Thermal-aware semi-dynamic power management for multicore systems with energy harvesting

In this paper, we focus on power and thermal management for multicore embedded systems with solar energy harvesting as the power source and a periodic hard real-time task set as the workload. We design a novel semi-dynamic scheme, which reschedules tasks at the beginning of specified time epochs. By rejecting job instances of certain tasks until the next rescheduling point, our scheduler dispatches a subset of tasks that comply with the predicted energy budget and thermal conditions. Our approach reacts to run-time energy harvesting power variation without losing the consistency of the periodic task set, which helps to scale processor speed evenly by utilizing slack time efficiently without the need for complex slack reclamation algorithms, as in prior work. When applied to a multicore platform, our approach offers a chance to shut down cores and reassign tasks for superior energy efficiency. As a result, experimental results show up to 70% miss rate reduction compared to prior work. Unlike any prior work, our approach also integrates thermal management to reduce peak temperature while minimizing miss rate for energy harvesting embedded systems.