Exploring the trade-off between performance and energy consumption in cloud infrastructures

Abstract

Emerging fog and mobile edge computing paradigms will create distributed pervasive virtualization environments, where computing, storage, and networking resources will be deployed at the network boundary in a capillary way. To effectively tackle the large dynamic fluctuations in workload engendered by user-centric services, effective energy management schemes must be in place to modulate power consumption according to the actual processing load in each installation. In this respect, service orchestrators and multi- and cross-cloud energy management systems need proper tools to understand how power consumption would change with different placement decisions, both in the single as well as in federated clouds. In this paper, we describe a framework for exploring the trade-off between performance and energy consumption. Our work builds on the availability of both resource usage and power consumption measurements in Cloud Management Software, and makes proper correlation between these values to effectively support energy-efficiency strategies. We describe the implementation of energy monitoring framework in OpenStack, which leverages available features in the Ceilometer component.