Hard real-time preemptively scheduling with high context switch cost

Abstract

One of the most important aspects in the development of real-time systems is the scheduling policy to guarantee the response time of a task with timing constraints. Some real-time scheduling techniques use the advantage of stack scheduler properties to bound the number of context switches. In general purpose real-time systems, the context switch does not introduce a significant overhead in the system execution, so their importance in the analysis is not relevant. However, in some applications such as multimedia where the tasks exhibit a very high context switch cost, their importance could be crucial. In this case, the context switch is mainly due to the latency time of the I/O devices. In this paper, an analytic model is presented to calculate off-line the number of preemptions which a set of periodic tasks suffer, scheduled by a static (Rate Monotonic, Deadline Monotonic) and a dynamic (Earliest Deadline First) scheduler. Moreover, the necessary and sufficient conditions to construct a task set in which a subset will never be preempted being scheduled under the scheduling policies described is shown. Finally, an evaluation and comparison of both approaches is developed.