Adaptive power management for energy-harvesting real-time embedded systems

Power dissipation remains a challenging issue for battery power systems such as notebook computers, PDAs, cellular phones and networked sensor nodes. This is particularly true for systems where replacing or recharging batteries manually is impracticable. One example is sensor nodes deployed in radioactive surroundings, where the energy constraint has become a main obstacle for increasing its lifespan. In order to solve the energy problem and prolong the system operating duration, a new technology called energy harvesting, also known as energy scavenging, has recently been explored and it is considered as a promising approach for battery- powered systems to achieve energy autonomy.;However, it is unclear how to perform power management for energy harvesting real-time embedded systems in an energy efficient manner. Particularly, it is unknown how to schedule real-time tasks energy-efficiently. In order to resolve these issues, we conduct substantial research. We first propose an energy-aware dynamic voltage and frequency selection (EA-DVFS) algorithm. The EA-DVFS algorithm adjusts the processor's behavior depending on the sum of the stored energy and the harvested energy in a future duration. Specifically, if the system has sufficient energy, tasks are executed at full speed; otherwise, the processor slows down task execution to save energy. By slowing down the task executing when the system is in shortage of energy, EA-DVFS algorithm saves energy and improves system performance from three aspects: (1) the more remaining energy, (2) the less deadline miss rate; and (3) the smaller minimum storage capacity for maintaining zero deadline miss rate.;In order to fully utilize task slack time for more energy savings, an adaptive scheduling and voltage/frequency selection (AS-DVFS) algorithm is then proposed. This algorithm adjusts the processor operating frequency under the timing and energy constraints based on workload information so that the better energy efficiency is achieved. In AS-DVFS algorithm, we decouple the timing and energy constraints and simplify the original scheduling problem by separating constraints in timing and energy domains. The AS-DVFS algorithm aggressively trades task slack for energy saving and further improves system performance, compared to EA-DVFS algorithm.;For an energy harvesting system, the energy storage may overflow due to incoming harvested energy. But EA-DVFS algorithm and DVFS algorithm can not handle the energy overflowing properly and energy will be wasted when overflow occurs. In order to avoid energy wastage due to the overflow of energy storage, an overflow-aware dynamic voltage and frequency selection (OA-DVFS) algorithm is presented. OA-DVFS algorithm avoids overflowing the energy storage by speeding up task execution if the system predicts the overflow will occur such that system performance is ameliorated.;An energy harvesting real-time embedded system primarily consists of energy harvesting component, energy dissipation component, and energy storage component. EA-DVFS algorithm, AS-DVFS algorithm along with OA-DVFS algorithm focus on energy efficiency study of energy dissipation component; and neither of them optimize the energy efficiency of the other two. In order to obtain extra energy savings and achieve system-wide energy efficiency, we finally present a load matching adaptive power management (LD-APM) policy for energy harvesting systems which coordinates the operations of the three components and achieves system wide energy efficiency. The LD-APM algorithm first guarantees that the energy harvesting module always operates at the maximal output power point through load matching and that real-time tasks in embedded systems are scheduled in an energy efficiently manner. It further improves the system wide energy efficiency by considering the charging/discharging overhead when deciding if the harvested energy should be used to charge the battery or directly on the circuits.