Cooler Servers with Less Cooling
Many data centers supply cooling far in excess of what is required or have problems distributing cooling where it is really needed. The Uptime Institute conducted a study that pegged cooling at an average of 2.6 times what was required; the most overcooled sites often had the worst thermal problems. The most often cited cause for such inefficient cooling is a phenomenon known as mixing. Cool and warm air mix before entering server intakes and again before CRAC returns. Recent studies from various industry groups agree that managing airflow and reducing mixing can cut data center cooling costs dramatically and even eliminate the need to build new sites.
Poorly sealed cable cutouts, lack of blanking panels, and badly implemented hot aisle/cold aisle rack layouts provide mixing opportunities, Solving these issues does indeed help. However, a far less often considered, but equally important, component of mixing is the interplay between the computer room cooling units, server rack heat loads, and under-floor air distribution.
The area under the raised floor is hidden from view and often mistakenly considered to be at a relatively constant pressure and temperature. Knowledgeable data center professionals know that pressure can vary greatly. Less well known is how widely temperature can vary under the raised floor and why. This temperature variance is yet another form of mixing and can negatively affect server room cooling.
The phenomenon occurs in many typical data centers. One such facility had redundant cooling units running 24x7. Many of the computer room air conditioning (CRAC) units did not provide much cooling to the data center. The unintended side effect was that some of the CRACs instead injected vast quantities of relatively warm air under the raised floor. The under-floor temperature map (see figure 1) illustrates this phenomenon.
Turning off the lightly loaded CRACs is an obvious step, but this solution poses several problems. First, excess units often run as a redundant backup in the event of a cooling system failure. Turning a redundant unit off increases risk if one of the primary CRACs fails and the backup cannot quickly supply cooling. Second, under-floor air pressure is affected by turning CRACs off; there may not be enough pressure to force adequate air through the perforated tiles in some areas. Third, air is lazy; changing which cooling units are running will affect the return air paths. Turning off redundant cooling will increase some server temperatures and possibly reduce others. The data center’s temperature profile will vary depending on which CRACs are shut off, making manual management of this activity tricky to orchestrate – and today’s dynamic IT loads make manual management even tougher.
Providing close-coupled variable airflow to the server racks and intelligently controlling the CRACs will solve this dilemma. This solution puts excess cooling in a “Hot-Standby” mode. The expected benefits are reduced energy consumption. In addition, there is less mechanical wear on the infrastructure components. A surprising benefit is the IT equipment is cooler!
AdaptivCool’s Room Scale Intelligent Cooling, which consists of an intelligent network of sensors, air-movers, and intelligent control of the CRACs is an example of a close-coupled solution. Unlike similar solutions, it works with the existing infrastructure, allowing servers and racks to remain in place. The solution involves no liquids of any kind.
Under-floor and above-floor air movers dynamically manage airflow. The illustration in figure 2 shows the placement of an under-floor air-mover in front of a server rack. These air movers are placed strategically, based on the results of CFD analysis.
Figure 3 shows the results at a typical data center. The site experienced an average server temperature decrease of 4ºF, with 1/3 fewer CRACs online. The effect was verified repeatedly to ensure there was no time of day influence.
This result is counterintuitive, so it is useful to understand the contributing and reinforcing factors at work. Simply expressed, Delta T for a CRAC is the difference in return and supply temperature. It is greatly influenced by the heat load or total British thermal units being returned in the air and much less by the CRAC set point. The Delta T of a cooling system largely defines the system’s useable capacity and efficiency. CRACs that are highly heat loaded are at a high Delta T, run efficiently, and produce their maximum amount of cooling. Conversely CRACs that are lightly heat loaded run at a lower Delta T, have lower efficiency, and produce less cooling.
Using active air movers at key rack locations insures that cooling supply to the racks is right-sized. In addition, active air movers fix inherent distribution problems previously addressed by CRAC over-supply. Active air movers allow the use of varying combinations of CRACs so that others can be put in standby without greatly affecting airflow to the racks.
The remaining CRACs run at higher Delta T and deliver colder and undiluted air to the under-floor. Furthermore, since there is less over-supply of air, the amount of mixing above-floor is reduced. This has a reinforcing effect: Virtually all the cold air is delivered to IT rack intakes and then heated, and Delta T is further raised.
Cooling energy savings at this site is 37 percent as compared to its baseline. The savings comes from a variety of factors. Reduced mixing both above-floor and below-floor allows higher Delta T and less latent cooling at the CRACs. Fewer energized CRACs means less wasted fan energy. The entire project was completed without a single server or rack being moved and without any changes to the infrastructure. In addition, to achieve this efficiency neither a hot/cold aisle containment strategy nor consideration of attendant fire safety issues was required.