Energy-efficient Cloud Computing Technologies and Policies for an Eco-friendly Cloud Market: Final Study Report (SMART 2018/0028)

Overview

The study is a 16-month service contract carried out by Environment Agency Austria and the Borderstep Institute. It addresses the growth in energy consumption caused by the spread of cloud services and the technical and structural base behind it. It has three main parts: an overview, a technology assessment and a policy analysis. The technology side covers modelling of EU28 data centre energy demand (Task 1), a technological analysis (Task 2) and an assessment of research and technological development (Task 4). The policy side covers policy instruments, the public procurement framework and a set of recommendations.

The model shows EU28 data centre electricity use rising from 53.9 TWh/a in 2010 to 76.8 TWh/a in 2018, 2.7% of EU28 electricity demand. Growth more than offset efficiency gains in hardware, software and infrastructure. A further 21% rise to 92.6 TWh/a is forecast for 2025. Cloud's share was 10% in 2010 and 35% in 2018 and may reach 60% in 2025, with edge data centres at 12%. Northern and Western Europe held 82% of consumption in 2018, rising to 87% in 2025; Northern Europe is forecast to grow 48%, from 26.3 to 38.9 TWh/a.

Modelled average PUE is 2.03 in 2010, 1.75 in 2018 and 1.51 in the 2025 forecast. The 2018 figure splits into partial PUEs of 1.51 for cooling, 1.20 for UPS and 1.05 for other loads. EU Code of Conduct participants report average PUE of 1.71 in the Nordic countries, 1.83 in the UK and Ireland, 1.72 in Northern/Central Europe and 2.0 in Southern Europe. In Germany in 2015, PUE was 2.1 for server rooms under 100 m2 and 1.6 above 5,000 m2. If all potentials are used, consumption could return to the 2010 level; the trend case reaches 98.5 TWh/a in 2030.

The technology analysis finds that remaining efficiency potentials are shrinking towards physical limits and are harder to unlock. Easy wins include performance per watt of processors and accelerators, direct free cooling, Kyoto and evaporative cooling, stricter scaling policies in cloud management tools and energy-aware software. Warranty terms, colocation contracts and regulation slow the spread of efficient cooling, and remaining savings in modern sites above 1 MW are relatively small, so the focus is on small and older sites. Cloud also carries a high rebound risk because of fast scalability and low entry cost.

In the policy part, 195 information sources were screened. The EU has no cloud-specific procurement framework, and uptake of the EU Code of Conduct and the EU GPP criteria for data centres lags in Member States. The study offers seven summary recommendations for RTD policy and detailed policy recommendations. Case studies include the Frankfurt LOEWE-CSC cluster at PUE 1.076 and the Telia Helsinki heat reuse project, which lifts cooling water from about 30 °C to 70-80 °C with two-stage heat pumps. It is a policy and market study, not a design standard, and its 2020/2025 values are forecasts made in 2020.

Key points

1. EU28 data centre electricity use rose from 53.9 TWh/a in 2010 to 76.8 TWh/a in 2018, about 2.7% of EU28 electricity demand in 2018.
2. A 21% rise to 92.6 TWh/a is forecast for 2025; cloud's share was 10% in 2010 and 35% in 2018 and is forecast at 60% in 2025.
3. Edge data centres are expected to account for 12% of EU28 data centre energy consumption by 2025 (an edge site is defined as a remotely operated data centre of up to 10 racks).
4. Northern Europe is forecast to rise 48% between 2018 and 2025, from 26.3 to 38.9 TWh/a; Northern and Western Europe held 82% of consumption in 2018 and 87% in 2025.
5. EU Code of Conduct participants report average PUE of 1.71 in the Nordic countries, 1.83 in the UK and Ireland, 1.72 in Northern/Central Europe and 2.0 in Southern Europe.
6. German PUE in 2015: 2.1 for server rooms under 100 m2, 1.9 for 100-500 m2, 1.7 for 500-5,000 m2, 1.6 above 5,000 m2.
7. Modelled EU28 average PUE: 2.03 in 2010, 1.75 in 2018, 1.51 in 2025 (forecast); 2018 partial PUEs are 1.51 cooling, 1.20 UPS and 1.05 other.
8. If all technical potentials are exploited, consumption could be reduced to the 2010 level; the trend case reaches 98.5 TWh/a in 2030.
9. Remaining savings are small in modern sites above 1 MW; easy wins lie in small and older sites, software and cloud management tools.
10. Heat reuse case: cooling water at about 30 °C is raised to about 70-80 °C by two-stage heat pumps; the Frankfurt LOEWE-CSC cluster reached PUE 1.076 and saves 3.37 GWh a year.

Using it in design

• Use the regional and size-based values for first-pass PUE assumptions, but treat them as 2018-and-earlier averages and as historical context, not as a current target.
• Start a load budget from the partial-PUE split (cooling, UPS, other): in the 2018 model 1.51, 1.20 and 1.05; calculate each component's share separately for your own design.
• In small and older facilities consider existing technologies such as direct free cooling, Kyoto or evaporative cooling, and reflect in the specification that warranty terms and colocation contracts can block them.
• When planning heat reuse, verify early the heat-pump lift scenario (about 30 °C to 70-80 °C) and the existence of a district heating offtaker.
• For public or public-linked projects, reference the EU GPP criteria for data centres and the EU Code of Conduct in the specification; the study notes their national uptake lags.

Who should read it

Designers, planners and policy teams who want to understand data centre energy demand, regional PUE differences and the EU policy context.

Document contents

1. Abstract
2. Executive Summary
3. Introduction
4. Use of Cloud Computing - Current Situation and Framework Scenario
5. Task 1 - Analysis and Modelling across the EU28
6. Task 2 - Technological Analysis
7. Task 3 - Policy Analysis
8. Task 4 - Assessment of Research and Technological Development (RTD)
9. Task 5 - Green Public Procurement for Cloud Computing Services
10. Task 6 - Policy Recommendations
11. Annex 2 - Modelling the Energy Consumption of Data Centres in the EU28
12. Annex 3 - Screened Initiatives from Task 5
13. Annex 5 - All Recommendations for RTD

Read the document

Publication details

Authors

F. Montevecchi; T. Stickler; R. Hintemann; S. Hinterholzer

Publisher

European Commission, DG Communications Networks, Content and Technology (DG CONNECT)

Year

2020

Document language

English

Source type

Government / public

Access

Open licence — downloadable

DOI

10.2759/3320

Licence

CC BY 4.0 — share with attribution

Licence evidence

doi.org

Citation