In its continuing quest to improve the efficiency of its data centers, Google has found a new solution: machine learning. Jim Gao, an engineer on the company’s data center team, has been hard at work on building a model of how data centers perform based on a variety of factors, according to post to the company’s official blog.
The model is able to predict a data center’s Power Usage Effectiveness – a measure of how much power is being used for computing, and how much is being wasted – with 99.6 percent accuracy. That means Google can now use Gao’s system to figure out how it can best make use of its power and allocate resources in order to minimize how much power is being wasted.
In one case, the company had to take a couple of servers offline for maintenance in one of its data centers, and was able to maintain the center’s efficiency during the process by tweaking its cooling system according to Gao’s model.
Gao has helped solve a significant challenge: as he pointed out in a white paper about his system, modern data centers are incredibly complex.
“The sheer number of possible operating configurations and nonlinear interdependencies make it difficult to understand and optimize energy efficiency,” he wrote.
Now, companies interested in improving their own data centers can follow Google’s lead. Gao’s white paper lays out how he was able to use a neural network to learn about improving efficiency for Google, and provides a framework for other people to build their own systems.
As Internet use continues to grow, tech companies need to find ways to make their ever-growing data centers more energy efficent and environmentally friendly. Microsoft Research has been developing a method for directly powering servers using fuel cells, rather than relying on a larger central power plant to supply juice to a rack of computers. Those innovations will be necessary to help minimize the impact of running tons of servers to power users’ demand for information.
Energy efficiency is a contentious issue in the cloud services industry. Greenpeace released a report last month that took a number of internet companies, including Amazon and Twitter, to task for how much non-renewable energy they were using to power data centers.