Companies like Google and Microsoft could save millions of dollars a year by quickly shifting their computing workloads to data centers where electricity is cheapest, according to a new study from MIT, Carnegie Mellon University and networking company Akamai.
But that result would depend on a lot of what-ifs, such as having data centers with servers that use less energy when idle than when they're processing data, the study notes.
Still, even a modest reduction in the prices paid for electricity could save a lot of money for big companies, according to the study, "Cutting the Electric Bill for Internet-Scale Systems."
Google, for example, uses about $38 million in electricity per year, the researchers estimated. A 3 percent cost reduction would be in excess of $1 million for an energy bill of that size, the report said.
The study tracked electricity prices in 29 U.S. locations, as well as information from Akamai's operations.
But there's a catch – Akamai does "not use aggressive server power management," meaning that the study's results are less useful to it than to other companies with "systems where more energy elasticity exists and workloads are computationally intensive."
That's a point made by Michael Manos, senior vice president of data center company Digital Realty Trust.
In an interview with MIT Technology Review, Manos noted that most data center hardware doesn't have the aforementioned "energy elasticity" to allow it to use much less power when idle, making it difficult to achieve the potential savings laid out in the report in real world settings.
Still, companies serving the data center market are already exploring shifting computing loads to take advantage of differences in electricity prices.
Cassatt, for example, advertises the ability to "follow the moon" to use data centers in countries where it's nighttime, and thus have lower, off-peak electricity prices, as an ability of its new edition of software.
And thin client maker Wyse has come out with software to speed virtualization across international networks, which could expand the ability to shift computing power to faraway locations with cheaper power (see Wyse: Continent-Spanning Virtualization Saves Data Center Costs).
