Trends in data centre energy consumption under the European Code of Conduct for Data Centre Energy Efficiency (JRC108354, EUR 28874 EN)

Overview

The report introduces European data centre markets (London, Frankfurt, Paris, Amsterdam and others), explains the Code of Conduct participation scheme and the PUE metric, then analyses the registration forms submitted by companies voluntarily applying to the programme. Up to December 2016, 325 data centres from more than 120 businesses had applied for Participant status; the analysis uses a reduced sample of 289 facilities that were approved and submitted complete energy data. The data are confidential and used only for research. The report stresses that this is a unique database based on real data rather than estimates.

The method uses PUE as the ratio of total data centre input energy to IT load. Total energy is measured from the point where it is purchased (electricity, chilled water, gas, fuel oil and other purchases, including renewable generation); IT load is measured at the PDU output or at a minimum the UPS output. Because energy entries often do not cover the 12-month reporting period, annual consumption was extrapolated from the available data. PUE values below 1.0 are impossible and treated as incorrect data, and values above 3.0 are treated as outliers, leaving 268 facilities in the distribution; of 12 outliers, 3 were PUE<1.0 entries and 9 were PUE>3.0.

The main finding is an average PUE of 1.80 for the 289 approved facilities (average floor area 2,616 m2, average rated IT load 1,956 kW, average annual consumption 13,684 MWh, average set points 19.5-25 C and 35-59% RH). By application year, average PUE fell from 1.96 in 2010 to 1.64 in 2016, and every yearly average stayed below 2.0. Most facilities report PUE of 1.6-1.8, followed by 1.8-2.0, and four sites report excellent efficiency of 1.1-1.2. Medium-sized (1,000-5,000 m2) sites perform worse at 1.84 than the other size classes at 1.77, and the report concludes IT capacity has no real effect on efficiency.

The geographical analysis uses five zones: the Nordic countries average PUE 1.71, the UK and Ireland 1.83, Northern/Central Europe 1.72 and Southern Europe/Mediterranean 2.00, a difference explained by cooler ambient conditions facilitating economiser use. The free cooling review describes four basic types: direct air, indirect air, direct water and indirect water. 59% of facilities have no free cooling or did not report it. Average PUE by type ranges from 1.81 to 1.89, and the report concludes that indirect water, followed by direct water, contributes most to energy savings. It also notes that most facilities implement between 26 and 50 practices.

The limits are visible. The data are self-reported and energy entries do not always cover 12 months, so outliers were screened out and some consumption was estimated. Changes in temperature and humidity set points could not be correlated with PUE. Low PUE values for old construction periods such as 1960-1980 rest on very few facilities and are not statistically reliable. The most implemented practices are low-capital "low-hanging fruit", with about 97% producing periodic written energy reports. The study covers only facilities that joined a voluntary programme.

Key points

1. Up to December 2016, 325 data centres from more than 120 businesses applied for Participant status; the analysis covers 289 approved facilities.
2. Average PUE of the 289 facilities is 1.80; average area 2,616 m2, average rated IT load 1,956 kW, average annual consumption 13,684 MWh and IT consumption 7,871 MWh.
3. Average PUE by application year fell from 1.96 in 2010 to 1.64 in 2016, and every yearly average is below 2.0.
4. PUE below 1.0 (impossible) and above 3.0 (outlier) were excluded, leaving 268 facilities; of 12 outliers, 3 were PUE<1.0 and 9 PUE>3.0.
5. Most facilities report PUE of 1.6-1.8, followed by 1.8-2.0; four sites are at 1.1-1.2.
6. Medium-sized (1,000-5,000 m2) facilities average PUE 1.84 against 1.77 for other size classes; IT capacity is concluded to have no real effect on efficiency.
7. Average PUE by zone: Nordic 1.71, UK and Ireland 1.83, Northern/Central Europe 1.72, Southern Europe/Mediterranean 2.00.
8. 59% of facilities have no free cooling or did not report it; average PUE by free cooling type ranges from 1.81 to 1.89.
9. The report finds that of the economiser types, indirect water contributes most to savings, followed by direct water.
10. Most facilities implement between 26 and 50 practices; the most implemented are low-capital practices, with about 97% producing periodic written energy reports.

Using it in design

• Use the CoC PUE distribution (mean 1.80, falling to 1.64) as benchmark bands in the PUE calculator, showing where a design sits against CoC participants.
• Implement the same sanity rule: reject or warn on computed or entered PUE below 1.0 and flag above 3.0 as an outlier.
• Define the PUE measurement boundary as total energy at the point of purchase and IT load at the PDU output (UPS output at a minimum), the method behind the report's database.
• Show an expected PUE band by climate zone (Nordic 1.71, Northern/Central Europe 1.72, Southern Europe 2.00) and judge the economiser type together with that zone's temperature and humidity range.
• Keep free cooling options (direct air, indirect air, direct water, indirect water) as a design catalogue; in the report indirect water and direct water types are associated with the most savings.

Who should read it

Engineers and operators looking for PUE benchmark data, comparing free cooling types and climate effects, or planning Code of Conduct participation.

Document contents

1. Introduction
2. European Trends
3. Data centre markets (London, Frankfurt, Paris, Amsterdam, Other Markets)
4. European Code of Conduct for Data Centre Energy Efficiency
5. Methodology and Metrics
6. Results and Analysis
7. PUE analysis
8. Sector Distribution Analysis
9. Geographical Distribution Analysis
10. Free Cooling Technologies
11. Best practices
12. Conclusions

Related standards

• ISO/IEC 30134-2
• EN 50600-4-2

Read the document

Publication details

Authors

Paolo Bertoldi; Maria Avgerinou; Luca Castellazzi

Publisher

European Commission JRC

Year

2017

Document number

EUR 28874 EN / JRC108354

Document language

English

Source type

Government / public

Access

Open licence — downloadable

DOI

10.2760/358256

Licence

EU reuse (2011/833/EU) — reuse permitted with acknowledgement of the source

Licence evidence

op.europa.eu

Citation