Title: Multicore scheduling of real-time systems subject to permanent failure of one core with detection delay

Authors: Yves Mouafo Tchinda; Annie Choquet-Geniet; Gaëlle Largeteau-Skapin

Addresses: Laboratoire d'Informatique et d'Automatique pour les Systèmes, ISAE-ENSMA, Teleport 2, 1 avenue Clèment Ader BP 40109, 86961 Futuroscope-Chasseneuil, France ' Laboratoire d'Informatique et d'Automatique pour les Systèmes, ISAE-ENSMA, Teleport 2, 1 avenue Clèment Ader BP 40109, 86961 Futuroscope-Chasseneuil, France ' Laboratoire XLIM, Universitè de Poitiers, Tèlèport 2, 11 bd Marie et Pierre Curie, BP 30179, 86962 Futuroscope-Chasseneuil, France

Abstract: This work addresses the failure tolerance of a multicore real-time application with a non negligible detection delay and lost execution recovery. We propose three techniques having different scopes of use. The first one (SST) guarantees the validity and the fairness of the result but uses a high number of additional cores. The second (CRT) provides a valid schedule with a single additional core but cannot manage any system since several constraints are imposed. The third method (AFT), which can be applied to any system, uses the idle time units to recover the lost execution. However, a bounded number of temporal faults should be accepted. The experimental results for SST and CRT show that 100% of the simulated systems have a valid schedule. As for AFT, in 87% of the cases the tasks recover before their deadlines and the recovery delay is bounded for the 13% other cases.

Keywords: failure tolerance; multicore architecture; task re-execution; pfair scheduling; dynamic reconfiguration; idle time distribution; aperiodic flow.

DOI: 10.1504/IJCCBS.2018.096442

International Journal of Critical Computer-Based Systems, 2018 Vol.8 No.3/4, pp.258 - 281

Available online: 22 Nov 2018 *

Full-text access for editors Access for subscribers Purchase this article Comment on this article