Optimized DVFS Policies for Multi-Core Real-Time Embedded Systems

Energy consumption reduction is central to the design of modern embedded systems. Consequently, modern processors are equipped with software controlled mechanisms such as DVFS to allow the operating system to reduce the power consumption while meeting the required performance. This work presents two new policies for DVFS configuration which take into account the limited core DVFS range. Based on these policies we develop an OS scheduling algorithm for single and multiple simultaneous multi-threading (SMT) cores based embedded systems employing periodic real-time application. For a single SMT core, our new DVFS policies and corresponding scheduling algorithms consume 10%-50% less energy than current common algorithms. For CMP, we show that the optimal number of active cores can be derived by our analysis. The resulted scheduling algorithm can reduce the energy consumption by at least 10% when applied for realtime applications.