Short-Term Prediction Model to Maximize Renewable Energy Usage in Cloud Data Centers

The increasing demand for services offered by cloud providers results in a large amount of electricity usage by their data center sites and a high impact on the environment. This has motivated many cloud providers to move towards using on-site renewable energy sources to partially power their data centers using sustainable sources. This way, they can reduce their reliance on brown electricity delivered by off-site providers, which is typically drawn from polluting sources. However, most sources of renewable energy are intermittent and their availability changes over time. Therefore, having short-term prediction helps the cloud provider to make informed decisions and migrate the virtual machines (VMs) between data center sites in the absence of the renewable energy. In this chapter, we propose a short-term prediction model using Gaussian mixture model (GMM). The model uses the previously observed energy levels to train itself and predict the energy level for many-steps ahead into the future. We analyzed the accuracy of the proposed prediction model using real meteorological data. The experiment results show that the GMM model can predict up to 15 minutes ahead into the future with nearly 98% accuracy around ±10% of the actual values. This helps the cloud provider to perform online VM migration with performance close to the optimal offline algorithm, which has the full knowledge of renewable energy level in the system. Moreover, the accuracy of the model has been verified using the workload data from Amazon biggest region in US East (N. Virginia). However, due to the confidentiality of that data set, we only rely on the results of the carried experiments using real meteorological renewable energy traces. Keywords— Cloud computing, Green computing, Renewable energy, Data center, Renewable Prediction