Reducing the fault vulnerability of hard real-time systems - CNRS - Centre national de la recherche scientifique Accéder directement au contenu
Article Dans Une Revue Journal of Systems Architecture Année : 2022

Reducing the fault vulnerability of hard real-time systems

Résumé

With the progress of the technology, the presence of transient faults (e.g. bit-flipping errors) in cache memories becomes a challenge, especially in embedded real-time systems. These are mission critical systems that are often subject to both fault-tolerant and real-time constraints. To reduce the impact of transient faults, hardware protection mechanisms are usually proposed. However, these mechanisms introduce too much pessimism in the computation of the worst-case execution time of a task, decreasing the overall system performance. In this paper, we propose a methodology to evaluate and reduce the vulnerability of hard real-time applications to soft errors in IL1 cache memories. We use static analysis tools to analyze a binary program and compute the overall vulnerability of its instructions. Then, we propose to reduce this vulnerability by invalidating some cache blocks at specific instants during the execution, thus forcing vulnerable instruction blocks to be reloaded from higher layers of memory. Since adding invalidation points will likely increase the WCETs of the tasks, we perform a static analysis to guarantee that the application deadlines are respected Finally, we analyze how our methodology can be combined with hardware protection mechanisms as ECC memories, and we evaluate the performance on a set of benchmarks.
Fichier principal
Vignette du fichier
main.pdf (486.96 Ko) Télécharger le fichier
Origine : Fichiers produits par l'(les) auteur(s)

Dates et versions

hal-03842393 , version 1 (07-11-2022)

Identifiants

Citer

Fabien Bouquillon, Smail Niar, Giuseppe Lipari. Reducing the fault vulnerability of hard real-time systems. Journal of Systems Architecture, 2022, 133, pp.102758. ⟨10.1016/j.sysarc.2022.102758⟩. ⟨hal-03842393⟩
65 Consultations
50 Téléchargements

Altmetric

Partager

Gmail Facebook X LinkedIn More