic Microbenchmarks 2.3 Proxy Memory-Bound Parallel Applications 2.4 Observables for Analysis 3 Simple Timeline Metrics for Analysis 3.1 Rank/ccNUMA-wise Timelines and Histogram of MPI Time and Performance 3.2 Timeline in Compact Representation 4 Advanced Metrics for Analysis 4.1 Correlation Coefficient 4.2 Phase Space Plots 5 Machine Learning Techniques for Analysis 5.1 Principal Component Analysis (PCA) 5.2 K-means Clustering 6 Summary and Future Work References Cost and Performance Analysis of MPI-Based SaaS on the Private Cloud Infrastructure 1 Introduction 2 The Governing Relations of the Discrete Element Method 3 DEM SaaS Deployed on OpenStack Cloud 3.1 Parallel DEM SaaS 3.2 OpenStack Cloud Infrastructure 4 The Cost and Performance Analysis 5 Conclusions References Building a Fine-Grained Analytical Performance Model for Complex Scientific Simulations*-4pt 1 Introduction 2 Performance Modeling Methodology 3 Modeling Hemocell 3.1 Hemocell 3.2 Performance-Relevant Functions and Parameters 3.3 Model-Building 3.4 Model Calibration 4 Scenario Analysis 4.1 Scenario: Balanced Workload 4.2 Scenario: Imbalanced Subdomains 4.3 Scenarios: Imbalanced Hematocrit 4.4 Scenario: Imbalanced Communication 5 Related Work 6 Conclusion References Evaluation of Machine Learning Techniques for Predicting Run Times of Scientific Workflow Jobs 1 Introduction 2 Related Work 3 Data Set Characterization 4 Models for Execution Time Prediction 4.1 Two-Stage Prediction Architecture 4.2 Model Training and Building Pipeline 4.3 Symbolic Regression Model 5 Evaluation 5.1 Prediction Architecture Evaluation 5.2 Impact of Granularity of the First-Stage Prediction 5.3 Increasing Specialization of Predictors 5.4 Employing Symbolic Regression to Prediction Tasks 6 Conclusion References Smart Clustering of HPC Applications Using Similar Job Detection Methods 1 Introduction 2 Background and Related Work 3 Solving the Problem of Smart Job Clustering 3.1 Proposed Method 3.2 Tuning of Proposed Solution 3.3 Evaluation of Proposed Method 4 Using Proposed Solution in Practice 5 Conclusions References Scheduling for Parallel Computing Distributed Work Stealing in a Task-Based Dataflow Runtime 1 Introduction 1.1 Contributions 2 Related Work 3 Adding Work Stealing to PaRSEC 4 Experiments 4.1 Benchmarks 4.2 Potential for Work Stealing 4.3 Thief Policy 4.4 Victim Policy 5 Conclusion References Task Scheduler for Heterogeneous Data Centres Based on Deep Reinforcement Learning 1 Introduction 2 Background 3 DRL for Scheduling in Heterogeneous Data Centres 3.1 Observation and Action Spaces 3.2 Agent Architecture 3.3 Size Reduction Through Clustering 4 Evaluation 5 Conclusions References Shisha: Online Scheduling of CNN Pipelines on Heterogeneous Architectures 1 Introduction 2 Motivation and Problem Definition 3 Background and Related Work 4 Shisha Exploration Approach 4.1 Seed Generation 4.2 Online Tuning 5 Experimental Setup 6 Evaluation 6.1 Comparison of Shisha with Exploration Algorithms 6.2 Analysis of Optimality 6.3 Importance of Seed in the Auto-tuning Phase of Shisha 6.4 Assignment and Balancing Schemes in Shisha 6.5 Sensitivity Analysis of 7 Conclusion References Proactive Task Offloading for Load Balancing in Iterative Applications 1 Introduction 2 Related Work 3 Preliminaries and Motivation 4 Online Load Prediction and Proactive Task Offloading 4.1 Online Load Prediction 4.2 Proactive Algorithm and Offloading Strategies 5 Evaluation 5.1 Environment and Online Prediction Evaluation 5.2 Artificial Imbalance Benchmark 5.3 Realistic PDE Use Case with Sam(oa)2 6 Conclusion References Environments and Frameworks for Parallel/Cloud Computing Language Agnostic Approach for Unification of Implementation Variants for Different Computing Devices 1 Introduction 2 Background and Insights 3 Program Synthesis 3.1 In FLASH 3.2 Modifications in Flash-X 4 Macros and Macroprocessor 5 Spark Variants 5.1 Variants 5.2 Unifying with Macros 6 Conclusions References High Performance Dataframes from Parallel Processing Patterns 1 Introduction 2 Dataframe Systems 2.1 Engineering Challenges 2.2 System Considerations 3 Distributed Memory Dataframe Framework 3.1 Distributed Memory Dataframe 3.2 Building Blocks 3.3 Generic Operator Patterns 3.4 Runtime Aspects 4 Cylon 4.1 Architecture 4.2 Features 5 Experiments 6 Related Work 7 Limitations and Future Work 8 Conclusion References Global Access to Legacy Data-Sets in Multi-cloud Applications with Onedata 1 Introduction 2 Problem Statement and Related Work 3 Data Indexing Subsystem 3.1 Policies and Options 3.2 Data Consistency 3.3 DIS Optimizations 4 Exposing Legacy Data Collections with Onedata 5 Evaluation 5.1 DIS Performance 5.2 Transparent Global Data Access 6 Conclusions References Applications of Parallel and Distributed Computing MD-Bench: A Generic Proxy-App Toolbox for State-of-the-Art Molecular Dynamics Algorithms 1 Introduction and Motivation 2 Related Work 3 Background and Theory 4 MD-Bench Features 4.1 Optimization Schemes 4.2 Benchmark Test Cases 4.3 Tools 5 Examples 5.1 Assembly Analysis 5.2 Investigate Memory Latency Contributions 5.3 Compiler Code Quality Study 6 Conclusion and Outlook References Breaking Down the Parallel Performance of GROMACS, a High-Performance Molecular Dynamics Software 1 Introduction 2 Background 3 Related Work 4 Methodology 5 Results 6 Discussion and Conclusion References GPU-Based Molecular Dynamics of Turbulent Liquid Flows with OpenMM 1 Introduction 2 Related Work 3 Software: OpenMM as a Flexible MD Framework 4 Constant Temperature Open Boundary Conditions 5 Open Boundary Conditions Implementation 6 Grid Aggregation 7 Performance Analysis 8 Conclusions References A Novel Parallel Approach for Modeling the Dynamics of Aerodynamically Interacting Particles in Turbulent Flows 1 Introduction 2 Methodology 3 Parallel Performance 3.1 Number of Particles 3.2 Size of the Particles 3.3 Size of the Short-Range Interaction Region 3.4 Number of CPU Cores 4 Conclusions References Reliable Energy Measurement on Heterogeneous Systems–on–Chip Based Environments 1 Introduction 2 Related Work 3 Reliable Energy Benchmarking 3.1 EML Java Native Interface 3.2 Reliable Benchmarking 4 Experimentation 5 Conclusion References Distributed Objective Function Evaluation for Optimization of Radiation Therapy Treatment Plans 1 Introduction 2 Background and Related Work 2.1 TROTS Dataset 2.2 Dose Influence Matrices 2.3 Radiatiation Therapy Plan Quality 3 Methodology 3.1 Serial Version and Data Preprocessing 3.2 Parallelization 3.3 Experimental Setup 4 Results 4.1 Performance and Parallel Scaling 4.2 Plan Quality 4.3 Performance Analysis and Execution Tracing 5 Discussion and Conclusion References Soft Computing with Applications GPU4SNN: GPU-Based Acceleration for Spiking Neural Network Simulations 1 Introduction 2 Background 2.1 General Flow of SNN Simulation 2.2 Izhikevich Neuron Model 2.3 Network Dynamics and Modes 2.4 Grid-Stride Loop 3 SNN Simulation Algorithms 4 Performance Evaluation 5 Discussion 6 Conclusion References Ant System Inspired Heuristic Optimization of UAVs Deployment for k-Coverage Problem 1 Introduction 2 The Optimization Problem 2.1 The Model of a Wireless Communication System 2.2 Hypergraph Representation of the System 2.3 Representation of Solution 2.4 The Optimization Criteria 3 The Optimization Method 3.1 The Problem–Specific Step: Generation of a Solution 3.2 The Main Loop 4 Experiments 4.1 Benchmark 4.2 Plan of Experiments 4.3 The Results 5 Conclusions References Dataset Related Experimental Investigation of Chess Position Evaluation Using a Deep Neural Network 1 Introduction 2 Related Work 3 Data Used for Experiments 3.1 Data Preparation 3.2 Board Representation 3.3 Data Labeling 4 Test Methods 4.1 Neural Network Architecture 4.2 Experiment 4.3 Training Method 5 Results 6 Discussion 7 Summary and Future Work References Using AI-based Edge Processing in Monitoring the Pedestrian Crossing 1 Introduction 2 Experimental Environment 3 Case with Stationary Cameras 3.1 Data Collecting 3.2 Network Training and Detection 4 Mobile Cameras 4.1 Analysis of Different Neural Network 4.2 Data Collecting 4.3 Network Training and Detection Results 4.4 Accuracy Analysis 5 Conclusions References Special Session on Parallel EVD/SVD and its Application in Matrix Computations Automatic Code Selection for the Dense Symmetric Generalized Eigenvalue Problem Using ATMathCoreLib 1 Introduction 2 Operation of ATMathCoreLib 3 Algorithms for the Dense Symmetric GEP and Their Implementations 4 Automatic Code Selection for the Dense Symmetric GEP Using ATMathCoreLib 5 Conclusion References On Relative Accuracy of the One-Sided Block-Jacobi SVD Algorithm 1 Introduction 2 One-Sided Block-Jacobi Algorithm with Preconditioning 3 Test Matrices 4 Implementation in MATLAB 5 Discussion of Numerical Results 6 Conclusions References Author Index