Comparing Distributed Termination Detection Algorithms for Task-Based Runtime Systems on HPC platforms

George Bosilca (1) , Aurélien Bouteiller (1) , Thomas Hérault (1) , Valentin Le Fèvre (2) , Yves Robert (1, 3, 4) , Jack J Dongarra (1, 5)

1 ICL - Innovative Computing Laboratory [Knoxville]
2 BSC-CNS - Barcelona Supercomputing Center - Centro Nacional de Supercomputacion
3 ROMA - Optimisation des ressources : modèles, algorithmes et ordonnancement
4 LIP - Laboratoire de l'Informatique du Parallélisme
5 University of Manchester [Manchester]

This paper revisits distributed termination detection algorithms in the context of High-Performance Computing (HPC) applications. We introduce an efficient variant of the Credit Distribution Algorithm (CDA) and compare it to the original algorithm (HCDA) as well as to its two primary competitors: the Four Counters algorithm (4C) and the Efficient Delay-Optimal Distributed algorithm (EDOD). We analyze the behavior of each algorithm for some simplified task-based kernels and show the superiority of CDA in terms of the number of control messages. We then compare the implementation of these algorithms over a task-based runtime system, PaRSEC and show the advantages and limitations of each approach on a practical implementation.

Keywords: Termination detection credit distribution algorithms task-based HPC application control messages; Termination detection; credit distribution algorithms; task-based HPC application; control messages

Domains: Computer Science [cs]

George Bosilca, Aurélien Bouteiller, Thomas Hérault, Valentin Le Fèvre, Yves Robert, et al.. Comparing Distributed Termination Detection Algorithms for Task-Based Runtime Systems on HPC platforms. International Journal of Networking and Computing, 2022, 12 (1). ⟨hal-03920388⟩