Séminaires
Séminaire du 14 novembre 2014
Lieu
Telecom ParisTech, amphi Grena, - comment y aller
Programme
14h00-15h00: Mathieu Jan, CEA, "Real-Time Scheduling for time-triggered mixed criticality systems",
Abstract: 
The Time-Triggered (TT) paradigm is one solution to build hard real-time systems subject to certification constraints. In the TT paradigm, the tasks are triggered by the advancement of time. While the TT paradigm provides a predictable execution, the static scheduling approach is considered to lead to a poor resource utilization in the average case. The design of TT systems and the associated schedulability demonstrations must indeed be performed in the worst-case situation. These unused processing capabilities motivate the adding of Mixed-Criticality (MC) scheduling techniques within TT systems. This is achieved by hosting low-criticality tasks on a same hardware architecture in addition to the classical high-critical tasks, when considering two-criticality levels. The goal of MC scheduling is to increase the schedulability of the low-criticality tasks, while still guaranteeing in the worst-case scenario the schedulability of the high-criticality tasks. In this talk, after an introduction on the TT paradigm, we will show two different approaches for extending TT systems in order to cope with MC scheduling requirements.
15h00-15h30 : Romain Gratia, Telecom ParisTech, "Adaptation of real time scheduler RUN to mixed criticality".
Abstract: Mixed-criticality scheduling provides a method to improve CPU usage between tasks, assuming the task set can be decomposed in two subsets of different importance with respect to system requirements. Then, considering that tasks have different CPU consumption modes, the core idea is to accept that when tasks enter higher CPU consumption modes then lesser criticality tasks may no longer be scheduled. Hence, it allows deploying more tasks on fewer processors accepting that in the worst mode only most critical tasks will respect their deadlines.
This presentation details a method to adapt RUN, an optimal scheduling algorithm for multicore processor scheduling, to cope with mixed-criticality task sets. The RUN approach deeply relies on a data structure called reduction tree computed off-line to elect tasks. We adapt this tree to define a scheduler compatible with mixed-criticality requirements. The scheduler requires to be able to bind so called low task to scheduling servers. Finding which task can be bound to servers could be solved through reachability analysis on automata representing the binding problem.