Abstract
Improving energy efficiency of data centers is an important research challenge. Web services are an important part of data centers' workload, and a large contributor to their energy footprint. This paper contributes an approach that, leveraging statistical data over web services usage patterns, dynamically predicts the resources required by the web service application. Our framework, SOPRA, uses these predictions to constantly adapt the allocation of resources to minimize the energy utilization of the data center. We demonstrate the viability of our approach by executing SOPRA over a synthetic workload. We compare the energy savings achieved by SOPRA with the traditional over allocation strategy and with the saving achievable by using a static predictor. Furthermore, we show how different service level agreements (SLA) influence the ability to save energy. The results of our experiments show that, with our workload, we can save up to 52.49% of energy over the over-allocation approach while a static prediction can only achieve a 44.78% saving. Moreover, our results show that the SLA has a high impact on energy savings. Using a more demanding SLA, the energy saving SOPRA was able to achieve was only 28.29%.