Workload Prediction for Cloud Resource Provisioning using Time Series Data

Abstract

Analyzing and interpreting the fluctuating real-time workload by cloud providers is a challenging task for providing cloud services to end-users. Allocating computing resources (e.g., container machine) for running applications at right time solves the problem of the unavailability of service, maximizes response time, and minimizes throughput. Predictive analysis of data is important to identify the future trends and it also enables cloud providers to allocate the computing resources according to the demand for managing the workload. Predictive analysis has applications in different fields such as weather forecasting, stock price prediction, bio-informatics, and traffic over the Internet network (Cloud Computing). They use this analysis to assess the risk and predict future effect to avoid the different types of losses in the cloud computing area such as energy consumption, service delay, and customer loss. One of the methods to do predictive analysis is by using time series data. This paper presents the analysis of time series forecasting using the traditional classical Model (ARIMA) to predict the real-time fluctuating workload over the network. The predicted accuracy of model is measured using three metrics—RMSE, MSE, and MAE.