Google is using machine learning and 'neural networks' to make its vast data centres more energy efficient.
The search giant revealed some of its techniques to reduce energy consumption in a white paper, highlighting the use of artificial intelligence techniques to improve data analysis across its facilities.
Google said that it was already calculating PUE (power usage effectiveness) every 30 seconds, constantly tracking things like total IT load (the amount of energy its servers and networking equipment are using at any time), outside air temperature (which affects how its cooling towers work) and the levels at which its mechanical and cooling equipment were set.
However, simple manual cross-checking by humans in Google data centres could not make the most out of the figures generated to increase energy efficiency.
Joe Kava, Google vice president of data centres, said in a blog, that Google's new system "works a lot like other examples of machine learning, like speech recognition - a computer analyses large amounts of data to recognise patterns and 'learns' from them."
Kava said, that put simply, Google's data centre models "take a bunch of data, find the hidden interactions, then provide recommendations that help optimise energy efficiency".
He said Google's system was now 99.6 percent accurate in predicting PUE. This means, said Kava, that Google can use the models to come up with new ways to squeeze more efficiency out of its operations.
For example, a couple of months ago, Kava said, Google had to take some servers offline for a few days, "which would normally make that data centre less energy efficient". But, said Kava, "we were able to use the models to change our cooling set-up temporarily - reducing the impact of the change on our PUE for that time period".
"Small tweaks like this, on an ongoing basis, add up to significant savings in both energy and money," Kava said.
Find your next job with computerworld UK jobs