Run-Time Management for Multicore Embedded Systems With Energy Harvesting

In this paper, we propose a novel framework for runtime energy and workload management in multicore embedded systems with solar energy harvesting and a periodic hard real-time task set as the workload. Compared with prior work, our framework makes several novel contributions and possesses several advantages, including the following: 1) a semidynamic scheduling heuristic that dynamically adapts to runtime harvested power variations without losing the consistency of periodic tasks; 2) a battery-supercapacitor hybrid energy storage module for more efficient system energy management; 3) a coarse-grained core shutdown heuristic for additional energy saving; 4) energy budget planning and task allocation heuristics with process variation tolerance; 5) a novel dual-speed method specifically designed for periodic tasks to address discrete frequency levels and dynamic voltage/frequency scaling switching overhead at the core level; and 6) an extension to prepare the system for thermal issues arising at runtime during extreme environmental conditions. The experimental studies show that our framework results in a reduction in task miss rate by up to 70% and task miss penalty by up to 65% compared with the best known prior work.