Google Data Center Power Usage Effectiveness
(click for larger image)
Thanks to cool weather and ongoing efforts to optimize its data centers, Google reports that its data center energy efficiency improved during the fourth quarter of last year.
In October, Google claimed that its data centers use "nearly five times less energy" than conventional data centers to power and cool servers.
On Tuesday, Google announced further improvement in its data center efficiency metric -- Power Usage Effectiveness, or PUE.
PUE is a measure of total facility power usage divided by power consumed by IT equipment. With a PUE of 2.0, for example -- a typical industry figure -- for every watt consumed by servers, an additional watt is required to cool the equipment and to distribute power. Using less power for cooling and overhead brings the PUE number closer to 1, which represents perfect efficiency.
In the fourth quarter, "our average power and cooling overhead in these facilities was 16%, bringing the overhead for the trailing 12 months to 19% (down from 21% a quarter earlier)," said Urs Holzle, Google's senior VP of operations, in a blog post. "For comparison, a recent EPA report put the overhead of the average enterprise data center at 100% or higher" (PUE of 2.0 or higher).
Google's quarterly energy-weighted average PUE fell to 1.16 from 1.22 in the third quarter. As a point of comparison, Sun Microsystems in April said that its data center in Santa Clara, Calif., had achieved a PUE of 1.28.
Microsoft in December discussed plans for its Generation 4 Modular Data Center, which the company hopes will deliver an average PUE of 1.125 by 2012.
While Google's energy efficiency is widely admired, some see spin in the company's numbers. For instance, Google explains on its data center efficiency Web page that "we only show data for facilities with an actual IT load above 5MW, to eliminate any inaccuracies that can occur when measuring small values."
The author of the Tech Hermit blog, who doesn't identify himself but claims to have spent his career working in data centers, suggested in October that Google's statistical selectivity is a way to eliminate numbers that would make the company's average PUE worse. He argued that Google's numbers are further skewed by the company's decision not to include in its measurement of hardware power usage things like electrical losses in a server's power cord.
Google plans to disclose further information about its data center energy usage at the CeBIT Conference in Germany in March, Holzle said.