This week’s question is: Does hard containment fully contain hot and cold aisles in data centers? And for those of you who may have been following my contributions to this blog over the years know all too well, the answer has to be: It depends. It depends on what, in fact, we call full containment. It depends on how all the holes are accounted for that are not directly addressed by the containment architectural structures. It depends on the discipline applied to maintaining design intent. And, of course, it depends on the efficacy of the containment structures themselves.

Not surprisingly, containment is so self-evident that we generally don’t feel the need to say much more than it is what it is. The folks in California have taken a shot at a definition in the run-up to their Title 24 Energy Code attention to data centers with “… air barriers such that there is no significant air path for computer discharge air to recirculate back to computer inlets without passing through a cooling system.” Conversely, the good folks here at Upsite Technology who assigned me this subject tend to think of containment as a barrier to bypass airflow. In either case, the definition of containment is functional rather than structural, so the answer to our question then is…oh, I guess it still depends on the degree of functionality. For the sake of more precision, both paths to rampant non-containment can be addressed by a definition that cites the barriers to air paths and also cites the functionality of maintaining a minimal difference between the aggregate mean temperature rise of air moving through IT equipment and the macro temperature differential between supply air and return air across the cooling source, whether that be coils, wet media, air-to-air heat exchangers, polymer tubes or free air in-and-out paths. So, if we functionally define hard containment as such a barrier, then our answer is that, yes, it can fully contain hot and cold aisles. However, while most references to containment may assume such a functional definition, actual practice finds that we are usually talking about architectural structures regardless of any understanding or lack of understanding of functional distinctions. That being the case, we are back to “It depends.”

Whether hard containment fully contains hot and cold aisles will depend on how all the rest of the holes between the aisles are addressed that are not directly sealed by the containment walls, or barriers, as it were, if we include server cabinet chimneys as an incarnation of hard containment. Some of these holes are obvious, such as raised floor cut-outs for cable access toward the rear of the server cabinet footprint – essentially part of your hot aisle geography, or blanking panels in unused rack mount spaces. Plugging these holes is generally understood as a basic ingredient to aisle containment. I have for years emphasized the often forgotten bypass/re-circulation path around the perimeter of the equipment mounting area in cabinets and this hole is still often overlooked. If we have a one inch set-back for our equipment mounting rails and that gap is left open in a 42U cabinet, that total open area equates to about 5 ½ RMU’s, which most data center practitioners today know to seal up, but they still often miss this less visible space outside of the equipment rails. In addition, if cabinets are perched on leveling feet or casters, there is another potential hole in the integrity of the containment system. While early implementations of containment were usually associated with cabinets resting directly on the floor, thereby securing this area of the containment, I have more recently seen IT deployment strategies where the cabinet is determined to be a more economic unit of technology refresh than the individual server and casters are typically standard issue with these cabinets. For these installations, the containment is not complete until this bottom hole is effectively sealed. An even less obvious hole in containment is through the servers themselves. Several years ago, the folks at Tate Access Floors conducted some experiments with containment that produced some perplexing results – the overall temperature differential between supply air and return air was much less than what was predicted by the server equipment ΔT’s. What they found was that the servers themselves provided a path for bypass airflow between contained cold and hot aisles; out-of-service servers were the worst offenders, but operational servers also allowed bypass when static pressure in the cold aisles exceeded some derived thresholds. They found that at .05” H2O static pressure in a cold aisle, you could expect around 400 CFM leakage through the servers in a 42U cabinet. Whether that pressure is managed by managing fan speeds or managing variable floor tiles, static pressure management is part of a full containment solution.

The effectiveness of full containment also depends on the human element, i.e., the discipline by which design-intent is maintained over time. The most obvious discipline is just making sure doors at ends of aisles are kept closed when not providing actual ingress or egress. Some containment solutions now offer automatically closing doors to compensate for any lapses in discipline and “Were you raised in a barn?” inclinations. Other areas of discipline relate to maintaining design-intent during the normal flow of moves-adds-changes in the life of a data center: put in blanking panels when you remove a server, replace sealing under cabinet when we wheel a new one into a containment row. Again, a less obvious aspect of this discipline has to do with those servers which have been taken out of service. The Tate study showed that idle servers were a larger path for bypass than active servers and that off servers were dramatically worse for allowing bypass airflow in containment aisles. Therefore, part of the discipline required for maintaining design-intent containment is to remove out of service servers and replace them with blanking panels.

Finally, full containment depends on the quality of the containment solution itself. If we mean by “hard containment” panels and structures that are fully rigid, for example, we might not be able to truly say we are getting full containment. Unless we are looking at a fully integrated and exceptionally well designed containment solution from one vendor – that means all elements, from the containment panels and doors to the equipment racks and all the associated wire management accessories – intentionally designed as a single unit, then we are otherwise looking for plenty of opportunities for mis-matched surfaces. In general, aisle containment barriers are going to be more effective with edges constructed with material with some degree of flex to accommodate the vagaries associated with different vendors’ products or sometimes even between models within a single vendor’s portfolio.

We expert spokespersons have promised great efficiency and economic benefits from data center aisle containment. Those benefits are assumed in various standards and codes which require or strongly recommend containment. Title 24 in California mandates containment. ASHRAE TC 9.9’s process for optimizing data center temperature doesn’t let you past the first gate until airflow management best practices are implemented. Containment just happens to be one of those “latest-and-greatest” that delivers and even exceeds the hype – that is when containment is not just some walls and doors, but includes the total package of plugging all holes and exercising some basic discipline to sustain design-intent.