Energy Efficient Scheduling Algorithm For Real Time Embedded Multiprocessor Systems

Abstract

Multiprocessor systems provide high-performance scientific computing. These newlineare materialized to tackle sophisticated real-time applications. This carried a newlinehypothetical change in real-time systems research. Computational demands of newlinereal-time applications are growing continuously, which increases the necessity newlineof energy managing techniques. The main challenge here is the optimization newlineof energy consumption without violating timing restrictions. newlineThis thesis tackles the issues of energy efficiency and scheduling together to newlineprovide better performance for multiprocessor systems. It presents novel newlineenergy-aware schedulers for hard real-time systems. These schedulers adopt newlineglobal scheduling approach with preemption mechanism under homogeneous newlinemultiprocessor platforms. Our first contribution is E-token Energy Aware newlineProportionate Scheduler (EEAPS), which allows tasks among processors newlinebased on their energy consumption through auctioning mechanism. Our newlinesecond contribution is Thermal Energy Aware Proportionate Scheduler newline(TEAPS), which scales down the speed of the processor to diminish thermal newlineleakage energy dissipation on the processor. Our third and final contribution is newlineEnergy Aware Proportionate Slack Management scheduler (EAPSM). It newlineefficiently manages the slack through reclamation and shutdown mechanisms. newlineThese approaches minimize static power consumption and dynamic power newlineconsumption of the multiprocessor systems. It is integrated with task processor newlineaffinity and processor frequency affinity metrics to minimize the task migration and frequency switching overheads. It further increases the energy newlinesavings, which improves the battery life of multiprocessor systems. Results newlineillustrate the better performance of the recommended algorithm compared newlinewith the available optimal algorithms. newlineEAPSM reduces the energy consumption of 2 to 18% comparing with Early newlineRelease fair scheduler with suspension on multiprocessors (ESSM) and 20 to newline33% comparing with Hybrid scheduler. It reduces task migration overhead of newline2 to 27% with respect to