Work-in-Progress: System-wide DVFS for real-time systems with probabilistic parameters

Overview & Synopsis

Power consumption and the ever-increasing service demand are key issues that real-time embedded systems are facing nowadays. Approaches based on Dynamic Voltage and Frequency Scaling (DVFS) are capable of reducing the energy consumed by processors while still guaranteeing real-time constraints. In this paper we present short-comes on the state-of-the-art techniques that need to be leveraged in order to be more effective when reducing energy consumption: our experimental results clearly show that tasks' execution time is not proportional to processor speed, and that DVFS techniques could also be applied on other components like bus and memory without compromising time constraints. We also discuss the applicability of real-time probabilistic systems and DVFS, and argue that by adopting a frequency-aware model, we can (i) capture more detailed behaviors of tasks w.r.t. processor speeds and (ii) apply DVFS techniques to gain in energy consumption.