Asynchronous Many-Task Systems and Applications: First International Workshop, WAMTA 2023, Baton Rouge, LA, USA, February 15–17, 2023, Proceedings

Description

This book constitutes the proceedings of the Workshop on Asynchronous Many-Task Systems and Applications 2023 in Baton Rouge, LA, USA, February 2023. The workshop present  the advantages and challenges of task-based programming on modern and future HPC systems. The 6 full papers included in this volume were carefully reviewed and selected from 7 submissions. Preface Organization Contents Extending Hedgehog's Dataflow Graphs to Multi-node GPU Architectures 1 Introduction 2 Existing Approaches 2.1 Uintah 2.2 DPLASMA 2.3 SLATE 2.4 Hedgehog 3 Extending Hedgehog to Multiple Nodes 3.1 DataPacket 3.2 Sender Task 3.3 Receiver Task 4 Matrix Multiplication Using Hedgehog 4.1 Single-Node Setup 4.2 Multiple Node Setup 4.3 Communications 5 Results 6 Conclusions 7 Future Work References Command Horizons: Coalescing Data Dependencies While Maintaining Asynchronicity 1 Introduction and Related Work 2 Background 2.1 The Celerity Runtime System 2.2 Data State Tracking 2.3 Generative Data Access Patterns 3 Horizons 4 Evaluation 5 Conclusion References Shared Memory Parallelism in Modern C++ and HPX 1 Introduction 2 Related Work 3 HPX 4 Approaches 4.1 Futures and Futurization 4.2 Coroutines 4.3 Parallel Algorithms 4.4 Senders and Receivers 5 Performance Comparison 6 Conclusion References Framework for Extensible, Asynchronous Task Scheduling (FEATS) in Fortran 1 Introduction 2 Implementation 3 Advantages, Disadvantages, and Examples 3.1 Advantages 3.2 Disadvantages 3.3 Examples 4 Conclusion References Scalability of Gaussian Processes Using Asynchronous Tasks: A Comparison Between HPX and PETSc 1 Introduction 2 Related Work 3 Scientific Application 3.1 Gaussian Processes 3.2 System Identification 4 Parallel Algorithms 4.1 Tiled Cholesky Decomposition 4.2 Tiled Triangular Solve 4.3 Research Questions and Hardware Setup 5 Software Framework 5.1 HPX 5.2 PETSc 6 Results 6.1 Tile Scaling 6.2 Strong Scaling Comparison 7 Conclusion and Outlook References Scheduling Many-Task Applications on Multi-clouds and Hybrid Clouds 1 Introduction 2 Decentralized Cloud Scheduling Approach 2.1 Overlay Graph Construction 2.2 Three-Dimensional Vector Scheduling 2.3 Simulated Performance and Latency Variations 2.4 Using Virtual Machines 3 Multi-cloud Experiments 4 Related Work 5 Conclusion and Future Work References Author Index