ACCORDING TO YOUTUBE, more than 100 hours of video is uploaded to YouTube.com each minute. And that’s just a tiny portion of the trillions of pieces of information—from financial data to family photos—that is uploaded and stored on servers each day. These huge vaults of digital information are often housed in industrial-sized data centers, some in excess of a million square feet. Their daily energy consumption can be the equivalent to that of a small town. Companies like Apple, Facebook, and Google invest billions of dollars building massive data centers around the world. Globally, spending on these facilities will be an estimated $149 billion this year, according to the research group Gartner.
If one of these centers were destroyed by fire, however, the business loss could be many times greater than the hundreds of millions of dollars worth of equipment inside.
“It would be hugely destructive not only to the people who own the data center, but also to all the businesses it serves,” says Jonathan Hart, NFPA staff liaison to NFPA 75, Fire Protection of Informational Technology Equipment. “That’s why they are so well-protected and why there is such an emphasis on detecting and stopping fires when they are small and can’t do too much damage.”
However, critical gaps in knowledge about data center fire prevention persist. To close one of those gaps, the Fire Protection Research Foundation, in partnership with Hughes Associates and FM Global, recently undertook a project to study fire detection measures in these facilities. The results were published in September in a report titled “Validation of Modeling Tools for Detection Design in High Air Flow Environments.”
In the last five years there have been vast changes in how the equipment in data centers—servers, circuit boards, electrical cables, and more—is housed and maintained. The machines can get very hot from the large energy loads necessary to power them; to cool them more efficiently, companies have turned to a method known as “contained airflow,” where cool air is pumped in and hot air expelled through enclosed chambers between the machines. The resulting conditions create what’s classified as a “high airflow environment,” which Hart characterizes as a mild breeze. In the event of a fire, that moving air can dilute the products of combustion, making detection tricky. To make matters worse, circuit boards, wires, and other potential data center fuel sources tend to smolder, rather than fully combust, producing low amounts of smoke in a fire’s incipient stages.
The latest edition of NFPA 72, National Fire Alarm and Signaling Code, states that “there currently are no quantitative methods for estimating either smoke dilution or air flow effects on locating smoke detectors.” The new research aims at changing that and providing technical committee members for NFPA 72, NFPA 75, and NFPA 76, Fire Protection of Telecommunications Facilities, much-needed guidance as they consider revisions for the next editions.
Scenarios, data, and guidance
As part of the project, researchers at Hughes tested burn rates, heat, and smoke production of various fuel sources found in a typical data center, such as circuit boards and electrical cables. FM Global then built a mock-up data center, 16 feet wide by 24 feet long by 16 feet high, to see how the test data compared to a fire in a more real-world setting. The collected data was then plugged into a validated computer model “to do a series of simulations with various detector spacing, types of fires, and airflows,” says Amanda Kimball, a research project manager with the Research Foundation. “The simulations gave us guidance on detection placement and installation. ”
The study concluded that a smoke detector placed on the ceiling or in the ceiling plenum—a cavity space in the ceiling for air circulation—is most likely to detect smoke sources. Researchers also found that, in data centers where the air recirculates, any placed smoke detector should eventually detect the smoke. In centers with no recirculation, smoke detectors should be placed every 10 to 15 feet to reliably detect smoldering smoke sources, or every 20 feet to detect flaming sources.
“The study gives some useful guidance to people in the field,” Hart says. “The technical committee will have to figure out whether the findings are sound enough to create requirements or just offer guidance. But it is safe to say there will be something on this in the next edition of NFPA 75.”