Scheduling Fixed-Priority Tasks with Preemption Threshold

While it is widely believed that preemptability is a necessary requirement for developing real-time software, there are additional costs involved with preemptive scheduling, as compared to non-preemptive scheduling. Furthermore, in the context of fixed-priority scheduling, feasibility of a task set with nonpreemptive scheduling does not imply feasibility with preemptive scheduling (and vice-versa). In this paper, we use the notion of preemption threshold to develop a new scheduling model, which unifies the concepts of preemptive and non-preemptive scheduling, and subsumes both as special cases. The notion of preemption threshold was introduced by Express Logic, Inc. in their ThreadX real-time operating system. It allows a task to disable preemption from tasks up to a specified threshold priority. Tasks having priorities higher than the threshold are still allowed to preempt. In our new model, each task has a regular priority and a preemption threshold. We show how the new scheduling model can be analyzed for response times. We also develop algorithms for optimal assignment of priorities and preemption threshold. In this paper we also provide evidence that this new scheduling model provides substantial quantitative benefits over both preemptive and non-preemptive scheduling models. We show that the new model can result in substantial improvement of schedulability by taking advantage of aspects of both preemptive and non-preemptive scheduling. Furthermore, we show that the new model provides lower run-time costs by eliminating unnecessary preemptions. This work was presented in part at RTCSA’99, Hongkong, December 1999 and a brief version of this work was published in the conference proceedings of RTCSA’99 by the IEEE Computer Society Press.