Evaluation of Scheduling Algorithms for Embedded FreeRTOS-based Systems

Dynamic priority real-time scheduling (DPS), such as Earliest-Deadline First (EDF), offers high levels of system schedulability, implying that under this scheduling policy system processing capacity can be utilized in an optimized way. However, this schedulability gain can be compromised due to overheads associated with managing dynamic priority queues, making the use of DPS less appealing in low processing capacity embedded systems. In this paper we assess the overheads associated with two different implementations of EDF in FreeRTOS running on an ARM-M4 architecture, comparing them against Rate-Monotonic scheduling (RMS), a classic fixed-priority policy. The two EDF implementations differ from each other by the manner priority queues are implemented, based either on min-heap (EDF-H) or on multiple linked lists (EDF-L). Runtime overheads and schedulability are taken into consideration for different types of task sets and system loads. Results indicate that the higher overheads of EDF-H may lead to poor performance with respect to RMS in some scenarios. Even presenting slightly higher overheads than RMS, EDF-L was shown to perform consistently better in all considered experiments.