Keeping a data center cool is tricky because new servers require 10 times the power and 30 times the airflow of water-cooled mainframes with similar footprints. In addition, while legacy cooling designs can handle the length and width of a data center, designers must now contend with another factor—height.

For an idea of how hot air-cooled mainframes and racks of servers get, think of a 7-foot tower of toaster ovens. Also, power density has increased significantly as some servers are so compact, not even 2 inches tall, that a typical 84-inch high cabinet can contain up to 40 of them. Height is certainly an issue, too, as recent computer installations have been as tall as 3,500-10,000 W per 7 square feet of cabinet footprint.

In times of insufficient cooling, these critical servers can cease functioning. They require effective, redundant and dependable cooling distribution designs, which are not covered in the ASHRAE Handbook on HVAC applications for data centers, which was issued in 1987 and has remained unchanged since then. While the Handbook addresses the importance of static pressure, it doesn't take into account other factors, such as raised floor height, cooling load and air volume, among others, and how they interact. In fact, if you apply the Handbook's recommendations today, you would end up degrading the static pressure that is needed to convey air around the data center.

Comprehensive evaluations of over 3 million square feet of raised floors across the country have revealed that in order for an average data center load of 60 W per square feet to have sufficient airflow for cooling, it needs at least 0.025 inches w.c. static pressure. And the average static pressure throughout the data center floors tested did not meet this benchmark. (Keep in mind that static pressure is needed to let airflow stream upward through raised floor systems).

The Handbook advises designers to create a perimeter of inward-facing computer room air handlers (CRAHs). And whenever more cooling is needed, engineers place new CRAHs in areas of the floor, without orienting them in a specific direction. Real-world testing has shown, however, that the CRAH distribution design recommended by the Handbook creates distinct air plumes below the raised floor because of shear boundaries along each plume's edges. These shear boundaries prevent air distribution from being redundant, resulting in isolated air patches.

Because it creates these separate air plumes, the CRAH distribution design does not promote adequate cooling. In fact, even the data centers that currently produce enough static pressure will fail to sustain increased cooling loads if they continue to follow this flawed design concept.

Another misguided practice that the Handbook espouses is placing perforated tiles as close to the CRAHs as possible. While it may make sense at first to position critical cooling loads closest to the cooling source, this configuration endangers electronic equipment. A classic example of wrongly applied legacy design, this practice is based on a misconception of how air behaves underneath a raised floor.

This arrangement assumes that boosting CRAH output will propel air up through the tiles and that perforated tiles closest to CRAHs will convey the most cool air. Such expectations are based on the belief that raised floors can be likened to static pressure-filled balloons when in fact, raised floors have both areas of still air and areas of high air velocity.

In most of the data centers examined, there were areas of the raised floor in which warm air from the room actually streamed down into the raised floor plenum—instead of cold air flowing upward through the tiles to cool down the computers. Indeed, this is a surprising occurrence—since it seems strange that warm air could flow downward into a pressurized raised floor—but we can use fluid dynamic models to understand this tendency. It happens when the pressure above the cooling plenum exceeds the pressure below, creating a vacuum and allowing air to flow down through the tiles.

In areas of still air below the raised floor, vacuum pressure was not evident. But once the local air velocity of the area underneath the floor went above 566 fpm, vacuum pressure resulted and the warm air headed downward into the cooling plenum. This tendency means that placing a CRAH near a computer cabinet will not efficiently distribute cool air because if CRAHs blast chilled air at more than 2,500 fpm, the air won't be able to slow down to below 566 fpm to avoid creating vacuum pressure. Close placement will not allow it to sufficiently lose speed.

To improve cool air distribution, designers should place perforated tiles over still-air or low velocity areas under the floor. Since air loses speed as it spreads out, they should actually position the perforated tile as far from the CRAH as they can. Designers should always keep in mind that simply adding CRAHs will not bolster cool air distribution because this practice increases air velocity, creates isolated air plumes, compromises static pressure beneath the floor and thus, impairs air circulation.

These design complexities shouldn't deter designers from depending on raised floor cooling, however. It's still the most efficient, affordable and flexible method of keeping computer room airspace cool. Instead, designers should try a different approach to raised floor cooling called cascading static regain (CSR). Instead of inward-facing constant-volume CRAHs placed side-to-side, the equipment is arranged in rows, lined up front to back. In the latter arrangement, distinct plumes are not created; rather, this layout causes the plumes of upstream and downstream CRAHs to blend because of the turbulent flow from downstream CRAHs. And very importantly, static pressure actually goes up as the cascading CRAH plumes are conveyed along the length of the data center.

A CSR design fulfills the four criteria of an effective cooling system. First, it allows designers to predict how airflow and static pressure will be distributed in a raised floor environment. Second, it sustains static pressure and lets engineers place computer hardware over high velocity areas without worrying about inadequate static pressure. Third, because of the layout of CRAHs, underfloor air plumes can blend together, thus boosting access to redundant capacity. Finally, all this is achieved without increasing cooling output or occupying more floor space with CRAHs.

In fact, the CSR design has already yielded dramatic results for a large Midwest data processing company. The firm previously had a legacy air-distribution design that was causing the static pressure beneath the raised floor to deteriorate. For 16 years, this low static pressure provided inadequate cooling in certain areas of the company's 50,000-square-foot data center. Also, some areas would suddenly become warm and then cool again.

Through testing, it became evident that the previous cooling system produced separate air plumes. Since these underfloor plumes did not blend together, redundant capacity was not available for critical process loads.

After repositioning the CRAHs following the CSR design, air distribution improved dramatically. What's more, the renovation did not entail any additional air-conditioning equipment—just the relocation of existing equipment. Because the new layout allowed plume shear boundaries to mix, cool air was conveyed over wider areas of the data center. Maximum, minimum and average static pressure readings all went up. The most marked increase occurred in the areas with the lowest readings, which jumped up by over 200%. Even more significantly, the average static pressure value showed improvement, increasing by 26% throughout and revealing the substantial benefits of using a CSR design over the entire raised floor area, in the reconfigured data center.

Source: Data Center Cooling
Edward C. Koplin, P.E.
ASHRAE Journal, Mar. 2003
http://www.ashrae.org