In its quest for ever-greater energy efficiency, Google is now employing a new technique that adds a bit of artificial intelligence to its data centers: machine learning.
The web giant is using what it calls “neural networks” to optimize its data center operations by monitoring how its servers behave in real-time and adjusting certain parameters accordingly. The project, developed by Jim Gao, an engineer on the Google data center team, grew out of the routine daily collection of operational data.
“We calculate PUE, a measure of energy efficiency, every 30 seconds, and we’re constantly tracking things like total IT load (the amount of energy our servers and networking equipment are using at any time), outside air temperature (which affects how our cooling towers work) and the levels at which we set our mechanical and cooling equipment,” Joe Kava, vice president of the company’s data centers division, said in a blog post. “Jim realized that we could be doing more with this data. He studied up on machine learning and started building models to predict—and improve—data center performance.”
The resulting program, which analyzes data from different variables such as IT load and outside air temperature to identify patterns and learn from them, is 99.6 percent accurate in predicting PUE. The program is likely to lead Google engineers to come up with new methods of increasing operations efficiency.
“For example, a couple months ago we had to take some servers offline for a few days—which would normally make that data center less energy efficient,” Kava said. “But we were able to use Jim’s models to change our cooling setup 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.”
A white paper detailing specifics of the machine learning program can be found here.
