Bears: Building Energy-Aware Reconfigurable Systems

Energy efficiency has developed to one of the most important non-functional system properties. One keystone to building an energy-efficient system is the right system configuration, which is tailored to the currently running application and hardware. Finding such a right system configuration manually, however, is a complex and often unfeasible task due to the vast configuration space on the one side and the required hardware and application knowledge on the other side. This paper presents and refines an approach to automatically identify and select energy-efficient configurations in re-configurable systems. The approach relies on different machine-learning techniques and achieves energy efficiency improvements of up to 10.8 % out of 13.3 % by automatically adapting the system configuration on a Linux system. Additionally, we analyse the application knowledge required for selecting the configuration and make a proposal how to generate sufficient training data.