The amount of data processed and stored worldwide is growing rapidly. As a result, data centres require increasing amounts of electricity and cooling. Reducing their energy use requires detailed knowledge of where the energy goes.
In a new study, researchers Vlatko Mili膰, Patrik Thollander, Maria Andersson, Jim Enkel and Bahram Moshfegh examined energy use in a data centre room operated by Ericsson in Link枚ping. The room has a water-based cooling system that supplies cooling directly to the server rows.
The researchers have adapted a hierarchical classification system, known as a taxonomy, to the data centre environment. It makes it possible to trace energy flows from the supply of electricity and district cooling to individual processes within the data centre.
"The taxonomy works like an energy map of the data centre, showing where the energy actually goes 鈥� from electricity and district cooling to individual processes such as server operation, cooling and ventilation," says Vlatko Mili膰, Senior Lecturer and Associate Professor in Energy Systems at Link枚ping University.
More than 2.6 million data points analysed
Vlatko Mili膰 aims to provide companies with a better understanding of their energy flows, helping them make more informed decisions about energy-efficiency measures.
Ebba Nordqvist
The study is based on measurement data collected over a period of more than ten months in 2021. Data were recorded every five minutes at 29 measurement points. In total, the researchers analysed more than 2.6 million data points.
The taxonomy distinguishes between processes directly linked to data processing and those that support the operation of the room. Server operation, process cooling, pumping of cooling water and fan operation belong to the first category. Ventilation and lighting are classified as support processes because they serve the room as a whole.
Cooling and servers dominate
The data centre room used a total of 4,523 megawatt-hours (MWh) over one year. Of this, 2,154 MWh was used for process cooling and 2,135 MWh for server operation. Together, these two processes accounted for 95 per cent of the total energy use.
More than 99.9 per cent of the energy was used by processes directly linked to data processing. Ventilation and lighting together accounted for less than 0.1 per cent. Ventilation used the least energy, at 0.9 MWh.
"Because almost all the energy is used by a small number of processes, even modest efficiency improvements in server operation and cooling can noticeably reduce the data centre room鈥檚 total energy use," says Vlatko Mili膰.
Measures that make comparisons easier
The researchers have also developed indicators for measuring the energy performance of the data centre. These include energy use per square metre, peak power demand, the ratio between cooling and electricity use, and computing performance per watt.
The indicators were developed and reviewed during eleven workshops with Ericsson employees and consultants. Particular attention was given to process cooling because it accounted for almost half of the energy use and varied over time.
"The taxonomy shows where the energy is used. The indicators then allow us to monitor changes in energy performance and compare data centre rooms with similar infrastructure," says Vlatko Mili膰.
Can be adapted to other data centres
The taxonomy can be used in other data centres with similar infrastructure and access to detailed measurement data. It can also be adapted to other cooling systems by adjusting the processes and indicators to the design of the data centre.
The study covers one data centre room with a specific type of water-based cooling system. The measured energy use therefore cannot be directly generalised to the entire data centre sector. In addition, the energy use of some processes was calculated indirectly because detailed measurements were not available.
