What is it about?
This paper investigates various load-balancing algorithms suitable for dynamic load balancing in domain decomposition-based high-performance computing (HPC) simulations. A brute force algorithm was developed, optimized, and examined. The results indicated that while this algorithm can serve as a useful tool for evaluating load-balancing strategies, it is only practical for the smallest use cases as a dynamic load balancer. Among the tested approximation algorithms, the Painter’s partition-based algorithms outperform the original SFC-based strategies across all tested cases. Additionally, combination algorithms outperform their single-algorithm counterparts and should be evaluated for potential use in production-scale simulations.
Featured Image
Photo by Markus Spiske on Unsplash
Why is it important?
Load balancing is critical for successful large-scale high-performance computing (HPC) simulations. As modern supercomputers become increasingly complex and variable, dynamic load balancing is becoming more crucial for efficiently utilizing computational resources. In this study, we investigate various standard dynamic load-balancing algorithms.
Perspectives
This research has shown that substantial improvement is possible for current-generation load-balancing strategies, creating interest in further study of load-balancing for AMReX applications. Areas of interest for future work include: (1) further refinement of current algorithms by investigating different space-filling curves, such as Hilbert, (2) testing these new algorithms in an AMReX application to study the impact on communication performance (3) expanding to uniform-machine scheduling techniques to utilize heterogeneous architectures efficiently, such as cloud resources and (4) whether direct mathematical optimization strategies can be effective as iterative load balancers if implemented asynchronously to the simulation, such as on a separate thread or a separate accelerator, including quantum accelerators.
Amitash Nanda
University of California San Diego
Read the Original
This page is a summary of: Exploring Dynamic Load Balancing Algorithms for Block-Structured Mesh-and-Particle Simulations in AMReX, July 2025, ACM (Association for Computing Machinery),
DOI: 10.1145/3708035.3736022.
You can read the full text:
Contributors
The following have contributed to this page
