Energy-Aware Resource Reservation for Multicore Architectures

Power consumption is a key consideration when designing real-time embedded systems. In this work, we address the problem of energy minimization for the resource reservation algorithm Greedy Reclamation of Unused Bandwidth (GRUB) for multicore platforms. Our strategies mix Dynamic Voltage and Frequency Scaling (DVFS) and Dynamic Power Management (DPM) popular techniques for reducing total energy consumption. We propose two different approaches: (i) the first approach, Frequency First Approach (FFA), prioritizes reducing dynamic energy consumption, while the second one, Core State First (CSF), focuses on static energy consumption. We demonstrate that the two approaches are incomparable. We prove that with our techniques, temporal isolation is not compromised, and we can further reduce energy consumption compared to GRUB-PA (Power Aware), a resource reservation mechanism based on GRUB that aims to reduce only dynamic energy consumption. We evaluate the performance on a large set of synthetic task sets. In addition, we review the implementation of a new opensource simulator, which has been specifically designed to quickly prototype DVFS- and DPM-based schedulers.