We have talked about data center airflow management for high and low-density deployments, for raised floor and slab floor data centers, for legacy servers and newer heat tolerant servers, and many other applications and scenarios. Simply put, we have covered many variations on this theme, but for the most part, we have either focused on or assumed brick and mortar structures either built to house a data center or into which a data center has been established—stick-built facilities my father-in-law would have called them.

What about containerized data centers?

My apologies to all you designers, builders, sellers and users of modular containerized data centers for the oversight and the tardy step on board this train of thought. After all – it’s not like these data center containers have been hiding under a rock. Granted, there remains some disagreement on the scale and the critical mass of this phenomenon, but nevertheless, it is clear that modular containerized data centers are not a well-kept secret.

As might be expected, even if containerized data centers might not be for everybody, this forecasted growth is spurred on by promises of efficiency and effectiveness benefits. Typical benefits for these data centers focus on the ease of putting smaller, fully utilized increments of capacity into service to more closely scale capital investment to actual business demand and avoid operational inefficiencies frequently associated with low utilization during ramp-ups. Further, modular containerized data centers are touted for flexibility in site selection and speed in deployment – not critical attributes for most enterprise data centers, but certainly important for some industry segments. Providers of these data centers also make the case that the design and construction process for a manufactured solution is more controlled and repeatable than the process for a site-specific engineered solution involving a wide spectrum of unrelated and potentially uncoordinated building trades, thereby resulting in both lower costs and higher quality (at least in terms of conformance). In addition, according to a paper published by The Green Grid, these containerized data centers “can be expected to…have lower operating and capital costs, and be equipped with higher density and energy-savings targets.” The Green Grid paper further claims such data centers can optimize their airflow and cooling systems for better efficiency and reduce energy through tighter set point management and server inlet temperature variation, which brings us finally to today’s topic: airflow management considerations for containerized data centers.

And what are those airflow considerations? They are basically the same considerations as you would have for any data center. Upsite Technologies’ 4 R’s of Airflow Management define those considerations as well as you could ever want – manage the airflow under the raised floor; manage the airflow in the rack; manage the airflow in the row; manage the airflow in the room. The container modules may put some slightly different twists on these four R’s, but the discipline is essentially the same.

Some containerized data centers include raised floors and many do not. It doesn’t matter. The discipline is merely to make sure that whatever plenum or duct or space through which chilled air is distributed to the IT load is sealed to prevent the evacuation of any air anyplace except to the point of use by said IT load. Most containerized data centers will deploy some variation of a commercial server rack, though some will have custom or proprietary racks integrated into the container. It doesn’t matter. The discipline is still to make sure the front of the rack is completely isolated from the rear of the rack to prevent bypass and recirculation airflow, thereby assuring that the only path between supply air and return air is through IT equipment. Some containerized data centers may have multiple rows of racks and some may have only a single row of server racks. It doesn’t matter. The discipline is still for the row to define the boundary between cool air and warm air and provide a means for all IT equipment to move air in only one direction – from cold to hot. Finally, the data center room may include thousands of racks in many rows with heterogeneous equipment distributed randomly or a single row of homogeneous equipment in arrays of perfectly balanced racks. It doesn’t matter. The discipline is still to control temperature to allow it to approach as close as possible (within1-2˚F) to a specified maximum server inlet temperature and to control airflow volume to maintain an absolute minimal pressure differential between the cold air masses and warm air masses.

The considerations for airflow management for a traditionally built data center versus a modular containerized data center are virtually the same. Some of the detailed execution may differ, but the goals are the same. Those differences should become readily apparent within the context of the specific environment, just as there are differences between a traditional data center on a raised floor versus a slab data center with no mechanical plant.

The results of applying these disciplines are the same:

• Everything is adequately cooled
• All wasted fan energy is eliminated
• Cooling coil efficiency is maximized by those coils seeing higher ΔT’s
• Cooling plant efficiency is maximized by producing higher temperatures
• Free cooling proportionate hours are increased through higher supply temperatures

The vendors of containerized data center systems will make claims for the superiority of their solutions for delivering these effectiveness and efficiency results in comparison to traditional data centers. To the degree that “tradition” means poor discipline and control, those vendors would be right. However, all these practices have been preached by me and many others over the years and can deliver similar results in a brick and mortar data center when implemented correctly and guarded zealously. Interestingly, for example, the folks at Microsoft, after a much-publicized foray into containerized data centers have determined what they learned about airflow management and power distribution in containers can be applied to brick and mortar colo facilities and are therefore migrating back. Containerized data centers will have the singular advantage of having all the work done for us so that it may take us a little longer and a little more creativity to undo the discipline than it does in our stick-built facilities. Nevertheless, what constitutes good airflow design for a traditional data center and for a containerized data center is based on the same principles. What constitutes good airflow management discipline for a traditional data center and for a containerized data center also depend on the same principles.