by Cyta Hellas
Summary
This experiment looked to find the appropriate number of measurements needed for the evaluation of power consumption profiles of the data centre (DC) and their correlation with external weather conditions.
As a large portion of the energy consumption of DCs is driven towards cooling the IT infrastructure it is of great interest to investigate the factors that affect it. Air cooling systems that bring air from the external premises are usually deployed for the cooling of the interior of the DCs. This project focused on weather data (e.g. temperature, humidity, wind, atmospheric pressure) collected by weather station sensors in order to examine the correlation with the energy consumption of the DC. Additionally, we modelled the power consumption related to weather trends in order to effectively forecast the energy consumption and validated this through live measurements from the RealDC testbed.
Our analysis showed that only certain weather features have significant impact on the energy consumption. We then used the correlated data to build a forecast model using linear regression algorithm. The experimental results showed that the forecast energy consumption manages to predict the energy consumption from the weather conditions with adequate accuracy. These results are indicative as they could provide data centers operators and power distribution companies with tools to manage their power needs distribution.
Hence, a key outcome of the DC-IoT experiment is a web application that calculates the forecast values of the energy consumption of a data center, given a weather forecast for specific physical parameters, like the air temperature and the atmospheric pressure.
Our future work will include using weather data of longer periods of time to provide a more accurate forecast of the energy consumption.
Demo