Data Center World 2022: Using AI To Cool Data Centers Yields Big Cost Savings

Data centers are intensive users of energy, but a new generation of AI-powered tools is helping them become more efficient.

Case in point is a colocation data center in the Chicago area run by Evoque, a company that bought AT&T’s data centers for $1.1 billion in 2019.

At the Data Center World conference in Austin, Texas, John Diamond, vice president of design and construction at Evoque, said that AT&T’s data centers suffered from several issues.

These included legacy technology, high and increasing power usage effectiveness (PUE), overcooling of the data center space, and lack of optimization between the white space and the chilled water plant or chiller.

Evoque partnered with BGIS to conduct an assessment of its 5MW, 100,000-square-foot data center in the Chicago area. These were the findings: It had an unfavorable 48-degree Fahrenheit chilled water supply, 112 out of 121 computer room air handler (CRAH) units were in operation at 50% fan speed. Total fan energy consumption was 430 kW.

BGIS recommended using an AI-powered platform developed by software firm Vigilent to control and automate the cooling, linked to wireless sensors that monitor the data center environment in real time.

4567

Source: Evoque

‘Leap of faith’

The solution updated the facility data continuously and applied the learning to its system to optimize the outcome, said Patty Anderson, vice president of business development at BGIS.

Related:Google is Switching to a Self-Driving Data Center Management System

For example, if three server racks were removed, the AI/ML system would adjust the cooling needs and make those decisions on the fly, Diamond said.

Over three months, Evoque implemented the solution and observed the results. The AI/ML component of the platform was continuously learning about the data center’s activities and adjusting. It was constantly analyzing the airflow, balancing the cooling capacity to the IT load in real time, and performing temperature management.

Diamond said the data center reduced its fan energy consumption rate from 430 kW to 300 kW, a 30% decrease. He estimated the annual energy savings at nearly $194,000. Moreover, there was a 20% PUE reduction and 23% cooling carbon decrease. In addition, only 65 CRAH units are being used and only one out of six chillers needs to be operational.

Evoque will be deploying this solution at its other data centers, Diamond said.

Data center personnel at first were skeptical about the AI but soon adjusted as they saw the benefits. “It’s a little bit of a leap of faith, but it’s working out,” Diamond said. “We’re very happy.”

Related:Not Just for Google: ML-Assisted Data Center Cooling You Can Do Today

This story originally appeared on AI Business, a Data Center Knowledge sister publication.