Data Center Energy Efficiency Measurement Assessment Kit: Guide and Specification (DRAFT, July 25, 2014)

Overview

Written by Rod Mahdavi of LBNL for the Federal Energy Management Program, the guide addresses owners, IT and facilities managers. It introduces a temporary kit of temperature, humidity and pressure sensors plus power meters for CRACs, suitable only for data centers with air-cooled IT equipment. The motivation is that most federal data centers lack monitoring, a conventional assessment is a snapshot of power read from equipment displays, installing power meters usually needs a shutdown, and air management goes unexamined even though the guide says air management measures are the easiest and most cost-effective. The guide states it is not a detailed energy audit procedure and does not cover IT efficiency or savings calculations.

It defines the quantities to track: PUE, air temperature and relative humidity at IT intake, Return Temperature Index at each rack, cooling system efficiency, airflow efficiency and power distribution efficiency. RTI is the CRAC supply-to-return temperature difference divided by the rack intake-to-exhaust difference, a way to see bypass or recirculation without measuring airflow. Cooling efficiency is total cooling-system power divided by cooling delivered, and distribution efficiency is power reaching IT divided by power at the entry, better read at the switchgear than upstream of the UPS. PUE reporting levels follow the Green Grid, with UPS output usually taken as IT power at level 1.

The specification section justifies wireless mesh by flexibility and avoided cabling and sets requirements per component. The gateway runs from 120 V with at least 72 hours of battery backup; rack and CRAH/CRAC thermal nodes need +/-0.6 °F accuracy and reporting at least every 5 minutes; dew point accuracy is +0.9 °F; differential pressure spans -1.0 to 2.0 in. of water with zero-point accuracy of +/-0.002 in.; current transducers cover 10 A to 1000 A at +/-1% of full scale; voltage transformers are 600 Vac rated with +/-0.5% accuracy. Batteries should last at least four years, transmission power under 3 mW, and reliability on the order of 99.999%. Devices must not support IP protocols and traffic must be encrypted.

The LBNL kit is described with quantities: 20 rack thermal nodes with seven temperature sensors and one humidity sensor each, 3 sets for low-density racks, 10 CRAC node sets, 40 dew point sensors, 6 pressure transmitters covering 3,000 square feet and 8 current transmitter sets. Installation advice includes putting thermal nodes on every third rack in a row, installing current transformers in junction boxes to avoid shutting CRAC units, estimating cooling at 0.285 tons per kW of IT load when no plant meters exist, and noting that reducing underfloor pressure to around 0.03 in. can cut airflow. A four-day sample project raised rack intake temperature through sealing, tile moves and a higher supply setpoint, allowing warmer chilled water. It is a short 2014 draft.

Key points

1. Return Temperature Index (RTI) is the CRAC supply-to-return temperature difference divided by the rack intake-to-exhaust temperature difference, indicating bypass or recirculation.
2. The wireless gateway (120 V house power) must have at least 72 hours of battery backup.
3. Rack and CRAH/CRAC thermal nodes require accuracy of +/-0.6 °F and must report to the gateway at intervals not exceeding 5 minutes.
4. Differential pressure transmitters should span -1.0 to 2.0 in. H2O with zero-point accuracy of +/-0.002 in. H2O.
5. Current transducers should cover 10 A to 1000 A with +/-1% full-scale accuracy; voltage transformers are 600 Vac rated with +/-0.5% accuracy.
6. Wireless data reliability should be at least on the order of 99.999%; nodes should transmit below 3 mW and batteries should last at least four years.
7. Devices must not support IP-based protocols and the wireless network must encrypt its traffic.
8. The kit has 20 rack thermal nodes (7 temperature sensors and 1 humidity sensor each), 10 CRAC node sets, 40 dew point sensors, 6 pressure transmitters covering 3,000 sq ft, and 8 current transmitter sets.
9. Rack thermal nodes are usually installed on every third rack in a row; reducing underfloor pressure to around 0.03 in. can cut airflow and save fan energy.
10. Without plant meters, required cooling is estimated at 0.285 tons per kW of IT load, and PUE is computed rather than read directly.

Using it in design

• Borrow the metric set (PUE, intake temperature and humidity, RTI, cooling efficiency, airflow efficiency, distribution efficiency) as the minimum list of measurements a monitoring design must support.
• Use the sensor accuracy and reporting figures (+/-0.6 °F, 5-minute interval, +/-0.002 in. H2O, +/-1% CTs) as a starting specification for DCIM or environmental monitoring procurement.
• Plan sensor density per the kit: rack nodes on every third rack, several intake and exhaust points per rack, supply and return nodes on every cooling unit.
• Provide junction-box access for current transformers so cooling unit power can be metered later without shutdowns.
• Specify wireless gateways with battery backup, encryption and no IP-based protocol support where the site security policy requires it.
• Treat the document as a 2014 draft: check the figures against current vendor data and standards.

Who should read it

Energy assessors, facility and IT managers, and engineers specifying environmental monitoring or planning a temporary assessment of an air-cooled data center.

Document contents

1. I. Summary
2. II. Guide/Specification Scope
3. Target Audience
4. What This Guide Does
5. What This Guide Does Not Do
6. III. Data Center Monitoring
7. Principal Parameters: what, why and where to measure
8. Data Center Visual Trending
9. IV. Wireless Network Specification
10. Why Wireless?
11. Components and Features Specifications
12. Scope of Work
13. V. Assessment Kit
14. Assessment Kit Components
15. VI. Assessment Kit Installation and Commissioning
16. System Data Presentation
17. Sample Presentation
18. VII. References

Publication details

Authors

Rod Mahdavi

Publisher

Lawrence Berkeley National Laboratory

Year

2014

Document language

English

Source type

Government / public

Access

Citation only

Licence

Licence conditions not established — confirm the conditions with the publisher

Licence evidence

datacenters.lbl.gov

Citation