Engineering energy-aware web services toward dynamically-green computing

With the emergence of commodity computing environments (i.e. clouds), information technology (IT) infrastructure providers are creating data centers in distributed geographical regions. Since geographic regions have different costs and demands on their local power grids, cloud computing infrastructures will require innovative management procedures to ensure energy-efficiency that spans multiple regions. Macro-level measurement of energy consumption that focuses on the individual servers does not have the dynamism to respond to situations where domain-specific software services are migrated to different data centers in varying regions. Next-generation models will have to understand the impact on power consumption for a particular software application or software service, at a micro-level. A challenge to this approach is to develop a prediction of energy conservation a priori. In this work, we discuss the challenges for measuring the power consumption of an individual web service. We discuss the challenges of determining the power consumption profile of a web service each time it is migrated to a new server and the training procedure of the power model. This potentially promotes creating a dynamically-green cloud infrastructure.