Schedulability Analysis of General Task Model and Demand Aware Scheduling in Mixed-Criticality Systems

Nowadays, the embedded systems are undergoing an unprecedented trend towards integrating components or tasks of different criticality levels onto a common computing platform, as the task integration can reduce the “SWaP” (Size, Weight, and Power) related costs. Those systems, commonly referred to mixed-criticality systems, consist of functionalities with two or more distinct criticality levels, e.g. safety criticality and mission criticality; it is of vital importance for systems to meet tasks’ requirements corresponding to their own criticality levels. To achieve the mixedcriticality guarantee, less criticality tasks are assumed to be degraded or dropped once they overrun their given execution budgets, and tasks are commonly considered being activated sporadically because sporadic activation model is easier to be analyzed than other complex activation models like sporadic burst model. These two assumptions are however pessimistic because in fact the system may have some slacks to allow tasks overrun; and today’s real-time systems are embracing a growing variety of activation patterns whose activation features may deviate a lot from the sporadic activation assumption. In this thesis, we focus on adaptively postponing the mode-switch online by exploiting the system’s static and runtime slacks; and we address the problem of schedulability analysis towards a more general task model in mixed-criticality systems. Specifically, we first propose an on-the-fly fast overrun budgeting mode-switch scheme to online postpone the mode-switch. Then, we extend the current sporadic task model to the arbitrary activation task model (arrival curve) in mixed-criticality systems, based on which we furthermore propose an approach that can adaptively shape the arriving events of task activation. We also present a case study showing the application of some basic mixed-criticality scheduling concepts in an autonomous driving system.