Prediction of Overall Energy Consumption of Data Centers in Different Locations

Overview

The study starts from the need to analyse the economic budget comprehensively against energy requirements and potential energy cost before building a hyperscale data centre (HDC), and proposes a global energy consumption prediction framework based on PUE calculation that considers all heat sources and power consumption. The PUE model has three free-cooling scenarios: airside economisers with adiabatic cooling (AE), waterside economisers using cooling-tower evaporative capacity (WEC) and waterside economisers using seawater (WES); a chiller covers the remaining hours. Facility heat load is modelled as IT power plus UPS losses (1 - eta_UPS) x P_UPS, plus power-distribution losses and lighting, with fan power added for total facility power.

Using hourly NOAA Integrated Surface Database weather, the framework estimates annual PUE for 60 regions (49 in China and 11 abroad). Seasonal results show higher PUE in summer and lower in winter, and colder sites such as Harbin benefit more from free cooling than Nanjing. Chinese data centres usually combine a cooling tower and a plate heat exchanger (the WEC configuration), which was used for comparison with reported PUEs. Carbon (CUE) and electricity-cost estimates are derived from PUE using regional grid mixes and tariffs.

For an assumed 100 million kWh/yr of IT energy, estimated annual electricity cost ranged from $8.1376M (Delinha) to $9.6742M (Yinchuan) among the five regions in Table 5, showing tariff differences can outweigh PUE differences. UPS efficiency had a large effect on modelled results, and parameter sensitivity rankings changed with location. The authors argue that PUE alone is a poor siting criterion, because low-PUE sites can have high electricity prices or carbon-intensive grids.

Limits: the framework is for HDCs with air-cooled IT equipment; predictions were compared with actual reports, but model parameters use limited ranges and cost estimates rest on tariff assumptions. In general, sites in colder climates can take advantage of free cooling and cut energy cost.

Key points

1. The PUE model has three free-cooling scenarios: airside economisers with adiabatic cooling (AE), waterside economisers using cooling-tower evaporative capacity (WEC) and waterside economisers using seawater (WES); a chiller covers the remaining hours.
2. Facility heat load is IT power plus UPS losses (1 - eta_UPS) x P_UPS, power-distribution losses and lighting, with fan power added for total facility power.
3. Annual PUE was estimated for 60 regions using hourly NOAA ISD weather; PUE is higher in summer and lower in winter, and colder sites such as Harbin benefit more from free cooling than Nanjing.
4. Chinese data centres usually combine a cooling tower and a plate heat exchanger (WEC), which was used for comparison with reported PUEs.
5. For an assumed 100 million kWh/yr of IT energy, estimated annual electricity cost ranged from $8.1376M (Delinha) to $9.6742M (Yinchuan) among five regions.
6. UPS efficiency had a large effect on modelled results, and parameter sensitivity rankings changed with location.

Using it in design

• In siting, do not use PUE as the only criterion; evaluate it together with electricity cost (tariff) and carbon (CUE, grid mix) estimates derived from PUE, since a low-PUE site may have a high tariff or carbon-intensive grid.
• In site comparison, compute annual PUE from hourly weather data and show the seasonal difference (higher in summer, lower in winter).
• Treat UPS efficiency as a sensitive input; it strongly affected results and sensitivity rankings changed with location.

Who should read it

Planners and energy modellers doing hyperscale data-centre siting and energy budgeting.

Document contents

1. 1. Introduction
2. 2. Related Work
3. 3. Energy Consumption Prediction Framework (3.1 PUE Prediction, 3.2 IT Equipment Energy Consumption Model)
4. 4. Evaluation (4.1 Sensitivity Results, 4.2 Annual PUE Estimation Analysis, 4.3 Carbon Emissions Prediction, 4.4 Electricity Cost Estimation)
5. 5. Conclusions

Related standards

• ISO/IEC 30134-2 (PUE)
• ISO/IEC 30134-8 (CUE)

Read the document

Publication details

Authors

Yiliu Zhang; Jie Liu

Publisher

Sensors (MDPI AG)

Year

2022

Volume

22

Issue

10

Pages

3704

Document language

English

Source type

Scientific

Access

Open licence — downloadable

DOI

10.3390/s22103704

Licence

CC BY 4.0 — share with attribution

Licence evidence

www.mdpi.com

Citation