Real Time Non-Linear Cloud Workload Forecasting using the Holt-Winter Model

Abstract

Cloud computing facilitates on-demand services to its users. This elastic nature of the cloud environment-attracts more users ranging from individual ones to big enterprises. The increasing use of cloud resource necessitates the need for better and efficient resource management techniques for the cloud. In this paper, we focus on one such key resource management technique which is demand forecasting. It is important to forecast the demand because, under provisioning of resources results in violation of the service level agreement with the cloud users. Contrarily, over provisioning of resources results in resource wastage such as network and power usage. Therefore, we propose a seasonality based demand forecasting system. This system uses Holt-Winter's time series approach for demand forecasting in the real time non-linear cloud environment. Then we analyze the performance of our model for short term and long term forecasting. We also measure the accuracy of our proposed system using MAPE values.