PUE and Marketing Mischief
July 13th, 2009 By: Rich Miller
The reporting of Power Usage Effectiveness (PUE) has been a hot button issue in the industry for some time. PUE has become established as the leading “green data center” metric, but its value has become fuzzy due to a disconnect between companies’ desire to market their energy efficiency and the industry’s historic caution about disclosure. As a result, there have been many public reports of low PUE numbers that have generated skepticism.
Mike Manos of Digital Realty discussed this trend in a recent post at Loose Bolts about the current problems with PUE reporting. “PUE is poised to be a victim of its own success, in my opinion, unless the industry takes steps to standardizes its use in marketing material and how it is talked about,” Mike writes. “These days, I view each and every public statement of PUE with a full heaping shovel-full of skepticism regardless of company or perceived leadership position.”
Manos also reports a worrisome sign: a recent data center RFP from a regional government entity stipulating that the facility must have a PUE of 1.2 or lower, a level of efficiency achieved by a handful of working data centers at some of the largest operators. These kind of stipulations represent the sort of misuse of PUE that critics – most notably Ken Brill of the Uptime Institute – have warned about. Brill’s concern has been that executives yearning for low PUE numbers would set unreasonable expectations for data center managers.
Manos proposes a solution: “Lets re-claim PUE and Metrics from the Marketing People,” he writes. Mike writes at length about the challenges of PUE reporting and possible solutions, including four different types of PUE ratings that he believes can provide common reporting standards and address concerns about apples vs. oranges comparisons. There’s no shortage of either detail or opinion in the lengthy post, which is worth reading if you’re interested in PUE.
Manos isn’t alone in suggesting additional data and/or refinements to PUE. The Green Grid, which popularized the PUE metric, has outlined three tiers of PUE metrics while Amazon’s James Hamilton has offered an enhanced standard called tPUE. The government may also have a seat at the table, as the U.S. Environmental Protection Agency has indicated that it will use PUE as the basis for its Energy Star for Data Centers ratings.
How will this play out? Manos cites the need for “an honest, open and frank conversation around this topic.” It remains to be seen whether the industry can come together on an approach that addresses the current concerns without undermining the usefulness of the metric for audiences beyond the walls of the data center. It’s worth remembering that one motivation in advancing PUE was to simplify discussions of energy efficiency, a goal that could become more elusive if the standard becomes increasingly complex.
Interesting post – thanks. I agree with Mike that PUE is very useful, when used correctly. It seems the Green Grid hasn’t stepped up to defining a standard methodology similar to those required by auto-makers when making MPG claims.
To further complicate things, the EPA EnergyStar for Data Centers is calling it’s metric PUR or “PUE Ratio”, but I would call it more like EUE: Total Energy/IT Energy. I believe they use a montly total of energy consumed (electricity, diesel fuel, natural gas, etc) all normalized to kWh.
Keep up the great work.
President – Archinetix, Inc.
I’m glad to see this type of press on PUE. I did read Mike Manos’s post. There is something else I would add that wasn’t mentioned in these posts. In addition to his other questions regarding “Which PUE do you use?”, another question should be at what data center load percentage. As Mike showed in the tables, the PUE in a lightly loaded data center is much higher than one that is fully loaded.
If the Green Grid were to come up with a standard way to state PUE we would see less marketing hype. So for example, reporting PUE would require something like this: Annual average PUE of 1.9 at 65% load in New York City. Of course, along with this, we would need a data collection standard. Like Mike mentioned, how often was the data collected and how long were the measures made for.
Apart from this, I need a standard that tells us what should be measured as part of a valid PUE. Are security gates included? Are the mechanical plant air compressors included? Are the cooling tower basin heaters included? My guess is that if we had an audit done at some of these super efficient data centers, we would quickly find out that they aren’t measuring everything. We should be especially skeptical of PUE numbers for data center in a shared building, since backing out the proportion of shared devices like central chiller plants is tricky. It’s one thing to use inaccurate PUE numbers as a means to gauge improvements to your own data center, it’s another thing to market it and skew the reality of others.
Finally, has anyone ever tried measuring the losses across a UPS, switchgear, PDU, or between the utility entrance and PDUs? I have, and I can tell you that it’s easier said than done. Those who talk about measuring PUE like it’s easy to do, would be surprised at the amount of “head-scratching” that goes on when you try to compare your meter reading against the LCD panel of these devices.
Excellent points on PUE’s use and misuse here, and kudos to Mike Manos for his July 4th post as well.
Another issue is that even if all data centers magically agreed to measure PUE in exactly the same way, a “good” PUE target for one type of data center may be wildly inappropriate for another type of data center.
For example, I think it makes sense for companies like Google and Microsoft to shoot for PUEs of 1.2, since their data centers are the core of their search business, and they can run them like optimized factories that really excel at delivering a few key applications with maximum CapEX and OpEx efficiency. I would also hypothesize that they can pursue certain energy-saving strategies very aggressively, totally secure in the knowledge that if an odd server fails, a hundred (or a thousand?) similar servers running the same app or an entirely different data cetenr can pick up the slack. Most users will assume any gap in service is due to their internet connection and not Google or Microsoft or Yahoo deficiency.
A data center run by a financial services or a retailer or a logistics company has an entirely different uptime mission and different consequences for downtime. They are often burdened with a mix of heterogeneous legacy and custom IT equipment, so it’s much harder to optimize facility performance while still meeting the diverse needs of all the IT boxes. Sometimes a custom box is the only box running a specific mission critical application, so data center management often has to treat every box on the floor as if it’s made of gold. This makes it harder to pursue innovative efficiency strategies that carry even minor operational risk. As a result, these companies appropriately make different business tradeoffs between cost and reliability, In addtion, these business oriented data centers have disaster recovery and other legal Sarbox overhead processing requirements that search data centers do not. For data centers of this type, perhaps the best attainable PUE might be 1.7 or 1.8 depending upon free-cooling opportunities. Which means at light loads, they will be running well above 2.0 much of their life.
At Uptime’s 2009 Symposium held last April, Bill Weihl from Google correctly pointed out that differences in data center types shouldn’t be used as an “excuse” for operators at some types of facilities to slack off on energy efficiency. Still, I’m not yet convinced that it’s reasonable to expect all (or even most) data centers to be able to achieve the same level of efficiency as someone like Google or Microsoft has achieved for their search functions (what we never hear quoted is the PUE for their own internally focused data centers – I strongly suspect it is not 1.2).
I’m curious what others across the industry think about the notion that data centers with different business missions should have different targets for what constitutes a “good” PUE.
Thanks again, Rich, for your thoughtful post and for advancing the discussion on consistent efficiency measurements.
Ken Brill, Executive Director, Uptime Institute
Great Article and great comments. As I am on the cooling side, I would suggest that in some cases physics plays a large part in the best possible efficiency of a data center. When maintaining the preferred vendor server environment in some cases it’s not even possible to achieve a PUE of less than 1.32 assuming a 10% loss in utilities (power conversion, lights, batteries) in some area’s of the world such as the south eastern U.S. At some point there is a cut off of what can be done and what can’t, while maintaining sustainability of the facility and its equipment. I think a more holistic approach should be taken while evaluating data center efficiency. Energy, Water and E-Waste. As well as lost productivity and server side utilization.
I do believe the PUE or DCIE are great metrics to evaluate the mechanical infrastructure, why because with your UPS trends and a facility utility bill you can validate any claims of your facilities efficiency in 2 minutes. Data Center efficiency suffers no fools, you can’t hide anything in the data center energy is energy, a kW is a kW, a btu is a btu and water is wet.
Since I’m on the cooling side I have a question, is there a matrix for the server side. Something like bit per watt? I hear all kind of claims for server efficiency that to be honest really don’t matter if that server is running, but not being utilized…
Were heading down a great path and things have a way of working themselves out. The data center industry is full of really intelligent people and it doesn’t take long before a charlatan is proven a charlatan.
Rick Cockrell, CTO Core4
Thank you for sharing your perspective on PUE. As you mentioned, The Green Grid has been developing metrics/best practices for improving and measuring energy efficiency. While we do understand the various industry concerns surrounding the metric that you have highlighted above, we believe that using these metrics leads to consistent, apples-to-apples reporting, and ultimately allows the industry to identify and promote exceptional results, and provide a level of quality assurance for all data center managers.
Dave/Vector/Ken/Rick – would also like the opportunity to discuss your comments further regarding ways The Green Grid can improve their efforts. We are always looking to broaden the industry discussion around data center efficiency and encourage organizations to join The Green Grid to contribute their ideas and perspective.
Let me know if you would be interested in continuing this discussion via email.
Director, The Green Grid
[...] PUE and Marketing Mischief « Data Center Knowledge – The reporting of Power Usage Effectiveness (PUE) has been a hot button issue in the industry for some time. PUE has become established as the leading “green data center” metric, but its value has become fuzzy due to a disconnect between companies’ desire to market their energy efficiency and the industry’s historic caution about disclosure. As a result, there have been many public reports of low PUE numbers that have generated skepticism. [...]