Cloud Workload Forecasting with Holt-Winters, State Space Model, and GRU

Abstract

Cloud computing has become increasingly important in the modern world, and the ability to accurately predict cloud computing workloads is essential for optimizing resource utilization, cost efficiency, performance improvement, and service availability. This study applied three models to predict cloud workloads: Holtwinters Exponential Smoothing, Exponential Smoothing State Space Model (ETS), and Gated Recurrent Unit (GRU). The Holtwinters model had the lowest errors and was the best-performing model among the three. The Holtwinters Exponential Smoothing model was used to predict cloud workloads. The model used a combination of exponential smoothing techniques and a linear trend to forecast future values. The model was evaluated on both short-term and long-term predictions. On short-term predictions, the model had a mean absolute error of 5.3%, which was lower than the errors for the ETS and GRU models. On long-term predictions, the Holtwinters model had a mean absolute error of 8.4%, which was also lower than the errors for the other two models. The results of this study demonstrate the importance of accurately predicting cloud computing workloads. The Holtwinters Exponential Smoothing model was found to be the best-performing model among the three models evaluated, with the lowest errors. This model can be used to make accurate predictions of cloud workloads, which can be used to optimize resource utilization, cost efficiency, performance improvement, and service availability.