Data Center Knowledge: How to Make Your Data Center PUE Calculation More Accurate

September 17, 2015 Schneider Electric

Data Center Knowledge: How to Make Your Data Center PUE Calculation More Accurate
By Industry Perspectives on September 14, 2015

Original article: 

http://www.datacenterknowledge.com/archives/2015/09/14/how-to-make-your-data-center-pue-calculation-more-accurate/

Victor Avelar is a Senior Research Analyst for Schneider Electric’s Data Science Center.

The quest to conserve energy in the data center is ongoing. It has motivated data center managers to have a simple, standard means for tracking a facility’s total power usage against the amount of power used by the IT equipment. To address the need for an industry-wide benchmark, The Green Grid developed the power usage effectiveness (PUE) calculation in 2007 as a principle way to measure data center infrastructure efficiency.

Devices are encountered in data centers that draw power, but how (or if) their power data should be counted in the efficiency calculations is unclear.While PUE has become the de facto metric for measuring infrastructure efficiency, data center managers must clarify three things before embarking on their measurement strategy: There must be agreement on exactly what devices constitute IT loads, what devices constitute physical infrastructure, and what devices should be excluded from the measurement. Without first clarifying these three things, it can be difficult for data center managers to ensure the accuracy of their PUE. However, this process is easier said than done, as there are a number of issues that can make classifying power-consuming subsystems as 1) IT loads, 2) physical infrastructure, or 3) neither, problematic:

  • Various power-consuming data center subsystems aren’t present in some data centers such as outdoor lighting or the Network Operation Center (NOC).
  • Some subsystems support a mixed-use facility and are shared with other non-data center functions (for example, cooling towers and chiller plants) so fractions of the power attributable to the data center cannot be directly measured.
  • Some practical power measurement points include loads that are unrelated to the data center, but cannot be separated from the measurement.

Compounding the issue further is the fact that commonly published efficiency data is not computed using a standard methodology, and the same data center can have a different energy efficiency rating when different methodologies are applied. So what can a data center or facility manager do?

A Three-Pronged Solution to PUE Calculations

Since most data center operators who attempt to determine PUE will encounter one or more of the above problems, a standard way to deal with them should be defined. The three-pronged approach outlined below can be used to effectively determine PUE.

This methodology defines a standard approach for collecting data and drawing insight from data centers. It also helps data center managers understand how to use this approach to calculate PUE, with a focus on what to do with data that is either misleading or incomplete.

One: Establish a Standard for Classifying IT Loads and Physical Infrastructure

The first part of this methodology is to establish a standard to categorize data center subsystems as either (a) IT load or (b) physical infrastructure or (c) determine whether the subsystem should be excluded in the calculation. While it’s fairly simple to designate servers and storage devices as an IT load, and to lump the UPSs and HVAC systems into physical infrastructure, there are subsystems in the data center that are harder to classify. For example, the personnel spaces, switchgear, and the NOC, which all consume power, do not clearly fall into these categories. However, if these subsystems are not uniformly classified forall data centers, it’s not possible to directly compare computed efficiency results across different data centers within your data center portfolio. Since many customers, government bodies, utilities, and data center providers are looking for a standard benchmark for data center efficiency, clear guidelines for what is classified as an IT load or physical infrastructure are critical to determining a benchmark that can used across various data centers.

Two: Calculate PUE for Shared Devices

Some devices that consume power and are associated with a data center are shared with other uses such as a chiller plant or a UPS that also provides cooling or power to a call center or office space.

Even an exact measurement of the energy use of these shared devices doesn’t directly determine the data center PUE, since only the device’s data center-associated power usage can be used in the PUE calculation. One way to handle this is to omit the shared devices from the PUE, but this approach can cause major errors, especially if the device is a major energy user like a chiller plant.

A better way to measure this shared device is to estimate the fraction of losses that are associated with the data center, and then use those losses to determine the PUE. There are three ways to do this using a chiller plant as an example:

  • Measure/estimate the thermal load on the chiller using all the electrical losses of all the other data center loads, then measure/estimate the chiller performance. This approach will provide you with a good estimate of how much of the chiller’s power the data center is using.
  • Measure the fractional split of the thermal load between the data center and the other loads. Using water temperature, pressure, pump setting, etc., measure the chiller input power, and then allocate the traction of the chiller power to the data center according to the fractional split.
  • Shut off the non-data center loads on the chiller, and then measure it to determine the power offset for the data center. These indirect estimates are best made during an expert data center energy audit, and once the technique is established it can frequently be used over time when efficiency trending is important.

Three: Provide an Estimate for Devices that are Impractical to Measure

While every device in the data center that uses energy can be measured, it can be impractical, complex, or expensive to measure its energy use. Consider a power distribution unit (PDU). In a partially loaded data center, the losses in PDUs can be in excess of 10 percent of the IT load. These loss figures can significantly impact PUE, yet most data center operations omit PDU losses in PUE calculations because they can be difficult to determine when using the built-in PDU instrumentation.

Fortunately, the losses in a PDU are quite deterministic and can be directly calculated from the IT load with precise accuracy if the load is known in either watts, amps or VA. In fact, this tends to be more accurate than the built-in instrumentation approach. Once the estimated PDU losses are subtracted from the UPS output metering to obtain the IT load, they can be counted as a part of the infrastructure load. This method improves the PUE calculation, as opposed to ignoring PDU.

With this three-pronged standard methodology, data center managers can accurately and effectively determine PUE to ensure their data centers meet not only energy efficiency regulations but larger business goals as well.

Industry Perspectives is a content channel at Data Center Knowledge highlighting thought leadership in the data center arena. See our guidelines and submission process for information on participating. View previously published Industry Perspectives in our Knowledge Library.

Previous Article
[Infographic] How Micro Data Centers Help Win The Race For Scale
[Infographic] How Micro Data Centers Help Win The Race For Scale

Scaling to meet demand almost always means balancing capacity with time and cost of deployment.

Next Article
Micro data centers doing the bridgework for IoT
Micro data centers doing the bridgework for IoT

The fight for data centre efficiency is a never ending one. Yet despite the best laid plans, data centre ar...

×

Ready to talk to an expert? Fill out your info and we'll be in touch:

First Name
Last Name
Company - optional
Would you like to sign up for our newsletter?
Phone Number
Country
Comments - optional
Thank you!
Error - something went wrong!