DISTRIBUTED MODEL EXPLORATION WITH EMEWS
View abstract on PubMed
Summary
This summary is machine-generated.This tutorial introduces advancements to the Extreme-scale Model Exploration with Swift (EMEWS) framework for high-performance computing. New features enhance accessibility and workflow distribution for large-scale computational model analysis.
Area Of Science
- Computational Science
- High-Performance Computing
- Software Frameworks
Background
- Increasing availability of high-performance computing (HPC) resources enables new simulation and computational tool applications.
- Model exploration (ME) frameworks are crucial for large-scale analyses like calibration and optimization of computational models.
Purpose Of The Study
- Present recent advancements to the Extreme-scale Model Exploration with Swift (EMEWS) framework.
- Highlight new capabilities for improved accessibility, distributed workflows, and project creation.
Main Methods
- Focus on three use-inspired EMEWS capabilities: binary installation, decoupled architecture (EMEWS DB) with task API, and enhanced project creation.
- Demonstrate EMEWS DB connecting a Python Bayesian optimization algorithm to heterogeneous compute resources (local and remote).
Main Results
- Improved accessibility via binary installation.
- New decoupled architecture (EMEWS DB) and task API enable distributed workflows on heterogeneous resources.
- Enhanced capabilities for creating and managing EMEWS projects.
Conclusions
- The presented EMEWS advancements facilitate large-scale model exploration on HPC resources.
- The worked example demonstrates the framework's utility in connecting diverse computational tools and resources.
- Further details and code are publicly available, promoting broader adoption and research.