A Little Data Center Containment Can Go A Long Way14 min read
Some data center airflow containment is better than no containment at all. Usually. Under the right conditions. I have always been a proponent of maximizing cold and hot air separation in data centers. After all, a mere 10% loss in full containment could cost a couple of months or more of free cooling and could increase chiller and cooling fan energy costs by 20% or more. That being the case then, why would anyone consider partial containment? For a new space purpose-designed and purpose-built for a data center, it never makes sense to compromise on maximum separation containment. Just in case you have missed any of my previous fifty-plus blogs or fifteen years of white papers and conference presentations, let me make that as transparent as possible: It NEVER makes sense to compromise full containment in a purpose-designed and purpose-built data center space. Nevertheless, there are plenty of legacy data centers hanging around from the days when containment was still a leading edge concept. Also, spaces are ranging from old central offices to historic urban dairies that are being converted to data center space. Sometimes these upgrade projects or retrofit projects come complete with a veritable plethora of obstacles to full containment. In these situations, partial containment such as cabinet chimneys and CRAH plenum extensions that don’t connect to enclosed return air spaces, or partial barriers mounted on top of cabinets that don’t intersect a ceiling, or end of aisle doors may be all the containment that is readily feasible. Each of these containment approaches comes with challenges, and each can provide a pathway to significant energy savings and considerable density increases when applied with a little intelligence. Today, I am just focusing on end-of-aisle doors and what considerations to apply to different room layouts and configurations for the best impact on data center containment.
Before we can reasonably evaluate the efficacy of a solution for different application environments, we need to be sure we have criteria for reasonably making comparisons of likely outcomes. Nothing new here: these criteria have been my go-to benchmark variables for years. First, are there hotspots, based on the stakeholders’ definition of a hotspot. Second, what is the difference between the cooling equipment supply temperature and the highest measured server inlet temperature? Third, how much air is being supplied? Fourth, what does the solution cost or, more importantly, what is the ROI and payback or internal rate of return (IRR)? For this particular exercise, we shall assume that any hotspot issues have already been solved (possibly at great expense) so our partial containment will merely contribute to reducing the expense of preventing hotspots. Also, since the solution I will be discussing is end-of-aisle containment doors, that cost will not vary from scenario to scenario, so cost is not a critical criteria in this benchmarking and comparison of scenarios. Therefore, air volume and ΔT between supply and maximum applied temperature are the criteria which will indicate whether or not this particular solution is beneficial for a reviewed application criteria.
Air volume is a relatively straightforward metric for data center efficiency because it pegs directly to fan energy. Almost directly. Since the fan affinity laws tell us that any reduction in fan flow results in energy savings that is the cube of that flow reduction (80% fan volume = 51.2% rated energy > .83 = .512), small decreases in airflow volume produce more substantial savings in energy dollars. If some baseline target has value for the analysis part of a project, an excellent place to start is to note the current total air volume (CFM) supplied into the room and then to calculate the actual demand. For a reasonable estimate of real demand, multiply the UPS load times 3100 and divide the product by 20. The chances are that calculated demand will be half or less than half of what is being supplied. If on the other hand, we are evaluating scenarios for a new data center planned for a retrofitted space, a CFD study would produce estimates of airflow volume requirements with and without partial containment.
The ΔT between supply air temperature and maximum server inlet temperature tells us what the supply temperature will be, or could be. There will frequently be a certain apple to oranges element to this comparison, but that is nothing about which to be dismayed. More than likely, in an existing data center with no containment, the set point is aligned with the cooling units’ return air intake, and it has been dialed in to make sure no server sees a higher inlet temperature than whatever has been determined for the operating specification or service level agreement. The supply temperature will be somewhere around 18˚F lower than that set point. Therefore, a 70˚F set point will likely result in 52˚F supply air being discharged from the cooling units. While the ΔT between supply and maximum server inlet should be measured, it should be no surprise if that difference approaches or even exceeds 30˚F. With containment, the thermostat setpoint is typically abandoned, and a supply temperature is set. The benchmark for full containment is for that ΔT to be in the area of 2˚F – 3˚F. In partial containment that ΔT will be somewhere greater than 3˚F but significantly less than 30˚F. That ΔT is an important metric because in the legacy “before” condition with a 70˚F set point and a maximum server inlet temperature approaching 80˚F, that supply will be in the 50’s. On the other hand, in partial containment where that ΔT may be 10˚F for example, we can maintain that same 80˚ maximum server inlet temperature with a supply temperature of 70˚F. The economic impact of that 18˚F higher supply temperature will be a dramatic increase in chiller efficiency – I would put that at 27%, but it could be anywhere from 18% to 70%, depending which vendor is trying to sell you something.
Cold aisle containment on a raised floor is typically cited as a preferred retrofit path for upgrading an existing data center, and this application can also deliver excellent results for partial containment with end-of aisle doors. Without any overhead boundaries, such a partially contained cold aisle has the potential for producing bypass airflow, which can be mitigated by a good pressure feedback system to control cooling unit fan speed or more directly with directional floor tile grates that increase the likelihood for chilled air to be ingested before escaping the cold aisle. Having air handlers well-placed facing down the length of hot aisles, where supply and return paths are parallel to the alignment of the rows of cabinets will enhance this partial containment layout by using the cabinets themselves as a containment barrier while the end-of-aisle doors prevent wrap-around re-circulation at the ends of the rows. Typically, we do not like to see air flow paths perpendicular to the rows of cabinets. If we had a single cold aisle with just two rows of cabinets and cooling units facing the backs of cabinets in both hot aisles, the direct return air path essentially makes the cabinets part of the partial containment barrier. Conversely, a similar setup but with cooling units on only one wall with a vertical airflow would result in return air passing over the cold aisle, necessitating an increase in airflow volume as well as producing an increased ΔT between supply and maximum server inlet temperature by re-circulation.
Hot aisle partial containment on a raised floor is going to have somewhat more limited opportunities for reducing airflow fan energy and increasing supply temperature. The ideal scenario would be the presence of a suspended ceiling equipped with return air grates and directly coupled to the return air intakes of the cooling units with plenum extensions. Such an airflow configuration would allow the maximum flexibility for locating cooling units about the rows of server cabinets. On the other hand, if the return air path remains in the room, hot aisle partial containment becomes problematic. If the cooling units are located at the ends of the aisles, hot return air has to rise over the end-of-aisle doors, thereby reducing barriers to re-circulation into the cold aisle. Similarly, if the cooling units are perpendicular to the rows of cabinets, return air has to pass over the cold aisles. In both cases, lower temperatures will be required to compensate for the re-circulation. Partial hot aisle containment on a raised floor can be useful with ducted or suspended ceiling return air path; otherwise, you will not see any containment benefits.
Cold aisle partial containment on a solid floor in conjunction with up-flow cooling units and ducting to deliver supply into the partially contained cold aisle can minimize cooling fan energy and utilize higher supply air temperatures to capture chiller and free cooling savings. With airflow parallel to rows of cabinets, the end-of-aisle doors will minimize return air re-circulation into the partially contained cold aisle. Locating cooling on walls to create return and supply paths perpendicular to the rows of cabinets will necessitate re-circulation on the return path unless the data center consists on only two rows of cabinets and one partially contained cold aisle and there are cooling units on both bounding walls for return air path segregation. This separation cannot be maintained in a space with three or more rows of cabinets.
Hot aisle partial containment on a solid floor will be much more restricted in ways to deliver fan-energy savings and further savings associated with higher supply temperature. Downflow cooling units mounted on pedestals with the airflow perpendicular to the rows of cabinets, with cooling units facing the fronts of both rows of cabinets will contribute to fan-energy savings and allow for significantly increased supply temperature. Some degree of supply over-production would likely be required to help provide a barrier to return air passing over the cold aisle. The end-of-aisle doors protect against end-of-row wrap-around re-circulation. This scenario would only work in a space with two rows of cabinets. Delivering chilled supply air to additional rows would be problematic to downright impossible. Downflow cooling units at ends of rows would be susceptible to short-cycling and necessitate compensatory extra flow volume and lower temperatures. Upflow cooling units are a bad idea with partially contained hot aisles; regardless of location, re-circulation would require higher air volumes and more moderate temperatures.
End-of-aisle doors represent a viable partial containment solution for many application scenarios. The objective is to minimize fan energy and to minimize the differential between supply temperature and maximum measured server inlet temperature, therefore enabling increased supply temperatures to realize chiller efficiency savings and access to more free cooling hours. Raised floors and slab floors present different opportunities and hot aisle partial containment versus cold aisle partial containment present different restrictions. In all cases, there is a scenario by which end-of-aisle doors can reduce the gap between airflow volume supply and demand and can enable chiller and economizer savings through optimized temperature settings.
Data Center Consultant
Let's keep in touch!
Airflow Management Awareness Month
Free Informative webinars every Tuesday in June.
Cooling Capacity Factor (CCF) Reveals Data Center Savings
Learn the importance of calculating your computer room’s CCF by downloading our free Cooling Capacity Factor white paper.