Metrics for Sustainable Data Centers

Overview

There is a multitude of metrics for analysing individual data-centre key performance indicators; to predict growth or set effective goals it is important to choose the correct metric and know its expressivity and limitations. As cloud services and ICT infrastructure grow globally, continuous monitoring and measurement of data-centre facilities is becoming essential. For each metric the study records the unit, whether to minimise or maximise, the optimal value and the scale at which it applies (facility, HVAC, rack, server, IT equipment, UPS), and maps 'uses' and 'based on' relationships between metrics.

The authors argue that no existing metric was designed for comparing one facility with another, and that PUE in particular depends on IT load, weather and location. PUE was never intended as a cross-facility comparison metric; it was meant as an internal improvement tool, and being strongly affected by IT load, weather and location, PUE comparisons between data centres are usually not representative. Metrics such as ITEU (an input to DPPE) need the exact power drawn by fans, voltage regulators and other components inside IT equipment, which is unclear how to measure, especially in heterogeneous colocation sites.

Combining existing metrics into new ones usually does not overcome the flaws of the originals and can amplify them; DPPE, for example, combines DCiE, GEC, ITEE and ITEU. Estimating power and cooling capacity from IT nameplate ratings may be inaccurate, and heat density varies within and between racks; operators relying on nameplate and modelled loads do not capture actual capacity requirements. Green metrics such as CUE, ERE, WUE and GEC need seasonal and regional benchmarking before facilities can be compared fairly. Thermal and air-management metrics (RCI, RTI, RHI/SHI, bypass ratio, recirculation ratio, balance ratio) share common inputs; Balance Ratio can be expressed as a function of Recirculation Ratio and Bypass Ratio. Tier availability figures quoted: Tier I 99.671%, Tier II 99.741%, Tier III 99.982%, Tier IV 99.995%.

The authors call for new comparison-oriented metrics that normalise for utilisation, location, climate, age and facility typology such as colocation. Limits: a taxonomy and critique with no new measurements; the accuracy of metrics depends on the accuracy of input data.

Key points

1. Metrics are grouped into nine dimensions (energy efficiency, cooling, greenness, performance, thermal and air management, network, storage, security, financial impact); each entry lists unit, objective, optimal value and scale (facility, HVAC, rack, server, IT equipment, UPS).
2. PUE was never intended as a cross-facility comparison metric but as an internal improvement tool; it is strongly affected by IT load, weather and location, so PUE comparisons between data centres are usually not representative.
3. Metrics such as ITEU (an input to DPPE) need the exact power drawn by fans, voltage regulators and other components inside IT equipment, which is unclear how to measure and hard in heterogeneous colocation sites.
4. Combining existing metrics into new ones usually does not overcome the flaws of the originals and can amplify them; DPPE combines DCiE, GEC, ITEE and ITEU.
5. Estimating power and cooling capacity from IT nameplate ratings may be inaccurate, and heat density varies within and between racks.
6. Green metrics such as CUE, ERE, WUE and GEC need seasonal and regional benchmarking before facilities can be compared fairly.
7. Tier availability figures quoted: Tier I 99.671%, Tier II 99.741%, Tier III 99.982%, Tier IV 99.995%.

Using it in design

• In data-centre comparisons, do not use PUE as a cross-facility metric; normalise for IT load, weather and location, or state the same conditions.
• Report green metrics (CUE, ERE, WUE, GEC) together with seasonal and regional benchmarks.
• Do not estimate power and cooling capacity from IT nameplate alone; verify heat density that varies within and between racks by measurement.
• Check that the inputs of composite metrics (such as DPPE) are actually measurable.

Who should read it

Designers, operators and researchers selecting, comparing and reporting data-centre metrics.

Document contents

1. 1 Introduction
2. 2 Taxonomy and relationships of metrics
3. 3 Metric dimensions (3.1 Energy Efficiency Metrics, 3.2 Cooling Metrics, 3.3 Green Metrics, 3.4 Performance Metrics, 3.5 Thermal and Air Management Metrics, 3.6 Network Metrics, 3.7 Storage Metrics, 3.8 Security Metrics, 3.9 Financial Impact Metrics)
4. 4 Discussion
5. 5 Conclusion

Related standards

• Uptime Institute Tier Classification
• ASHRAE Thermal Guidelines for Data Processing Environments

Read the document

Publication details

Authors

V. Dinesh Reddy; Brian Setz; G. Subrahmanya V.R.K. Rao; G.R. Gangadharan; Marco Aiello

Publisher

IEEE Transactions on Sustainable Computing (Institute of Electrical and Electronics Engineers (IEEE))

Year

2017

Volume

2

Issue

3

Pages

290-303

Document language

English

Source type

Scientific

Access

Open licence — downloadable

DOI

10.1109/tsusc.2017.2701883

Licence

CC BY 3.0 — share with attribution

Licence evidence

doi.org

Citation