Data centers are the heart that maintains the flow of information across our public and private sectors. To process a mind-boggling amount of data in just a few seconds (and stay cool while doing it), data centers consume enormous amounts of power. And as our demand for artificial intelligence (AI) and machine learning increases, so will the energy demands placed on data centers.
Figuring out how to meet these energy demands efficiently and sustainably is the mission of the U.S. Department of Energy (DOE). Federal Energy Management Program (FEMP) leads the way. DOE’s National Renewable Energy Laboratory (NREL) is dedicated to research and development in energy efficiency and renewable energy. So FEMP turned to NREL to develop a guide that federal agencies can follow to optimize energy and water usage in their data centers.
NREL researchers David Sickinger and Otto Van Geet applied their combined knowledge and experience in data center design and operations to upgrade the FEMP systems. Best Practices Guide for Designing Energy-Efficient Data Centers with contributions from Magnus Herrlin of Lawrence Berkeley National Laboratory.
The updated guide applies to traditional air-cooled data centers, “all the way up to edge data centers that have higher rack power densities and use liquid cooling,” Sickinger said. The U.S. Department of Defense’s Unified Facilities Criteria references the original version of the guide, “and that has made a dramatic difference in how they build and operate their data centers,” Van Geet said.
But federal agencies are not the only ones who can benefit from this guide. “The private sector owns data centers and can also use this guide – hyperscalers like Google, Microsoft, as well as universities, healthcare and banking sectors, for example,” Van Geet said.
Best practices are more appropriate than a uniform mandate because there is no one-size-fits-all solution for effective data center design. A data center located in the desert will have different design temperature parameters than a data center located a kilometer above sea level, for example. The guide provides specific effectiveness measures and their calculations, as well as benchmarking goals and additional resources to reference in different circumstances.
As a former facilities engineer and senior planner, Kendall Kam, FEMP’s energy program manager, is “a strong supporter of guides like these.” Data centers are expensive, so whether you’re building one or renovating one, if you can plan for energy – and water usage efficiency in advance, you can get what you pay for.”
Optimize component-level efficiency, reuse heat and maximize renewable energy
According to the guide, the main priority is to optimize a data center’s energy consumption by making its internal systems as energy efficient as possible (measured by energy consumption efficiency). The guide expands on sections focused on how to improve the efficiency of IT equipment, electrical systems, and air and liquid cooling systems. The second priority is to reuse as much waste heat as possible (measured in terms of energy reuse efficiency), then reject unusable waste heat via dry coolers where possible to save water (measured in terms of energy reuse efficiency). water use efficiency). Finally, maximize the use of energy drawn from renewable systems on-site or in the grid region (measured in terms of carbon use efficiency).
Adjust heat reuse and cooling methods
“Overall, from the compute side to the facilities side, data center technologies have changed dramatically since 2011,” said Van Geet, a mechanical engineering researcher. But the energy saving principles reflected in the 2011 version of the best practice guide are still applicable today. “We just needed to modernize it, especially with rack compute densities increasing, which was one of the main triggers for the update,” said Sickinger, a high-performance computing researcher. “And things that were just starting to emerge in 2011, like hot water liquid cooling, for example, we have a lot more detail on that now.”
Heat reuse methods are a hot topic across the industry – one of the authors’ top priorities. “Europe is way ahead when it comes to data center heat reuse, but we are seeing very growing interest in this area,” Sickinger said. “Otto Van Geet and I are involved in a heat reuse group of the Open Compute project.” And while the guide doesn’t delve into organizational structure and operations, the authors “make people think about the business rationales for keeping something on-premises versus moving it to the cloud,” Sickinger said.
Prepare for AI demands on data centers
The growing use of AI is also on the authors’ radar. “AI influences the growth of data center load, so energy and water consumption are also growing rapidly. A data center that uses AI works much like NREL’s high-performance computing data center,” Van Geet said. NREL’s data center was efficiently designed to devote only 6% of its energy consumption to cooling equipment (a typical data center needs 70% to perform this task).
“We are investigating the next generation of cooling technology for data centers, thinking about how to handle the ever-increasing power densities of racks and their processing power,” Sickinger said. The authors explore these critical questions in cooling technology through the Advanced Research Projects Agency-Energy (ARPA-E) COOLERCHIPS program.
Start a conversation with data center peers
Sickinger and Van Geet credit their continued participation in DOE-funded technical assistance projects, collaborative projects with other national laboratories, and working groups for staying abreast of the latest industry challenges and solutions. And they look forward to the conversations this updated guide will inspire.
Kam agrees, and he hopes data center stakeholders will feel inspired to reach out to FEMP. “We have many resources available to them, and this guide is just one part of the whole.” Ultimately, he would like these products to serve as a starting point for professional connections. “I would love to see a cohort of data center leaders get together so they can compare notes. I think sometimes people in the data center world feel like they’re a lone wolf, and I want that they know they’re not alone,” Kam said.
In addition to the Best Practices Guide for Designing Energy-Efficient Data CentersFEMP helps agencies build and maintain energy-efficient data centers by providing resources through its Center of Expertise for Energy Efficiency in Data Centers and encourage participation in Better Buildings Challenge And Data Center Accelerator.