Data Centers Want an MPG Rating for Energy Efficiency
- 15 April, 2009 09:26
These days, with the shock of US$150-per-barrel oil only a year old, consumers in the market for a car will likely pay much more attention to a pair of numbers: The vehicle's two miles-per-gallon ratings.
Companies in the market for efficient data center equipment, however, find themselves cursed with a scarcity of such data. While some basic efficiency measures exist for designers of data centers, as of yet, there is no standard method or metric for comparing energy efficiency of the computing hardware that fills the complexes.
"That has always been the challenge of the data center-you cannot compare one data center to another, because they do different work," says Michelle Bailey, vice president of market researcher IDC's Enterprise Platforms and Datacenter Trends group.
However, the industry is actively searching for an answer. Their first attempt compares the power used by the data center overall to the power consumed solely by information-technology equipment. Known as the power usage effectiveness (PUE), the metric gives data-center architects a data point with which to holistically compare the overall efficiency in delivering power to computers and routers in the building. In the past, data centers typically use a third to a half of all energy for cooling and other non-computer functions, giving them a PUE of 1.5 to 2.0.
Such efficiency matters. In 2005, the total power consumed by servers accounted for 0.6 percent of all electricity consumed in the United States, according to research done by AMD. Add to that the cost of cooling those servers and the requirements doubled, to 1.2 percent of U.S. annual power consumption. (See CIO.com's Five Energy Trends Driving Your Data Center for more background on this topic.)
Still, PUE is not that useful for companies in the market for new information technology. Instead, what is needed is a proxy for productivity-a technical term for a metric that acts as a good estimate of the energy consumer for a given amount of work. The miles-per-gallon rating used for cars is a such a proxy: A vehicle's miles-per-gallon rating does not actually correspond to the efficiency of a car under all conditions, but acts as a generally accepted measure of the amount of energy a car consumes to run.
Another example is the Dow Jones Industrial Average, according to Mike Monroe, director of sustainable computing for Sun Microsystems and a director of The Green Grid, a collection of computer hardware firms, enterprise application makers and data center providers focused on energy efficiency.
"The DJIA is a relatively simple indicator of the economy," he says. "If the economy goes up the index goes up. If the economy goes down, the index goes down."
Formed two years ago, the Green Grid currently maintains the PUE measure of data-center efficiency, and a related measure known as data center infrastructure efficiency (DCiE). The group, which now counts more than 200 companies as members, is actively seeking an improved metric of efficiency known as data center energy productivity (DCeP).
"The goal is a proxy for productivity that would measure many different kinds of data centers but not all kinds," Monroe says. "If we had a metric that covers 70 percent of the data centers out there, we would be happy. If it was 85 percent, we would be ecstatic."
The Green Grid initially attempted to create its own measure of productivity, but decided that it needed more input and created eight possible measures, which it has published for public comment.
"We pretty much, like everyone else, got our heads boggled when we tried to do it ourselves," Monroe says. "In some cases, useful work is easy to measure, such as a Google data center when work can be measured in searches served. Other data centers are more difficult."
Each of the eight different possible proxies deals differently with the problem of measuring the work done by a data center. Several of the possible ratings repurpose data collected during routine monitoring of the data center, while other would require that software vendors add additional code to their applications to monitor work. Still other proposals would benchmark data center products using common computer benchmarking suites such as SPECint and SPECpower. At least one proposal further abstracts the data center by equating work done with the number of bits output by the data center.
No matter which proposal or proposals are adopted, companies will have to resolve themselves to keeping track their use of energy in much greater detail, says David Cappuccio, managing vice president for infrastructure research at business intelligence firm Gartner.
"They are going to have to monitor their energy output at a much more specific level then ever before," Cappuccio says.
Moreover, the end result only tackles a single data-center consideration, says John Steigerwald, director of product management for IBM's performance management group. While energy efficiency, no matter how it is measured, will be increasingly important in the future, it will not be the sole factor in data-center decisions, he says.
"I don't think it will be the only measure that people will try to optimize," Steigerwald says. "It will be a combination of what it takes for IT to deliver the proper mix of availability, ease-of-use and cost to the rest of the company."