Schedulability Analysis of Real-Time Systems Using Metaheuristic Search and Machine Learning

Schedulability analysis aims at determining whether task executions complete before their specified deadlines. It is an important activity in developing real-time systems. However, in practice, engineers have had difficulties applying existing techniques mainly because the working assumptions of existing methods are often not valid in their systems. Specifically, uncertainties in real-time systems and hybrid scheduling policies that combine standard scheduling policies have not been fully studied in the literature. This project develops an approach that analyzes the schedulability problem of real-time systems by accounting for such uncertainties and complex scheduling policies applied in practice. Our approach combines a metaheuristic search algorithm for generating worst-case scheduling scenarios with a machine learning technique for inferring a probability of deadline misses. To evaluate the practical usefulness of our work, we apply our approach to real systems developed by our industry partner, BlackBerry.

Faculty Supervisor:

Shiva Nejati;Lionel Briand

Student:

Jaekwon Lee

Partner:

Blackberry

Discipline:

Engineering - computer / electrical

Sector:

Manufacturing

University:

University of Ottawa