A material social view on data center waste heat: Novel uses and metrics

Overview

Today's data centres use substantial amounts of the world's electrical supply, and in line with circular-economy concepts much of this energy can be reused. Reuse includes heating of buildings but also commodity dehydration (wood pellets, food, fodder, seaweed), electricity production and energy storage. The paper accounts for three cases from different societal contexts (urban Malaysia, rural Costa Rica and semi-urban Sweden), drawn from field visits. In Sweden, EcoDataCenter supplies waste heat to Falu Energi & Vatten for drying wood powder that is pressed into heating pellets, a basis for its 'climate-positive' claim.

Server cooling is cited as typically adding 15-100% overhead, with hyperscale data centres achieving the lowest figures. The authors argue that PUE ignores energy source and heat reuse, and propose a composite DESS that weights sustainable energy input, compute overhead and heat reuse: DESS = 3 × EnergyIn + EnergyCompute + 2 × EnergyReuse, where each constituent lies between 0 and 1.

In DESS, EnergyIn is the sustainability-rated share of total energy (rating 1-10 covering scopes 1-3), EnergyCompute derives from the ISO PUE overhead factor, and EnergyReuse is the share of heat energy reused. The authors argue that an energy-efficient coal-powered data centre scores well on PUE but is properly penalised by DESS, and that reusing heat can bring much greater benefit than ever-smaller in-building efficiency gains.

The conclusion is that site choice and local context matter as much as in-building efficiency. Limits: the three cases are qualitative examples from field visits and DESS is a proposed metric, not standardised or widely validated; it does not give answers for every location or size. The authors note that the decade-long quest for a PUE successor has largely failed because it is not merely an engineering problem.

Key points

1. Server cooling is cited as typically adding 15-100% overhead, with hyperscale data centres achieving the lowest figures.
2. In Sweden, EcoDataCenter supplies waste heat to Falu Energi & Vatten for drying wood powder pressed into heating pellets, a basis for its 'climate-positive' claim.
3. Proposed metric: DESS = 3 × EnergyIn + EnergyCompute + 2 × EnergyReuse, where each constituent lies between 0 and 1.
4. In DESS, EnergyIn is the sustainability-rated share of total energy (rating 1-10 covering scopes 1-3), EnergyCompute derives from the ISO PUE overhead factor, and EnergyReuse is the share of heat energy reused.
5. An energy-efficient coal-powered data centre scores well on PUE but is properly penalised by DESS; reusing heat can bring much greater benefit than ever-smaller in-building efficiency gains.

Using it in design

• In data-centre siting, look beyond PUE to energy source (EnergyIn) and heat reuse (EnergyReuse), which PUE ignores.
• Investigate options for connecting waste heat to nearby users beyond district heating (for example commodity drying) at the siting stage.
• Use composite metrics such as the proposed DESS as a comparison tool, noting that it is not a standardised metric.

Who should read it

Planners and policy specialists evaluating data-centre siting, waste-heat use and sustainability metrics.

Document contents

1. 1. Introduction
2. 2. Materials and methods
3. 3. Results (3.1 Applications: commodity dehydration, energy storage solutions, marine energy; 3.2 Societies: building a sustainable future for Malaysia, drying coffee beans in Costa Rica; 3.3 Metrics: power usage effectiveness, the quest for new data center energy)
4. 4. Discussion (4.1 Key findings, 4.3 Limitations, 4.4 Prior research, 4.5 The path forward)
5. 5. Conclusion

Related standards

• ISO/IEC 30134-2 (PUE)

Read the document

Publication details

Authors

Petter Terenius; Peter Garraghan; Richard Harper

Publisher

Frontiers in Sustainability (Frontiers Media SA)

Year

2023

Volume

3

Pages

1008583

Document language

English

Source type

Scientific

Access

Open licence — downloadable

DOI

10.3389/frsus.2022.1008583

Licence

CC BY 4.0 — share with attribution

Licence evidence

doi.org

Citation