How Google Uses Machine Learning And Neural Networks To Optimize Data Centers

Google has released some new research about it efforts to maximize performance and minimize energy use at data centers through machine learning today. Long story short: Google is building superintelligent server farms that can learn from their past performance and improve themselves in the future.

Google’s AI data centers are a 20 percent project – the result of an employee, Jim Gao, working on something he found interesting that falls outside of his standard job description. Google is famous for allowing its employees 20 percent of their work time to come up with passion projects and things that they wouldn’t otherwise be able to work on. Thinking, learning data centers happened to be Gao’s main area of interest.

Gao researched machine learning and then worked on building models that take in a huge amount of data Google was already tracking about its data centers including how much energy is being used at any given time by servers and other equipment, outside air temperature and more. Computers then crunch all this data, analyzing the interplay that may be impossible for a human mind to grasp, and predicting Power Usage Effectiveness, or how to use available power most efficiently for maximum computing return.

The model means that where once Google had to shut down entire server banks in order to perform service or for other reasons, it can now temporarily tweak another variable like cooling in order to maintain a much higher general level of output, saving time, energy and money.