Data-Driven Adaptive Prediction of Cloud Resource Usage

Abstract Predicting computing resource usage in any system allows optimized management of resources. As cloud is gaining popularity, the urgency accurate prediction reduced as resources can be scaled on demand. However, this may result excessive costs, and therefore there a considerable body work devoted to optimization which significantly reduce costs computing. The most promising methods employ load scaling based forecast values. quality depends method selection, different characteristics require forecasting mechanisms. This paper presents novel approach that incorporates data-driven adaptation algorithms generate short- long-term predictions enables proposed solution readjust well both temporary permanent changes. First, preliminary tests were performed yielded results – up 36% better quality. Subsequently, fully autonomous, multi-stage was proposed. evaluated using real-life historical data from various production servers. Experiment demonstrate 9.28% 80.68% when compared static algorithm selection.