Using time-series analysis to provide long-term CPU utilization prediction

Abstract

Time-series analysis has been a recognized method of prediction for years. In computing, prediction has revolved around scheduling problems and is used in computing time to manage resources in the short-term. Long-term prediction using these methods has proven to be problematic and of little use. However, with the rise of Information Technology (IT) and the need to manage resources in a business atmosphere, the need to provide long-term resource management remains a difficult problem without a good solution. At this time, IT professionals use experience and their best judgment to manage equipment and frequently purchase systems based on maximum requirements. The need to take the knowledge gained by computer science in scheduling now needs to be expanded into the realm of IT to facilitate a more economical use of resource availability. This work is just the first step and provides one path to accomplishing long-term prediction in computing.

Using LSTM and SARIMA Models to Forecast Cluster CPU Usage

“…Currently, prediction is often based on unquantifiable industry expertise. Instead, machine learning techniques could be applied to more accurately predict hardware need [21]. Therefore, there is a need for both long term and short term CPU usage forecasts.…”