Improved Cache-Related Preemption Delay Estimation for Fixed Preemption Point Scheduling

摘要

Cache-Related Preemption Delays (CRPD) can significantly increase tasks' execution time in preemptive real-time scheduling, potentially jeopardising the system schedulability. In order to reduce the cumulative CRPD, Limited Preemptive Scheduling (LPS) has emerged as a scheduling approach which limits the maximum number of preemptions encountered by real-time tasks, thus decreasing CRPD compared to fully preemptive scheduling. Furthermore, an instance of LPS, called Fixed Preemption Point Scheduling (LP-FPP), defines the exact points where the preemptions are permitted within a task, which enables a more precise CRPD estimation. The majority of the research, in the domain of LP-FPP, estimates CRPD with pessimistic upper bounds, without considering the possible sources of over-approximation: (1) accounting for the infeasible preemption combinations, and (2) accounting for the infeasible cache block reloads. In this paper, we improve the analysis by accounting for those two cases towards a more precise estimation of the CRPD upper bounds. The evaluation of the approach on synthetic tasksets reveals a significant reduction of the pessimism in the calculation of the CRPD upper bounds, compared to the existing approaches.