- Sales & CRM
- Business Intelligence
Google is using machine learning in order to increase the efficiency of its datacentre operations around the globe.
The datacentre industry has long since adopted power usage effectiveness (PUE) as a measurement of efficiency. Google’s vice president of datacentres, Joe Kava, however, has said firms need to look beyond PUE, especially as they consume more power.
With the help of Google's mechanical engineer Jim Gao (nicknamed “Boy Genius”), the company has developed a method to analyse groups of data variables churned out by their real estate. The end result, Kava told Cloud Pro, is the creation of a predictive system that keeps the firm’s PUE level within a specific tolerance range.
Google tracked its PUE data every five minutes for more than five years, meaning there were lots of variables to be calculated.
Gao devised a form of machine learning, operating on a separate terminal, to churn through the data sets. What Google found was that it could predict the PUE value of any of its datacentres across the world to a high degree of accuracy.
With that information, Google has set up alerts for its engineers, telling them when their machines fall out of a certain range – effectively telling them it may need maintenance or recalibration.
Kava told Cloud Pro: “It’s impossible for you or me to spot trends that are percentile small, but through differential calculus [the machine] can identify those interactions. You don’t have to tell it what variables to focus on, it finds them itself.”
A refresh in the effectiveness of its operations can, according to Kava, save a company millions of dollars a year in maintenance and procurement.
The new efficiency method has already been rolled out to Google's datacentres across the globe. The methodology is also available to any datacentre operators who wish to implement machine learning into their own operations. “Anything that makes the global consumption of power become more efficient can’t be a bad thing,” Kava said.
"We publish several papers on how to achieve good efficiency at even smaller enterprise level set-ups. Part of what I feel very passionate about is trying to educate the rest of the industry as well," he added.