A Systematic Review of Energy Efficiency Metrics for Optimizing Cloud Data Center Operations and Management

Overview

Cloud data centres are an essential component of the infrastructure powering digital life, and their energy efficiency is a top priority because they consume significant electricity, contributing to operating cost and environmental impact. Energy-efficiency metrics measure how effectively a data centre uses energy for computing versus cooling and other overheads. The review covers more than 25 metrics and more than 250 literature references; for each metric it gives a definition, formula, units, best value and qualitative interactions with other metrics, then lists challenges and research directions.

Within a cloud data centre, IT equipment accounts for roughly 60% of electricity use, while cooling and power-delivery systems account for another 30-40%. PUE, DCiE and CUE give only a partial view because they usually exclude software-related inefficiency, dynamic workloads and grid-level CO2 intensity; finer-grained metrics such as Server PUE (SPUE) have been proposed but are not yet standardised. Listed PUE weaknesses: it can be manipulated (for example by shutting off unused equipment), it is often reported under optimal conditions, and it looks only at infrastructure energy, ignoring IT utilisation.

CUE depends on CO2 data from energy providers that is often unavailable or approximate, and it leaves out offsets and lifecycle (embodied) emissions; the authors suggest standardising Scope 2 reporting and using time-stamped carbon intensity and RECs. Distributed UPS designs (a battery cabinet serving several racks with DC fed to dual-input server PSUs, or a battery inside each server) avoid the AC/DC/AC double-conversion losses of a centralised UPS. Data Center Power Density (DCPD, W/m²) is linked with thermal hot spots and cooling bottlenecks and is said to correlate negatively with the Rack Cooling Index (RCI).

Case examples: Google reported a 2023 fleet-average annual PUE of 1.10 against an industry average of 1.58; Microsoft reports PUE and WUE together and stresses climate and humidity as location-specific variables. Limits: the analysis is descriptive rather than quantitative; its main value is as a taxonomy and a list of known metric weaknesses.

Key points

1. Within a cloud data centre, IT equipment accounts for roughly 60% of electricity use, while cooling and power-delivery systems account for another 30-40%.
2. Metrics are split into two classes: IT-related (APC, CPE, DWPE, EWR, ITEE, OSWE, PpW, ScE, SPUE, SWaP) and non-IT-related (CADE, DCA, DCeP, DCgE, DCPD, DCPE, DC-FVER, DH-UE, DH-UR, EBS, H-POM, PDE, PEsavings, SI-POM, TUE, PUE, DCiE, CUE).
3. PUE, DCiE and CUE give only a partial view because they usually exclude software-related inefficiency, dynamic workloads and grid-level CO2 intensity; finer-grained metrics such as SPUE are not yet standardised.
4. PUE weaknesses: it can be manipulated (for example by shutting off unused equipment), it is often reported under optimal conditions, and it looks only at infrastructure energy, ignoring IT utilisation.
5. CUE depends on CO2 data from energy providers that is often unavailable or approximate, and leaves out offsets and embodied emissions; the authors suggest standardising Scope 2 reporting and using time-stamped carbon intensity and RECs.
6. Distributed UPS designs (a battery cabinet serving several racks or a battery inside each server) avoid the AC/DC/AC double-conversion losses of a centralised UPS.
7. Google reported a 2023 fleet-average annual PUE of 1.10 against an industry average of 1.58; Microsoft reports PUE and WUE together.
8. DCPD (W/m²) is linked with thermal hot spots and cooling bottlenecks and is said to correlate negatively with RCI.

Using it in design

• In efficiency reporting do not make PUE the only metric; knowing its weaknesses (manipulation, optimal-condition reporting, ignoring IT utilisation), complement it with WUE, CUE and IT-side metrics (SPUE, PpW, ITEE).
• In CUE calculation use time-stamped grid carbon intensity and Scope 2 data, noting that provider data can be approximate.
• Compare distributed UPS architectures (battery cabinet or in-server battery) against centralised double-conversion UPS on conversion losses.

Who should read it

Designers, operators and policy specialists selecting and reporting data-centre energy-efficiency metrics.

Document contents

1. 1. Introduction (1.1 Background and Statistics, 1.2 Systematic Review Methodology, 1.3 Literature Review, 1.4 Existing Gaps and Contribution, 1.5 Paper Structure)
2. 2. Energy Management in Cloud Data Centers (2.1 Servers and Racks, 2.2 UPS, 2.3 PDU, 2.4 IT Rooms and Equipment, 2.5 HVAC, 2.6 Facility Security, Lighting, and Offices)
3. 3. Energy Efficiency Metrics in Cloud Data Centers
4. 4. Real-World Examples
5. 5. Challenges and Future Works (5.1 Challenges: IT-related and non-IT-related metrics; 5.2 Future Works)
6. 6. Conclusions

Read the document

Publication details

Authors

Ashkan Safari; Hoda Sorouri; Afshin Rahimi; Arman Oshnoei

Publisher

Electronics (MDPI AG)

Year

2025

Volume

14

Issue

11

Pages

2214

Document language

English

Source type

Scientific

Access

Open licence — downloadable

DOI

10.3390/electronics14112214

Licence

CC BY 4.0 — share with attribution

Licence evidence

www.mdpi.com

Citation