Environment data processing for a data centre

Abstract

There are rising concerns over the carbon footprint of data centre (DC) in Singapore as their total electricity consumption has increased from 5.3% in 2019 to 7% in 2020. This figure is projected to increase, in line with the growing demand for DC. In this study, we will focus on the DC’s information technology (IT) system which is one of the major energy consumers. Five types of predictive models: Multilayer Perceptron, Linear Regression, Decision Tree, Support Vector Regression and Stacking-based Ensemble were developed to predict total IT power consumption using IT facilities' operating conditions and meteorological parameters as inputs. This research aims to investigate the relationship between a DC’s IT power consumption with its IT facilities and environmental factors outside the DC. This paper also presents a feature importance analysis and a detailed comparison of the performance of different models. The results of the feature importance analysis indicate that CPU utilization is the most significant factor that will affect the total IT power consumption in the DC. As for the model evaluation, the Stacking-based ensemble performs the best. Based on the Stacking-based ensemble model, a set of recommendations on the most optimal IT operating conditions was made.