Data Center Infrastructure Efficiency

The Green Grid paper on data center power efficiency metrics covers two key measures of data center infrastructure efficiency: PUE and DCiE (formerly DCE). The basic notion is that the most efficient facility is one in which most of the power goes directly to power the IT equipment, rather than to cooling, etc. But there are complexities.

PUE Overview

Simply put, PUE (Power Usage Effectiveness) is defined as data center power consumption divided by ICT equipment power consumption. (DCiE is the inverse of PUE.) Put another way, data center PUE = (ICT Power + non-ICT Power) / ICT Power. Over 70% of non-IT power goes to cooling in a typical facility.

Legacy PUE values are typically above 2.0. "Mega data centers" are claiming to be driving PUE down below 1.3.

Commentators have questioned the potential gaming of PUE numbers for PR purposes and the subtleties of using PUE to compare facilities. There are ways to focus on the PUE to improve 'efficiency' while neglecting total energy consumption. While these cautions are valid and useful, we should not lose sight of PUE's potential value to each individual facility seeking to improve its own performance.

Support of for PUE

PUE got a boost in February 2010 from a group of eight US government agencies, industry consortia, and standards bodies. The Agreement on Guiding Principles for Energy Efficiency Metrics states, "Power Usage Effectiveness (PUE) using source energy consumption is the preferred energy efficiency metric for data centers."

PUE got another boost in April 2010 when US, European, and Japanese groups reached an agreement that "designated Power Usage Effectiveness (PUE) as the industry’s preferred energy efficiency metric." There is a little more detail in the Guiding Principles released earlier this year:

- Power Usage Effectiveness (PUE) using source energy is the preferred energy efficiency metric. PUE is a measurement of the total energy of the data center divided by the IT energy consumption
- The industry should improve the IT measurement capabilities to ultimately enable taking the measurement directly at the IT load (e.g. servers). At a minimum IT energy measurements should be measured at the output of the UPS
- For a dedicated data center, total energy measurement should include all energy sources at the point of utility handoff. For data centers in larger buildings, total energy should include all cooling, lighting, and support infrastructure, in addition to IT load

US-EPA's ENERGY STAR® for Data Centers program, launched in June 2010, has made PUE the center of its rating system, although it attempts to recognize the scale issues in data center size.

Alternatives to PUE

The limitations of PUE have given rise to calls for improving the metric.

One idea for improving PUE is to include a measure of the actual work being performed; the idea being that a low PUE data center lightly loaded may be less desirable than a high PUE center fully loaded. In 2009, South African company SustinableIT advanced a concept called PUEI - Power Usage Effectiveness and Inefficiency to address this.

A 2010 article by the Icelandic company GreenQloud focuses on the C02e of the power source and discusses two other alternative: PUEx and GPUE

Update 2011.03.07

Invest in Tools

Tackling Today’s Data Center Energy Efficiency Challenges – A Software-Oriented Approach, a white paper published by Schneider Electric, argues for investing in the tools required for quality data gathering and analysis.

Techniques for the reduction of energy consumption in the data center environment take many forms, and if these are implemented without an over-all strategy the maximum over-all energy savings will not be achieved. Such a strategy involves a large amount of energy usage data and a large number of computations, some of which must be performed for different sets of data. Doing this by hand is labor-intensive and may not give optimal results. Instead, the implementation of both a hardware monitoring layer and software monitoring and analysis layer allows the gathering of energy usage data and the analysis of this data to achieve maximum energy savings. Such a system can use elements of already-installed power monitoring hardware. Investing in such a system can allow data centers to realize energy savings today, and increase energy savings in the future as new energy-saving techniques are developed and implemented.