Predictive Cloud resource management framework for enterprise workloads

The study proposes an innovative Predictive Resource Management Framework (PRMF) to overcome the drawbacks of the reactive Cloud resource management approach. Performance of PRMF was compared with that of a reactive approach by deploying a timesheet application on the Cloud. Key metrics of the simulated workload patterns were monitored and analyzed offline using information gain module present in PRMF to determine the key evaluation metric. Subsequently, the best-fit model for the key evaluation metric among Autoregressive Integrated Moving Average (ARIMA) (1 ⩽ p ⩽ 4, 0 < d < 2, 1 ⩽ q ⩽ 4), exponential smoothening (Single, Double & Triple) and Hidden Markov Model present in the PRMF library were determined. Best-fit model was used for predicting key evaluation metric. During real time, the validation module of PRMF would continuously compare the actual and predicted key evaluation metric. Best-fit model would be re-evaluated if 95% confidence level of the predicted value breaches the actual metric. For experiments performed in the current study, Request Arrival and ARIMA (2, 1, 3) were found to be the key evaluation metric and the best-fit model respectively. Proposed predictive approach performed better than the reactive approach while provisioning/deprovisioning instances during the real time experiments. Keywords: Cloud computing, Predictive modeling, Resource management, Enterprise workloa