DALB-MC: dynamic Adaptive load balancing for efficient Monte Carlo particle transport on heterogeneous architectures

This paper proposes a Dynamic Adaptive Load Balancing Algorithm for Monte Carlo Particle Transport (DALB-MC algorithm) to enhance computational performance of Monte Carlo codes, particularly in heterogeneous environments. While Monte Carlo codes have adopted local thread parallelism and global paral...

Ausführliche Beschreibung

Gespeichert in:
Bibliographische Detailangaben
Veröffentlicht in:Journal of nuclear science and technology Jg. 62; H. 11; S. 1100 - 1111
Hauptverfasser: Wu, Jiajun, Zhang, Hui, Hu, Ankang, Pu, Yanheng, Wu, Zhen, Qiu, Rui, Li, Junli
Format: Journal Article
Sprache:Englisch
Veröffentlicht: Tokyo Taylor & Francis 02.11.2025
Taylor & Francis Ltd
Schlagworte:
Adaptive algorithms
Algorithms
Computer simulation
Load balancing
load balancing algorithm
Mcshield
Message passing
Monte Carlo simulation
multicore processors
parallel computing efficiency
Parallel processing
Resource utilization
ISSN:0022-3131, 1881-1248
Online-Zugang:Volltext
Tags: Tag hinzufügen
Keine Tags, Fügen Sie den ersten Tag hinzu!
Beschreibung
Zusammenfassung:This paper proposes a Dynamic Adaptive Load Balancing Algorithm for Monte Carlo Particle Transport (DALB-MC algorithm) to enhance computational performance of Monte Carlo codes, particularly in heterogeneous environments. While Monte Carlo codes have adopted local thread parallelism and global parallelism through message passing between nodes, traditional parallel algorithms face challenges in adapting to diverse hardware resources like tailing effect and long particle history in weight-window method. The proposed DALB-MC algorithm addresses this by dynamically partitioning tasks into sub-tasks and allocating them in real time based on each node's computational capacity, ensuring optimized resource utilization and reduced simulation time. Implemented on MCShield, the algorithm achieves approximately 20% computation time reduction in massively parallel simulations and significantly improves overall efficiency. To verify its effectiveness, self-defined arithmetic examples confirm the correctness of the algorithm, and additional tests on a Linux supercomputing platform with a complex benchmark case demonstrate its scalability and performance optimization under real-world engineering applications. The results highlight DALB-MC algorithm's capability to adaptively balance workloads, making it highly suitable for modern Monte Carlo particle transport simulations in diverse computational environments.
Bibliographie:ObjectType-Article-1
SourceType-Scholarly Journals-1
ObjectType-Feature-2
content type line 14
ISSN:0022-3131
1881-1248
DOI:10.1080/00223131.2025.2526784