Composable Preemptive and Non-preemptive Real-time Scheduling Eliminating Component Overhead

Our work required a model for scheduling real-time tasks that extended the state of the art, supporting a flexible mix of preemptive and non-preemptive scheduling. Mixing these two kinds of scheduling is useful because preemption permits flexible allocation of CPU time while nonpreemptive scheduling has many benefits in software engineering (since race conditions become much more difficult to create) and efficiency. To support this model we developed two new abstractions, task clusters and task barriers that are described in our paper from RTSS 2002 [3]. Limitations of current techniques for measurementor analysis-based estimation of worst-case execution time (WCET) caused us to develop a way of evaluating real-time systems for their robustness to timing faults, or tasks that run for longer than expected. In our RTSS paper [3] we also described some theoretical results about the robustness of real-time schedules produced by existing scheduling algorithms, and we developed algorithms for creating highly robust schedules that are as resistant as possible to timing faults. Finally, as part of this work we developed and released the static priority analysis kit (SPAK) that provides an open-source implementation of all algorithms described in our paper from RTSS 2002.