Fixed Priority Scheduling Strategies for Ambient Energy-Harvesting Embedded Systems

Abstract

The new generation of embedded systems will have the capability to harvest energy from the environment. The electrical energy which is available to power these devices changes over time and is limited by the size of the energy storage unit such as battery or capacitor and the size of the harvester such as a solar panel. In order to cope with this limitation, the system has to dynamically decide when to be active and when to sleep in order to provide the best quality of service without wasting the harvested energy. In this paper, we study this problem for a uniprocessor architecture where periodic tasks have to execute with deadline constraints according to a preemptive fixed priority rule. We evaluate and compare several scheduling approaches by means of simulation.

Periodic charging scheme for fixed-priority real-time systems with renewable energy

“…The total power P σ0 consumed by the system in the active state is the sum of processor power and total power consumption of the specific subsets of peripherals in use by the running task; thus, at any time, P CP U ≤ P σ0 ≤ P CP U + P dev . As in [7], [8], we do not assume DVFS capability in the system; i.e., task execution takes place at a constant frequency level.…”

“…In systems with renewable energy, the concept of feasibility is extended to consider also battery (or, energy) failures ( [7], [8]). Specifically, in addition to guaranteeing task completions no later than their respective deadlines, in order to ensure feasibility, the algorithm must also guarantee that the battery level never drops below a certain threshold E low : ∀t, E(t) > E low .…”

“…Moreover, thanks to its design principles, it explicitly considers the time/energy overheads involved in the processor state transitions, through the explicit analysis of the break-even times. In contrast, the energy-greedy algorithms (e.g., P F P st [7]) involve online computation of the existing slack to re-charge the battery, which is, in general, of pseudopolynomial complexity. Similarly, the computation-greedy algorithms (e.g., the theoretically optimal P F P ASAP ) result in very frequent invocation and processor state transitions with prohibitive costs on real systems that have non-zero transition overheads.…”

“…For fixed-priority real-time embedded systems which are more common in practice, the two well-known algorithms are P F P st (also called EDeg) [7] and P F P asap [8], that represent energy-greedy and computation-greedy algorithms, respectively. In particular, P F P asap is shown to be optimal in [8]: any task set that can be feasibly scheduled by any other fixed-priority energy-harvesting scheduling algorithm can be also scheduled by P F P asap .…”

The Fixed Priority Scheduling Problem for Energy Harvesting Real-Time Systems

“…Recent work shows that this hypothesis is not realistic [3]. Finally, the problem was addressed by Chetto et al in [4], [5]. A clairvoyant algorithm and several non clairvoyant heuristics was proposed.…”