Published on July 22nd, 2016 | by Joshua S Hill0
Google’s DeepMind AI Reduces Data Center Cooling Bill By 40%
July 22nd, 2016 by Joshua S Hill
Google’s 2014 acquisition of artificial intelligence company DeepMind has paid dividends, by helping to save the company 40% on its data center cooling electricity bills.
Google acquired the London-based artificial intelligence company in early 2014 for more than $500 million. Since then, DeepMind’s machine learning has been used for only a few research and intelligence projects. However, in recent months, Google put DeepMind in charge of aspects of Google’s data centers to “significantly improve the system’s utility.” Google had already been using machine learning to operate its data centers more efficiently, but only recently has DeepMind been charged with creating “a more efficient and adaptive framework to understand data center dynamics and optimize efficiency.”
As a result, DeepMind has managed to optimize Google’s data center operations, and to reduce the amount of energy used for cooling by up to 40%.
Google is intent on powering all its operations, including its data centers, from 100% renewable energy. But until it reaches that point, Google is also hoping to increase its energy efficiency levels. Google recently announced the purchase of 236 MW of wind energy for more of its European operations, adding to an existing 842 MW worth of Power Purchase Agreements made at the end of 2015.
The renewable energy Power Purchase Agreements are simply one facet of Google’s overall efforts to reduce its carbon footprint, with the work being done by DeepMind and other machine learning tools to increase energy efficiency being another. Cooling, as with any computer, makes up for a significant portion of its power usage. This is even more evident in data centers, with servers forced to operate continually at high usage requiring intensive cooling systems.
Prior to machine learning techniques, Google explained that “dynamic environments like data centers make it difficult to operate optimally for several reasons:
The equipment, how we operate that equipment, and the environment interact with each other in complex, nonlinear ways. Traditional formula-based engineering and human intuition often do not capture these interactions.
The system cannot adapt quickly to internal or external changes (like the weather). This is because we cannot come up with rules and heuristics for every operating scenario.
Each data center has a unique architecture and environment. A custom-tuned model for one system may not be applicable to another. Therefore, a general intelligence framework is needed to understand the data center’s interactions.”
DeepMind’s introduction into the equation served to help predict the Power Usage Effectiveness of a data center and to minimize the cooling to the necessary requirements. The graph below shows a typical day of testing for the introduction of DeepMind: