Scheduling analysis and correction for dependent real-time tasks upon heterogeneous multiprocessor architectures

Abstract

Current real-time systems applications are mostly large-scale, where tasks are intrinsically time-constrained and meaningfully dependent. Nevertheless, while the initial partitioning solution has a huge impact on scheduling results, most tasks-allocation problems are NP-complete and lack binding semantics. The scheduling analysis strategies are inevitably required to validate the system scheduling before its actual deployment. However, once the proposed scheduling is invalid, no possible corrections are provided, and a costing NP-complete partitioning regeneration is provoked as the only possible correction action. Whereas, sometimes nearby task re-allocations guarantee the scheduling correction. This paper proposes an approach to efficiently correct initial dependent task partitioning solutions while considering their allocation semantics. The ultimate objective is to conduct a correct schedule for multiprocessor real-time systems by means of a driven task re clustering and/or re-allocation without the need for extra complex partitioning regeneration. Simulation results show that our proposed approach is faster than classic partitioning regeneration, ameliorates the partitioning semantic results, and gives efficient scheduling results.

Type
Walid Karamti
Walid Karamti
ِAssistant Professor

My research interests include formal methods for the analysis of behaviors in real time systems and social networks.

Adel Mahfoudhi
Adel Mahfoudhi
Professor

My research interests include formal methods for Embedded Real Time System modeling, verification and Human Computer interaction, Cyber Security.