Energy-harvesting-aware federated scheduling of parallel real-time tasks

This paper presents HEARTS, a multicore energy scheduling approach utilizing a federated strategy designed for parallel real-time tasks of significant computational demands in embedded systems deployed in environments of unreliable power sources like surveillance and intelligent city infrastructures. HEARTS, specialized for high-utilization parallel tasks, divides the scheduling horizon into multiple windows, so that it dynamically allocates the cores to the tasks based on the energy availability. It introduces two schedulers based on the first-fit and last-fit approaches. We demonstrate the optimality of the last-fit-based scheduler when the battery capacity exceeds some specific threshold; further, we show scenarios where the first-fit-based scheduler performs better under lower capacities. Simulations using two setups—one with random harvested energy and task parameters, and the other with real solar energy and benchmark tasks—show a maximum deviation of 19.05% and 21.34% from two theoretical optimal solutions, respectively, and a substantial improvement of 28.24% over the energy partitioning approach.