Energy-Aware Real-Time Scheduling of Sporadic Tasks Using a Hybrid Genetic Algorithm with EDF and DVFS

Résumé

This paper outlines a hybrid scheduling approach for multiprocessor real-time embedded systems. The proposed hybrid scheduler for multiprocessor real-time embedded systems integrates Earliest-Deadline-First (EDF) priorities with Dynamic Voltage and Frequency Scaling (DVFS) inside a Genetic Algorithm (GA). This approach tackles the NP-hard problem of scheduling sporadic tasks while jointly minimizing makespan and energy, avoiding deadline misses, and maintaining load balance. EDF is applied during fitness evaluation to enforce deadline-aware dispatch, and DVFS is co-evolved per task. In a heavy-load scenario with 150 task instances on 4 processors, the proposed GA (EDF=On, DVFS=On) achieved a 66.07 ms makespan and 0.20113 J energy, scheduling 150/150 tasks with zero deadline misses. Compared against the Non-dominated Sorting Genetic Algorithm II (NSGA-II) baseline without EDF/DVFS (makespan 253.17 ms, energy 0.36813 J, 132/150 scheduled), the proposed method reduced makespan by ~74% and energy by ~45%, while eliminating deadline misses. In a lighter 100-task scenario, both methods met all deadlines with comparable makespan/energy. These results highlight the benefit of unifying EDF and DVFS within an evolutionary framework for energy-aware real-time scheduling.

Citer ce document

Accès au document

Texte intégral en lecture en ligne, réservé aux abonnés SPHAERO et aux membres de l'institution. Se connecter

Auteur(s)

Statistiques

Consultations : 1

Téléchargements : 0