PUE Still Above 2.0 for Most Data Centers
A 2010 version of this post was titled "No One Can Agree on Typical PUE". I wrote, "As more data centers measure their PUE, more and more ask what is typical? The industry does not seem to agree, so a wide range of numbers are out there." I updated the post in 2012 with the latest data, concluding that most data centers still appear to be operating above a PUE of 2.0."
I put the question to Vertatique's global Green ICT community in August 2013 via a tweet: "After years of #GreenICT, is there evidence that most #datacenters now operate below PUE 2.0?". This was one of our most-retweeted, but no one came forward with new evidence. Some replied in the emphatic negative. Three years of very enlightening survey results from Digital Realty, including the newly-posted 2014 data, confirm that the 2012 analysis, below, still stands - PUE of less than 2.0 remains an elusive goal for many data centers.
Digital Realty shared its February 2014 research results with us. The global population survey respondents are "mid-market" companies: under $500 million in annual revenue, with a majority having 2000+ square feet of data center space. Here is what is informative from a Green ICT perspective.
A general question on the "important" or "very important" factors in making decisions about new investments, 71% cited "Energy efficiency of the data center facility" and 61% cited "Access to green and sustainable energy sources".
A more specific question about plans for new or expanded data center space within the next year yield a much lower priority for these sustainability issues among the 88% planning to invest in the near future. "Want a more energy efficient data center" was cited by only 25% and "Access to green and sustainable energy sources" by only 16%. It appears that 'green is good - just not so much right now.'
Only 27% thought their firm measures PUE, while 42% said the plan to do so in the future. Of those that thought PUE was being measured, 73% could not cite the value. The median PUE for those that did know was 2.0. Bottom line is that only 20% of these "senior level decision makers with responsibility for data centers" knew the most basic indicator of their facilities' energy efficiency. (PUE is a basic indicator only if you care: 19% called PUE measurement "not important", 8% were "not familiar" with the term, and 4% responded "Don't know".)
Here are regional results for those general decision factors:
Europe: 72% "Energy efficiency of the data center facility", 56% "Access to green and sustainable energy sources."
North America: 76% "Energy efficiency of the data center facility", 61% "Access to green and sustainable energy sources".
Asia-Pacific: 67% "Energy efficiency of the data center facility", 68% "Access to green and sustainable energy sources".
I was surprised by the relative ranking across regions of "Access to green and sustainable energy sources" among the eleven options. It ranking highest - sixth - in Asia-Pacific and lowest - eleventh - in Europe.
Digital Realty Trust's January 2013 research continues to show many data centers operate above 2.0 PUE.
Europe: "12% is unfamiliar with the term PUE. 19% doesn’t know their PUE. The average reported PUE is 2.53. 6% report a PUE of 3 or more; 28% report a PUE below 2.0."
North America: "5% don’t know their PUE and 3% are unfamiliar with PUE. 20% report a PUE of less than 2.0. The average reported PUE is 2.9."
Japan: "The average reported PUE is 2.22. 10% reports a PUE of 3 or more. 41% report a PUE below 2.0. 29% do not know their PUE."
Signapore: "11% are unfamiliar with PUE and 5% don’t know their PUE. The average reported PUE is 2.61 - 24% reports a PUE of 3 or more and 32% report a PUE below 2.0."
Hong Kong: "Relatively few are unfamiliar with PUE (4%) or don’t know their PUE (4%). The average reported PUE is 2.42. 15% report a PUE of 3 or more. 39% report a PUE below 2.0."
Australia: "One in twenty is unfamiliar with PUE and one in eight doesn’t know their PUE. The average reported PUE is 2.25. 13% report a PUE of 3 or more. 51% report a PUE below 2.0. Fewer do not know their PUE (13%) than in 2012 (25%)."
Let's use the PUE values the EPA in 2007 predicted for four 2011 target scenarios to see where the United states is now, as well as look at a survey that compares three world regions.
Today's 'mega data centers' are striving for very low PUEs by combining "Best Practices" and "State-of-the-Art" with the advantages of scale. Some are even below 1.1 PUE. Mega data centers represent the pinnacle of energy efficiency.
Some findings seem to indicate most data centers are at least at the predicted 1.9 PUE based on 2007's "Current Trends", but I am skeptical.
EPA's ENERGY STAR® for Data Centers rating system is based on a mean PUE of 1.9. The EPA will not publish the names of the 61 data centers in the 2010 ENERGY STAR reference set and we suspect that it is skewed by some of those mega data centers.
The Uptime Institute 2012 Data Center Survey PUE finding has a similar problem with large data center skewing. "Average PUE of respondents’ largest datacenter is between 1.8 and 1.89." Adding in the survey participant's smaller data centers would like increase average PUE.
The EPA group responsible for ENERGY STAR for Servers implies in its estimates of energy and CO2e savings that a PUE of 2.5 is more typical. Capgemini in 2010 used 2.5 to describe an "average datacenter". ""
- North America - "Few (7%) don’t know their PUE and even fewer (4%) are unfamiliar with PUE. One in five (19%) reports a PUE of less than 2.0. The average reported PUE is 2.8."
- Europe - "One in twelve (8%) is unfamiliar with the term PUE. One in eight (12%) doesn’t know their PUE, compared with 18% in the last wave.The average reported PUE is 2.61. Nearly one in four (23%) reports a PUE of 3 or more and 28% report a PUE below 2.0."
- Asia Pacific - "One in nine is unfamiliar with PUE and a similar number doesn’t know their PUE. The average reported PUE is 2.52. One in five (18%) reports a PUE of 3 or more. One in three (31%) reports a PUE below 2.0." This results are from only three locations: Australia, Hong Kong and Singapore. Full participation by India and China would likely significantly alter these results.
An April 2008 IBM report "found that only 28 percent of every dollar spent on energy was actually used by the IT equipment. The rest was spent on air-conditioning and other infrastructure—not generating productive use of IT for the business." This implies an average PUE of 3.6!
The EU Code of Conduct on Data Centres Energy Efficiency (October 2008) references "the Draft UK Market Transformation Programme European Enterprise Server installed base model, and assumes an upper bound ratio of 1:2 between electricity consumed by the server equipment within the data centre or server room, against that consumed by cooling equipment and through power losses. The lower bound ratio of 1:1 gives total electricity consumption close to 37 TWh. The upper and lower bound ratio is based on several different sources of measurements of electricity consumption in the data centre." This implies PUEs in the 3.0 - 2.0 range.
It is also useful to remember that improving PUE-type indicators is only one measure of increased energy efficiency. Another is to reduce the electricity directly consumed by the gear through virtualization, data de-duplication and archiving, active power management. and other GreenICT techniques. Focusing exclusively on PUE risks falling into the trap of greater efficiency through increased consumption.
And carbon reduction can be advanced in even the most efficient facility by converting to the cleanest-generated electricity possible for what energy is needed.
We are confident our new headline is accurate. Most data centers have not fully implemented "Best Practices" or "State of the Art" and are still well above 2.0 PUE. The only upside is that cloud computing may be improving industry PUE by moving more computing to state-of-the-art mega datacentrs.