Data centers power most of our day-to-day life, because they’re the backbone for many of the services, apps and systems we rely on. Everything from your e-mail, to YouTube, to Google searches, to this article goes through a data center where the information is stored and processed.
But there’s one aspect at which data centers aren’t so great: energy efficiency. Running thousands of processors, hard drives, magnetic tapes and networking equipment takes a real toll on the grid. What’s worse is all of that equipment also needs a powerful cooling system to keep it running. The BBC reports that by some estimates, up to 2% of the world’s greenhouse-gas emissions are due to data centers.
It’s no surprise then that many companies are desperate to improve the efficiency of their data centers. Some try to create more efficient systems; others simply build data centers in colder climates; still others drop their servers to the bottom of the ocean.
Even a decrease of a few percentage points of energy use can mean huge savings for companies, so it’s very impressive to see Google announce that it managed to decrease the energy spent on cooling one of its data centers by up to 40%. The secret behind this success? Machine learning.
Google’s AI division, DeepMind, has been developing a machine learning system that used historical sensor data from Google’s data centers. After crunching that data, DeepMind developed a general purpose algorithm that learned how to control cooling in real-time inside of a data center to make it as efficient as possible.
The results speak for themselves, with DeepMind’s program being able to keep cooling energy use 40% below normal levels. The results were so impressive that the system will be deployed inside all of Google’s data centers by the end of the year.
But according to Rich Evans, Research Engineer at DeepMind, the real beauty of the system is it can be deployed in other data centers and environments with no changes – hence the general purpose moniker. It can even be applied to other domains like optimizing water usage, or the national energy grid.
DeepMind’s co-founder, Mustafa Suleyman, explained that the team is already talking to interested parties outside of Google, with its algorithm being a perfect fit for many industrial facilities.
The team announced it would be releasing a white paper detailing its results and how the system was developed and implemented in the near future.