Modeling and Forecasting Http Requests-Based Cloud Workloads Using Autoregressive Artificial Neural Networks

Abstract

For the proactive scalability/elasticity of a cloud computing platform, cloud workload forecasting is a key issue and has been studied by using conventional time series (TS) methods, such as autoregressive integrated moving average (ARIMA) models. In this paper, to modeling and forecasting http requests-based cloud workloads, we propose using autoregressive artificial neural networks (autoregressive ANNs), which are a machine learning (ML) technique widely used in diverse research problems and practical applications. Based on our empirical evaluation, compared with other representative TS methods and ML techniques, the proposed approach can achieve highest accuracy and stability for the addressed problem although it also consumes slightly more time to yield a predictor.