engleski

Optimizacija rasporedivosti i kvalitete usluge u sustavima za rad u stvarnom vremenu s mješovitom kritičnošču

This thesis investigates three different methods for optimization of schedulability and quality of service in real-time mixed-criticality systems. The research is motivated by requirements of safety-critical systems, typically encountered in transportation and industrial systems, which are often represented with various real-time mixed-criticality system models. The first method is focused on the schedulability testing in the adaptive mixed-criticality system model. A novel sufficient schedulability test for adaptive mixed-criticality task systems is devised, which improves the schedulability in comparison with the existing sufficient schedulability tests. Moreover, the devised test requires significantly less computing power in comparison with exhaustive exact methods. In addition, a framework for schedulability testing was devised, which ensures comprehensive and systematic experimental evaluation of the devised test as well as the validation of the existing schedulability tests. Using this framework, errors and inconsistencies in the existing schedulability tests were corrected. The results presented in the thesis consist of extensive experimental evaluation on a large number of different synthetically generated task sets as well as on small numerical examples. The second method deals with the harmonic period assignment from period ranges, which is of great importance for maintaining schedulability and quality of service in the safety-critical real-time systems. Unlike the existing harmonic period assignment methods from period ranges, the method for harmonic period assignment devised in this research is optimal, and it enables optimization of the number of different period values in the system, which is often of practical interest in real-world applications. Moreover, the devised harmonic period assignment method enables the optimization of arbitrary utilization values. The method is validated and compared to existing methods using an extensive experimental evaluation. It was shown that the usage of this method can significantly increase schedulability and quality of service in sense of utilization in systems of interest. The third method addresses the optimization of quality of service of non-critical or low-criticality tasks. The method is based on the genetic programming method, which is often used in solving various scheduling problems. In this research, the genetic programming is exploited to generate dynamic priority assignment functions for scheduling of low-criticality tasks in the adaptive mixed-criticality environment, which is typically overloaded. The extensive experimental evaluation on synthetically generated task sets demonstrates that the proposed method can generate heuristics for various system configurations, which dominate single-variable based heuristics that can be found in the literature.

real-time ; scheduling ; mixed-criticality ; period assignment ; genetic programming

nije evidentirano