Simultaneously exploiting dynamic voltage scaling, execution time variations, and multiple methods in energy-aware hard real-time scheduling

In this paper we present a novel energy-aware scheduling algorithm that simultaneously exploits three effects to yield energy savings. Savings are achieved by using dynamic voltage scaling (DVS), flexibility provided by slack time, and by dynamically selecting for each task one of several alternative methods that, can be used to implement the task. The algorithm is split up in two parts. The first part is an off-line optimizer that prepares a conditional scheduling precedence graph with timing conditions defining for any decision point in time which branch should be taken due to the assumed elapsed time. The second part is an efficient runtime dispatcher that evaluates the timing conditions. This separation of optimization complexity and runtime efficiency allows our algorithm to be used on mobile devices having only little energy resources and being driven to the edge by the applications that run on them, e.g., creating a video on a mobile phone. We show that our approach typically yields more energy savings than worst-case execution time based approaches while it still guarantees all real-time constraints. Our application model includes periodic non-preemptive tasks with release times, hard deadlines and data-dependencies. Multiple methods having different execution times and energy demands can be specified for each task, and an arbitrary number of processor speeds is supported.