Section: New Results

Monoprocessor real-time scheduling

Participants : Liliana Cucu-Grosjean, Laurent George, Patrick Meumeu Yomsi, Daniel de Rauglaudre, Yves Sorel.

Last year, we introduced a new model, called the otask model , in order to solve the general scheduling problem of hard real-time systems with various kinds of constraints. This new model allowed us to perform the schedulability analysis of a task set while taking the exact RTOS cost into account for any scenario of first release for all tasks and when priorities are given according to a fixed-priority scheduling policy. In this case, the schedulability analysis is based on an algebraic approach through the definition of a binary scheduling operation denoted by which is associative and not commutative. For a given set of otasks, operation is used as many times as there are tasks in the system.

This year, we have closely studied the impact of the scenario of first release times of all tasks [25] , [27] , [14] , then the impact of the priority assignment and finally the impact of the scheduler cost [26] , all on both the schedule and the schedulability analysis for a given system [12] . First, when tasks are scheduled according to any fixed-priority scheduling policy, we have proposed an algorithm to improve the sensitivity of the deadlines and then the quality of control of the system by using our approach for the computation of the worst response times. Here, the worst response time of a task is the maximum time elapsed between the activation times and the completion times of the task amongst all its instances. Second, because a dynamic-priority scheduling policy may produce a valid schedule for a system that can not be schedulable by any fixed-priority scheduling policy, we have extended our approach to tackle the scheduling problem where priorities are assigned to tasks (resp. otasks) according to a dynamic-priority scheduling policy such as Earliest Deadline First (EDF). This has been done while taking real-time constraints such precedence, strict periodicity and latency into account. As our main goal is to build a predictable RTOS, we have performed an extensive state of the art work on the mechanisms which are used to implement a scheduler. While doing so, we have come up with two main types of schedulers: the timer-driven schedulers and the event-driven schedulers . We have given the advantages and disadvantages of each type of schedulers.

Meanwhile, we continued our work on multiple latency constraints. We provided a schedulability test for dependant periodic task systems scheduled using non-preemptive policies, that must satisfy latency constraints [19] .