Driving data center PUE, efficiency

Overview

This article is an expert piece by Bill Kosik published in Consulting-Specifying Engineer in 2015. Its subject is what actually determines a facility's PUE. Kosik works step by step through the engineering inputs, starting with economizer control limits and supply and return temperatures and reaching transformer, UPS and PDU losses at part load.

The PUE basics section notes that well-designed, operated and maintained data centers typically land at PUE 1.20-1.60, that very efficient sites approach about 1.10, and that PUE cannot fall below 1.0. The author also compares the energy behaviour of central chilled-water plants, DX, evaporative, and direct and indirect economizer systems. A key point is that electrical losses are paid for twice: they become heat, which turns into extra load the cooling plant must reject.

The article also treats building envelope air leakage. The ranges cited include 0.11-0.33 cfm/sq ft for CIBSE TM-23/ATTMA TS1, 0.1 tight, 0.3 average and 0.6 leaky for ASHRAE Fundamentals, and 0.47-2.7 cfm/sq ft for a NIST survey. For the energy model it recommends 8,760 hourly climate values (ASHRAE IWEC2) plus extreme dry-bulb values for 5-, 10-, 20- and 50-year return periods. Typical-year files match ASHRAE 1% and 2% cooling and 99% heating conditions but not 0.4% and 99.6%, so equipment sizing should use ASHRAE Fundamentals design data.

On reliability, the reported five-year failure probability ranges from about 83% for an N system to about 4% for 2(N+1), and higher redundancy generally lowers part-load UPS and PDU efficiency. Even water-cooled IT still rejects roughly 10-30% of cabinet capacity to air, and such cabinets typically house 50-80 kW. The article closes with containment and close-coupled cooling options for dense racks. As a limit, it is a 2015 expert article and may not reflect current hardware densities.

Key points

1. Well-designed, operated and maintained data centers typically land at PUE 1.20-1.60; very efficient sites approach about 1.10; PUE cannot fall below 1.0.
2. The energy model should use 8,760 hourly climate values (ASHRAE IWEC2) plus extreme dry-bulb values for 5-, 10-, 20- and 50-year return periods.
3. Losses in transformers, switchgear, UPS, PDUs and RPPs reappear as heat the cooling plant must remove, so they affect energy use twice.
4. Envelope leakage: CIBSE TM-23/ATTMA TS1 0.11-0.33; ASHRAE 0.1 tight / 0.3 average / 0.6 leaky; NIST survey 0.47-2.7 cfm/sq ft.
5. Five-year failure probability is about 83% for an N system and about 4% for 2(N+1); higher redundancy lowers part-load UPS/PDU efficiency.
6. Typical-year weather files match ASHRAE 1%/2% cooling and 99% heating but not 0.4%/99.6%; sizing should use ASHRAE Fundamentals.
7. Water-cooled IT still rejects roughly 10-30% of cabinet capacity to air; such cabinets typically house 50-80 kW.

Using it in design

• Include economizer control limits, part-load loss curves for each electrical stage and the redundancy topology as energy model inputs.
• Add electrical losses back into the cooling load.
• Pull 0.4% and extreme weather design points for sizing separately from the 8,760-hour annual simulation file.
• Warn when a design PUE is quoted only at full IT load.

Who should read it

Engineers who do energy modelling, cooling and electrical design for data centers.

Related standards

• ASHRAE Thermal Guidelines for Data Processing Environments
• CIBSE TM23
• ASHRAE Handbook - Fundamentals

Publication details

Authors

Bill Kosik

Publisher

Consulting-Specifying Engineer

Year

2015

Document language

English

Source type

Expert / community

Access

Free to read

Licence

Copyright retained; citation only — the publisher’s terms apply to the original document

Citation