Global Scheduling of Parallel Mixed-Criticality Real-time Tasks

Abstract

Safety-critical real-time systems, for example automotive vehicles, are often subject to certification at different safety assurance levels (criticality levels). With a growing number of functions, it becomes less sustainable to implement high criticality functionality, such as breaking, in isolation from other, less critical functionality such as infotainment systems, which have been a solution in the past. Instead, all functions may be implemented on a shared mixed-criticality platform, with higher efficiency in terms of heat, space and power as a result. Research has gone into schedulability of mixed-criticality real-time systems but so far, no schedulability analysis has been done when the real-time tasks are parallel and scheduled using a global scheduler. A global scheduler is free to dispatch a process to any processor and the purpose of this thesis is to perform scheduability analysis on parallel mixed-criticality tasks assuming global scheduling. In this thesis, four schedulability tests are developed. Schedulability tests are used for statically verifying that timing constraints are met in real-time applications. The first test derived is a general test which is applicable for any work-conserving (i.e. greedy) scheduler. Then, from the general test we derive a particular test for EDF and another for fixed priority. Finally, adjustments are made to the fixed priority test resulting in a fourth test which is compatible with Audsley’s algorithm for optimal priority assignment. The schedulability tests are evaluated in two settings. The first is a pure global setting where tasks may execute on which ever processor. In the second setting we partition the processors into clusters, then assign tasks to run only in a particular cluster, still using global scheduling for each cluster. Results show that schedulability is on par with state-of-the-art test for federated scheduling. Further, results also show that mixing federated and global scheduling using clustering may improve schedulability. We conclude that, theoretically, global scheduling is a viable option to scheduling parallel real-time tasks in a mixed-criticality system and that clustering the processors can increase schedulability.