BGW: New task model for overloaded real time systems: Ordonnancement monoprocesseur
Mohamed Ould Sass · HAL (Le Centre pour la Communication Scientifique Directe) · 2015
Real-time embedded systems are found in variousapplication domains. They have to offer an increasingnumber of functionalities and to provide the highestQuality of Service despite possible failures due tofaults or processing overloads. In such systems,programs are characterized by upper bounds onfinishing times and the QoS is assessed by the ratio ofsuccessful deadlines. In this thesis, we deal with thisissue. We focus on a uniprocessor architecture in theframework of a firm real-time application that acceptsdeadline missing under some specified limits. Tasksare assumed to be periodic. Our first contribution liesin the proposition of a novel model for tasks which iscalled BGW model. It is drawn from two approachesrespectively known as the skip-over model and theDeadline Mechanism. The first one provides timingfault-tolerance through passive dynamic softwareredundancy with two versions. The second one copeswith transient processing overloads by discardinginstances of the periodic tasks in a controlled andpre-specified way. We give a feasibility test for thismodel. In a second part, we describe the behavior ofdynamic priority schedulers based on EDF (EarliestDeadline First) for BGW task sets. A performanceanalysis is reported which is mainly related to QoSevaluation and measurement of overheads (complexityof the scheduler). The following contribution concernsmore sophisticated schedulers that permit to enhancethe QoS as to improve service balancing.