An Optimal Priority

Hard real-time systems require predictable timing behavior, and priority-driven preemptive scheduling is increasingly being used in these systems. Resources in these enviroiments should ideally be allocated to the highest-priority task. Priority inversion is a situation in which a higher-priority job is forced to wait for a lower-priority job. Priority inversion degrades system schedulability. Hence, priority inversion should be minimized in a hard real-time environment. Unfortunately, a direct application of synchronization primitives such as semaphores, monitors, and Ada rendezvous can cause uncontrolled priority inversion, a situation in which a low-priority job blocks a higher-priority job for an indeenite period of time. In this chapter we investigate policies belonging to the class of priority inheritance policies that minimize priority inversion. We develop a priority inheritance policy called the Optimal Mutex Policy (OMP) which has two desirable properties: deadlocks are avoided and the worst-case blocking duration of a job is bounded by the duration of execution of a single critical section of a lower-priority job. Real-time systems operate under strict timing constraints and include applications such as avionics systems, space-related systems like the Space Shuttle and Space Station, production control, robotics, and defense systems. Timing constraints of diierent tasks in real-time systems can be either hard, soft, or non-existent. Introduction 247 A timing constraint is considered to be hard if it must be met at all times, or is considered to be soft if it must be met only most of the time. For example, the processing of a reactor temperature reading can have a hard deadline if it must be completed before the next reading becomes available. An operator query typically has a soft deadline with a desired average-case response time. In addition, background tasks such as on-line testing may have no associated timing constraints at all. The failure to meet hard deadlines in these systems can potentially lead to catastrophic results such as loss of life and/or property. Real-time systems tend to be embedded systems which are not generally programmed by the end-user. Unlike traditional time-shared systems, tasks in hard real-time systems are known a priori. In particular, the worst-case behavior of tasks with hard deadlines and the average-case behavior of tasks with soft deadlines are reasonably well-tested and understood. Given a set of tasks and their associated timing constraints, two distinct approaches to the implementation of real-time systems are possible. One, called the time-line approach, is typiied by the cyclical …