Configuration-aware and QoS-aware Task Scheduling in Real-Time Adaptive Embedded Systems

In order to cope with increase in demand for embedded applications under resource and power constraints, there is an increasing trend on incorporating system-level flexibility and runtime adaptation in embedded systems. Although a complete processor-based implementation provides ultimate flexibility, usually computationally intensive processing mandates hardware support in addition to the processors. However, the main drawback of reconfigurable hardware platform is delay and energy overhead during system reconfiguration. Ignoring the time and energy required for reconfiguration may lead to performance degradation and deadline misses.;In this thesis we provide different real-time scheduling techniques. First, we propose the first real-time scheduler that is aware of the transition time overhead due to configuration of underlying hardware. We consider a heterogeneous multi-processor system and there are data dependencies among tasks and communication overhead time during task execution.;In many applications deadline miss rate is not an accurate metric to model the application quality but the distribution of deadline misses is the main factor. We use a well-known Weakly Hard System QoS model to define the acceptable deadline miss distribution. We provide an online scheduler that considers the time overheads among task switching while the deadline misses caused by time overheads follow a predefined pattern.;Next, we focus on energy consumption of the system and provide a scheduler that considers the time and energy overhead during power management. In our work deadline miss rate is minimized as the primary objective and the energy consumption of the system is minimized by means of voltage scaling as the secondary objective.;Many adaptive systems need to adapt to varying delay and/or energy constraints such as fluctuation of energy source when system is powered by renewable energies like solar energy. In order to ensure the scheduler can handle the dynamicity in availability of the energy source, we exploit the QoS constraint so the application quality is not violated during job dropout when our power management scheme adjusts the energy consumption according to energy availability.;Each of these schedulers is accompanied by an extensive experimental evaluation that represents the effect of using the proposed algorithms.