High Performance Computing: 8th Latin American Conference, CARLA 2021, Guadalajara, Mexico, October 6–8, 2021, Revised Selected Papers (Communications in Computer and Information Science)

This book constitutes revised selected papers of the 8th Latin American High Performance Computing Conference, CARLA 2021, held in Guadalajara, Mexico, in October 2021. Due to the COVID-19 pandemic the conference was held in a virtual mode. The 16 revised full papers and 2 short papers presented were carefully reviewed and selected out of 45 submissions. The papers included in this book are organized according to the topics on ​high performance computing; high performance computing and artificial intelligence; high performance computing applications. Preface Organization Contents High Performance Computing TEPUI: High-Performance Computing Infrastructure for Beamlines at LNLS/Sirius 1 Introduction 2 Cluster Subsystems 2.1 High-Performance Computing 2.2 Storage 3 System Access 3.1 Scheduling System 4 Monitoring System 4.1 Infrastructure 5 Conclusion References Energy Consumption Studies of WRF Executions with the LIMITLESS Monitor 1 Introduction 2 WRF: The Standard for Weather Simulations 3 The LIMITLESS Monitor 4 Methodology 5 Results 6 Conclusions and Future Work References Improving Performance of Long Short-Term Memory Networks for Sentiment Analysis Using Multicore and GPU Architectures 1 Introduction 2 Related Work 3 Implementation 3.1 Technical Specifications 3.2 Training Data Pre-processing 3.3 Model and Training 4 Results 4.1 Accuracy of the Sentiment Analysis Model 4.2 Performance Improvements 4.3 Proposal Validation 5 Conclusion and Future Works References A Methodology for Evaluating the Energy Efficiency of Post-Moore Architectures 1 Introduction 2 State of the Art 3 ACDC Methodology Outline 3.1 Description of the ACDC Methodology 3.2 Case Study: Implementation of the ACDC Methodology 4 Conclusions References Understanding COVID-19 Epidemic in Costa Rica Through Network-Based Modeling 1 Introduction 2 Background 2.1 Epidemic Model Simulation 2.2 Corona++ Simulation Framework 2.3 Related Work 3 Modeling 3.1 Characterization 3.2 Calibration and Scenario Setup 4 Results 4.1 Experimental Setup 4.2 Validation 4.3 Evaluation of Scenarios 5 Conclusions and Further Work References An Efficient Vectorized Auction Algorithm for Many-Core and Multicore Architectures 1 Introduction 2 Related Work 3 Auction Algorithm 4 Vectorization of the Auction Algorithm for Multicore and Many-Core Architecture 4.1 Vectorization of the bid Phase 4.2 Vectorization of the assign Phase 5 Parallelization of the Auction Algorithm 6 Experimental Analysis 6.1 Experimental Analysis of the Vectorization 6.2 Experimental Analysis of the Parallel Vectorization 7 Conclusions References Green Energy HPC Data Centers to Improve Processing Cost Efficiency 1 Introduction 2 Development 3 Results 4 Conclusions References DICE: Generic Data Abstraction for Enhancing the Convergence of HPC and Big Data 1 Introduction 2 Data Containers 2.1 Input Containers 2.2 Output Containers 2.3 Storage Support for Containers 2.4 Interface 3 Reduction of I/O Interference 4 Evaluation 5 Related Work 6 Conclusions References A Comparative Study of Consensus Algorithms for Distributed Systems 1 Introduction 2 Implementation 2.1 Paxos 2.2 Raft 2.3 Practical Byzantine Fault Tolerance (pBFT) 3 Results 4 Conclusion References OCFTL: An MPI Implementation-Independent Fault Tolerance Library for Task-Based Applications 1 Introduction 2 Background and Motivation 2.1 Failure Detection and Propagation 2.2 Failure Mitigation 3 Related Work 4 An Implementation-Independent Fault Tolerance Library – OCFTL 4.1 Failure Detection 4.2 Handling Failures 4.3 Repairing Communicators 4.4 Gathering States 4.5 MPI Wrappers 4.6 Notification Callbacks 5 Experimental Results and Discussion 5.1 MPI Behavior 5.2 Empirical OCFTL Performance Evaluation 5.3 Internal Broadcast 5.4 Locality Problem 5.5 Checkpointing 5.6 Limitations 5.7 Stand-Alone Use 6 Conclusions and Future Work References Accelerating Smart City Simulations 1 Introduction 2 Related Works 3 SimEDaPE: Simulation Estimation by Data Patterns Exploration 3.1 Time Series Extraction 3.2 Clustering 3.3 Estimation 4 Low-Level Performance Optimizations 4.1 Optimizations 5 Experimental Results 5.1 SimEDaPE 5.2 Low-Level Optimizations 6 Conclusions References High Performance Computing and Artificial Intelligence Distributed Artificial Intelligent Model Training and Evaluation 1 Introduction 2 Implementation Auto Tuning for ML Training 2.1 Front End 2.2 Back End 3 Accelerated Prediction/Inference Implementation 4 Results 4.1 Manual Adjustment Testing 4.2 Hyper-parameter Scopes 4.3 Model Variations 4.4 Inference Results 5 Conclusions References Large-Scale Distributed Deep Learning: A Study of Mechanisms and Trade-Offs with PyTorch 1 Introduction 2 Related Work 3 Background 3.1 Deep Learning Neural Network Models 3.2 Deep Learning Frameworks 3.3 Distributed Deep Learning Training 4 Methodology 5 Experimental Results 5.1 Scalability Study 5.2 Accuracy and Scaling Trade-Off 5.3 Mixed-Precision Distributed Training 5.4 Adaptive Summation in Distributed Training 6 Discussion 7 Concluding Remarks References Wind Prediction Using Deep Learning and High Performance Computing 1 Introduction 2 Related Work 3 Theoretical Background 3.1 Time Series Wind Prediction 3.2 Convolutional Networks 4 Experiment Setup 4.1 Data 4.2 Experimental Framework 4.3 Technical Assessment and HPC Requirements 4.4 Hyper-parameter Tuning and Optimization 5 Experimentation 5.1 Multiple Input Multiple Output (MIMO) Approach 5.2 Common Parameters in all Architectures 5.3 Classic Convolutional Architecture 5.4 Separable Convolutional Architecture 5.5 Adding Skip and Residual Connections to the Models 5.6 Multi-head Architectures 6 Discussion and Conclusions 6.1 MIMO is a Reliable Approach for Deep Networks 6.2 Convolutional Networks are Superior to Baselines 7 Future Work References An Analysis of Neural Architecture Search and Hyper Parameter Optimization Methods 1 Introduction 2 Architecture and Hyper-parameters Optimization Overview 3 Optimization Methods 3.1 Bayesian Optimization 3.2 Population-Based Algorithms 3.3 Reinforcement Learning 3.4 Multi-objective Optimization 4 Occurrence Analysis 4.1 Cluster 1: Applications 4.2 Cluster 2: Learning 4.3 Cluster 3: Multi-objective 4.4 Cluster 4 and 5: Optimization Strategies 5 Conclusion References High Performance Computing Applications Solving the Heat Transfer Equation by a Finite Difference Method Using Multi-dimensional Arrays in CUDA as in Standard C 1 Introduction 2 Bidimensional Arrays 3 Tridimensional Arrays 4 Application to the Non-steady Heat Transport Equation 4.1 2D Case 4.2 3D Case 5 Performance Test 6 Conclusions References High-Throughput of Measure-Preserving Integrators Derived from the Liouville Operator for Molecular Dynamics Simulations on GPUs 1 Introduction 2 Methodology 3 Results 3.1 Code Validation 3.2 Code Performance 4 Discussion 5 Conclusions References An Efficient Parallel Model for Coupled Open-Porous Medium Problem Applied to Grain Drying Processing 1 Introduction 2 Numerical Simulation of Coupled Open-Porous Medium Problem 3 Methodology 3.1 The Algorithm 3.2 OpenMP Implementation 3.3 Experimental Setup 4 Experimental Results 4.1 Numerical Results 4.2 Preliminary Performance Analysis 4.3 Performance Evaluation 5 Conclusion and Future Works References Author Index