Containing the cost of powering the data center is one of the chief priorities for many IT and facilities professionals. But determining whether a data center is a paragon of green virtue or an unrepentant gas guzzler is trickier than it might seem at first glance.
That's why the Green Grid, a nonprofit trade group led by IT vendors, devised PUE (Power Usage Effectiveness), which compares the total power used in a facility with the power devoted specifically to IT equipment. But as with almost any new standard, there is confusion within the industry about how to measure PUE and in what context it is useful.
[ Keep up on green IT trends with InfoWorld's Sustainable IT blog and Green Tech newsletter. ]
PUE measures the number of watts needed power and cool IT gear, taking into account all power dedicated solely to the datacenter, an important distinction in mixed-use buildings that house both IT equipment and office space.
"A PUE of 2.0 indicates that for every watt of IT power, an additional watt is consumed to cool and distribute power to the IT equipment," according to a Google research paper.
The Uptime Institute measures PUE in a fee-based service for its members, and found an average rating of 2.0 to 2.5. But Kenneth Brill, the institute's founder and executive director, is wary of businesses misinterpreting the measurements.
Say an IT shop virtualizes its servers, reducing its IT power consumption by 20 percent. Few observers would consider this a bad thing, but it might result in a higher PUE, angering the facility director's boss, Brill says.
"Is it the facility manager's fault that the IT load is light?" Brill asks.
Alternative measures
The Green Grid's PUE is one of numerous attempts to analyze IT power usage and efficiency. The Uptime Institute has proposed alternative measures, including CADE (Corporate Average Datacenter Efficiency), which takes into account not only power usage but also a company's CPU utilization rates and the effectiveness with which the datacenter's IT equipment transforms energy into "useful" work.
Separately, the U.S. EPA is preparing an Energy Star specification for servers, which is expected to be finalized this spring. The Storage Networking Industry Association is developing a metric for analyzing the power efficiency of storage devices.
[ The EPA has been grappling with the Energy Star spec for servers for more than a year now. ]
There has to be some way of analyzing an individual datacenter's use of power, and PUE is a great step in that direction, says Enterprise Strategy Group analyst Mark Peters. But people who use PUE should recognize its limitations, he says. PUE looks at a datacenter's stats for wattage in and wattage out, but it cannot be used to say whether a datacenter is using its resources efficiently, he says.
A data center could achieve disk utilization rates of 99 percent and a PUE of 1.2, but if that same datacenter is failing to de-duplicate data, it's probably wasting storage space and energy, he notes.
"I absolutely love PUE," Peters says. "But datacenter efficiency is a much bigger topic than power usage efficiency."
Two years after being proposed, PUE is starting to catch on. Officials at the SuperNAP co-location center in Las Vegas use internal PUE measurements to argue that the data center is the most energy efficient in the world. PUE is being adopted by the European Union within its datacenter code of conduct standard, according to Green Grid officials. But PUE should still be seen as an interim metric, says Mark Monroe of the Green Grid's board of directors.
"What everyone wants to get to is what are the computers doing with that electricity, and how efficient is that computer?" says Monroe, who is also the director of sustainable computing at Sun. "How much useful work is coming out of that computer?"
Today, there's little sense in comparing one datacenter to another because companies might be measuring it differently, Monroe says. PUE is mainly useful as an internal metric to compare your data center to itself over time, he says. The Green Grid's Web site details what must be measured to calculate PUE, such as power delivery components like UPS, switching gear, generators, power distribution units, batteries, and power losses. But the organization is still working toward implementing detailed specifications and a certification program in which it will verify PUE at individual facilities. That should happen within a year, Green Grid officials say.
Gaining enterprise attention
While PUE has gained traction among industry powers such as Google and Microsoft, which both boast of low PUE ratings, Peters says enterprise customers for the most part aren't measuring it at their own data centers.
"It's too new for it to have filtered that far," he says.
Still, numerous vendors have stepped up to provide software tools for measuring both total IT power usage and the power devoted to individual devices. And some large enterprises have begun tracking PUE and how it changes over time.
James Carney, executive vice president of datacenter planning at Citigroup in New York, says he measures PUE at all the company's datacenters, but getting a meaningful measurement is difficult at facilities that include both data center equipment and office space.
"The only places you can very accurately measure it are where you have stand-alone, purpose-built datacenters," Carney says. In one such datacenter in Texas, Carney says Citi has achieved average PUE ratings of 1.58 to 1.62.
The keys for Carney include partnering with the company's real estate organization, using energy management software and control systems that track power entering all points of the datacenter, and using all the data to optimize cooling temperatures.
"People look at watts per square foot. They look at chilled water temperature. PUE gives a pretty holistic view of what's going on in the entire environment," Carney says. "We also like to measure [power usage] at the cabinet level and different areas of the datacenter floor."
Beyond the hype
While PUE is useful for measuring how power usage at an individual datacenter changes over time, public claims of incredibly low PUE ratings can't always be believed.
According to Brill, one company claimed to have a PUE of 0.9, which is impossible because that would mean its servers use more power than is actually delivered to them. The company said its use of solar power and windmills shouldn't count toward power usage, Brill says. Some companies might skew ratings by measuring UPS but not other power sources, or by measuring only during the cooler portions of the year, rather than calculating an average across 12 months, he says.
Meanwhile, Google has claimed an average PUE of 1.21 across six datacenters, and the most efficient of the six had an annual rating of 1.15 and a best quarterly rating of 1.13. If the numbers are accurate, only about two-tenths of a watt (for every watt of IT power) is needed to cool Google's servers and distribute power to IT equipment.
But how meaningful are Google's measures to your standard enterprise with just a few datacenters? If one Google datacenter fails, it has many others to pick up the slack. But if a company has a datacenter that must absolutely remain online at all times, lowering the PUE might require sacrificing redundant power supplies that maintain required uptime, Brill says.
Enterprises that want to measure PUE or analyze the energy consumption of individual devices have several options. Vendors like GDCM, Server Technology and Sentilla offer software products that measure and analyze power usage in the datacenter. Such tools can help reorganize IT loads to maximize power efficiency, or use chargeback to bill departments for their energy usage.
PUE, if not perfect, helps companies understand how IT load changes affect power usage, says Calvin Nicholson, director of product marketing for Server Technology.
"I think there's a lot of hype [around PUE]," he says. "But it's simple, it's easy for people to understand, and it makes sense."
Network World is an InfoWorld affiliate
This story, "Measuring datacenter efficiency easier said than done" was originally published by Network World.