Analyst Report: How a Snapshot of a Green Data Center Can Be Misleading


About a month ago, Facebook launched its Open Compute Project, which unveiled the nitty gritty of its energy-efficient data center in Oregon. The move is widely hailed as ground-breaking for an industry that keeps the location and size of its data centers top secret. But to me, Facebook’s touting of one specific efficiency metric shows just how a single snapshot of an energy-efficient data center can sometimes be misleading.

That efficiency metric I’m referring to is power usage effectiveness, or PUE, which looks at how efficiently a data center uses power and how much energy consumption is going to run its IT and servers. PUE rates data centers on a scale between one and two, with one being great and two not efficient. Think about it this way: If a lot of the data center’s energy is used for cooling and power conversion, the energy use is not very efficient and the data center will have a PUE closer to two; if very little of a data center’s energy is used for cooling and power conversion (and is instead mostly being used to power its servers), a data center is using energy much more efficiently and will have a PUE closer to one.

Looking behind the scenes of the Facebook PUE number is important because the Internet industry is only just starting to invest significantly into making its data centers far more energy efficient as a competitive advantage. According to the Environmental Protection Agency (EPA), data centers account for over 2 percent of energy consumption in the U.S.; that number is set to grow as more and more people buy up always-on gadgets and constantly connected computers. If an Internet company can cut its data center energy bill significantly, it can compete more effectively in the market by saving money, and also in smaller way, by being able to talk about its green ambitions.

Behind the Scenes of the Record Facebook PUE

The EPA lists data centers with energy-efficient best practices as having a 1.5 PUE. For Facebook’s data center in Oregon, the social network giant announced a PUE of 1.07 — one of the lowest I’ve heard about in the industry. The number is even lower than the company’s initial target of 1.15, and lower than the 1.5 PUE number that Facebook says is attributed to the data center space it already leases from third parties to run its web services.

After its unveiling, pundits immediately compared this super-low PUE number to Facebook competitor Google’s. Google has been publishing and lauding its own PUE numbers for years. It announced PUEs for its data centers back in 2008 of an average of 1.13, which was close to an industry record at the time. In April, this article on DataCenterDynamics declared that Facebook “seems to have pulled ahead” of Google and Yahoo for its PUE figure.

But take a second look at Facebook’s ultra-low PUE number, and you can see that it’s actually just a snapshot of the data center in a prime energy-efficient state. The number was calculated over an eight-hour period during the data center’s commissioning stage. In reality, data center PUE numbers consistently fluctuate throughout the year, often times rising in the warmer summer months and dropping in the colder winter months. The Facebook data center commissioning stage took place during the latter, in December 2010, when the weather was colder and generally needed less energy to cool and operate.

Google, which tracks its PUE on a quarterly basis, recorded an average PUE for its data centers in the third quarter of 2010 (July, August, September) of 1.2. Those same data centers dropped down to 1.13 in the fourth quarter of 2010, or the three end of the year winter months. Summer months generally require more cooling (servers can’t run well when they overheat), which requires more energy. Google says, “We measure throughout the year and not just during favorable seasons.”

The Importance of Being Efficient

In its defense, Facebook has acknowledged that the PUE metric is just a snapshot, and says it will be reporting its PUE on a quarterly basis in the future. No doubt, it will be updating the metric with more info soon. “We’re still in the early days of running the data center, so this number will change over time and we’re aiming to do even better in the future,” notes Facebook on its Open Compute website. However, it’s hard to do even better than 1.07, especially when that number was pulled during a prime energy efficient window.

Still, given the trend of energy-efficient data centers is somewhat new, it’s important to make sure that metrics like PUE keep their integrity and are used in the most telling manner. Standardized, independently reviewed metrics will be crucial to make sure Internet companies aren’t just greenwashing when they make claims about their green data centers, and that they are meeting certain requirements.

There will also be an education process as newer companies, like Facebook, start working on data center efficiency, while those that have spent years on the technology — like Google and Yahoo — help the newer ones by showing their lessons learned and case studies. PUE is only one of the first and more commonly used metrics to report data center efficiency, and groups like the Green Grid are working on a whole host of tools to track and show efficiency.

I’d also like to see companies like Google and Yahoo work on being as open as Facebook has been with its data center efficiency design. Facebook knows that it has things to learn about its data centers, and is embracing feedback as part of the process of making its data center designs transparent. In a video, Jonathan Heiliger, VP Technical Operations at Facebook, said: “Give us feedback, tell us where we screwed up, tell us where we made a bad decision, and help us make it better.” That mentality in itself is unprecedented in the data center industry, and will help make sure Internet companies become more educated about how PUE works, and will also ultimately help data centers integrate more energy efficient practices.

Table of Contents

  1. Summary

Join Gigaom Research! Become a subscriber and get reports like these, plus our collection of over 1,700 reports from world-class analysts for just $995 a year.