Section: New Results

Learning Power Models for Distributed and Virtualized Environments

Energy efficiency is a major concern for modern ICT infrastructures. The a priori estimation of the level of energy consumed by a given service is a difficult problem given the intricate nature of hardware and software that are involved. Consequently, even before considering saving, measuring the exact amount of energy consumed by a given software service or process is required. Over the last few years, a dozen of ad hoc power models have been proposed in the literature. Nevertheless they cannot cope with the constant evolution of software and hardware architecture. We have therefore defined and implemented a toolkit that automatically learns the power models of a given architecture, independently of the features and the complexity it exhibits. This toolkit considers traditional distributed environment as well as virtualized, cloud-based ones. This result has been achieved in the context of the PhD thesis, defended in November 2016, of Maxime Colmant [11].