Fault-tolerant numerical iterative algorithms at scale - Rapports de recherche et Technique de l'Inria
Rapport Année : 2025

Fault-tolerant numerical iterative algorithms at scale

Résumé

This work investigates how to protect numerical iterative algorithms from all types of errors that can strike at scale: fail-stop errors (a.k.a. failures) and silent errors, striking both as computation errors and memory bit-flips. We combine various techniques: detectors for computation errors, checksums for memory errors, and checkpoint/restart for failures. The objective is to minimize the expected time per iteration of the algorithm. We design a hierarchical pattern that combines and interleaves all these fault-tolerance mechanisms, and we determine the optimal periodic pattern that achieves this objective. We instantiate these results for the performance analysis of the Preconditioned Conjugate Gradient (PCG) algorithm: we report several scenarios where the optimal pattern dramatically decreases the overhead due to error mitigation.

Fichier principal
Vignette du fichier
RR-9567 new.pdf (831.05 Ko) Télécharger le fichier
Origine Fichiers produits par l'(les) auteur(s)
licence

Dates et versions

hal-04872041 , version 1 (08-01-2025)

Licence

Identifiants

  • HAL Id : hal-04872041 , version 1

Citer

Alix Tremodeux, Emmanuel Agullo, Anne Benoit, Luc Giraud, Thomas Herault, et al.. Fault-tolerant numerical iterative algorithms at scale. RR-9567, Inria Lyon. 2025. ⟨hal-04872041⟩
0 Consultations
0 Téléchargements

Partager

More