By now, everyone should know or at least suspect that there may be something to this business of data center airflow containment or maximum separation between supply and return air masses. After all, every data center standard identifies such separation as a fundamental best practice, and some state and municipal building and energy codes legislatively mandate it. Nevertheless, while one would surmise containment should be as ubiquitous as EIA-310 racks, that turns out not to be the case. As I reported earlier, a data center survey sponsored by Intel and HPE found almost 50% of the respondents did not employ some type of containment and the recent Uptime Institute data center survey found that 22% of respondents had not deployed some method of airflow containment. While the Uptime Institute survey was more current and larger, it was biased toward Uptime Institute members and tier rated data centers and therefore might be expected to represent a more sophisticated population sample. Therefore, realistically, somewhere around one-third of data centers may not have yet implemented some variety of air flow containment. The question remains: Why?

Indeed, there are a variety of reasons why there are data centers without good airflow separation. First, it costs too much and/or it doesn’t really work are two of the wrong answers. Airflow containment will always provide a data center environment that can support higher power densities and lower energy costs and will always pay for itself quickly.  So while those may be some of the excuses we hear occasionally, they do not hold up and I suspect that more often than not they are just euphemisms for a discomfort with the unfamiliar or the effort required. Nevertheless, there are some real obstacles to data center airflow containment. One of the more frequently mentioned has to do with fire suppression, ranging from the fire marshal proclaiming it violates code to just physical interference. While fire suppression and data center airflow containment is definitely compatible, the seriousness of the safety system at least merits some further discussion.

First, just as a general concept, code does, in fact, account for and allow airflow containment in data centers. NFPA-75 and NFPA-76 both contain sections on containment and specifically state, “Aisle containment shall be permitted,” and then go on to allow both integral and after-market containment and explain all the conditions which must be met. For those of my readers who have not been actively involved in the standard-writing or standard-enforcing communities, there is a hierarchy of discourse ranging from “shall” to “should” to “may” and then what we call our informational appendixes. “Shall” is the big guy – non-negotiable law that is used by standards and code writers only after lengthy meditation, deep soul-searching, interminable consensus-building and general angst. Important details include:

1. Containment materials must meet fire-proof test requirements
2. Containment structures and hot collars (chimneys) are not plenums
3. Sprinkler clearances must still meet NFPA-13 requirements
4. Heat sensors (fusible links) cannot be triggers for automatically removing barriers
5. Contained aisle with its own roof (e.g., cold air containment) is a separate room requiring its own fire suppression system.

Containment has been codified for some seven or eight years now so none of this should be a surprise to industry practitioners.

The more likely containment impediment from fire suppression is the complexity of dealing with physical interference. Physical interference should never be an issue for a new space. Data center layout should be included in the architectural plans so the fire suppression system can be designed around server cabinet row placement. Typically long runs of pre-action pipe on eight feet centers will accommodate hot air containment, whether that be aisle containment or chimney (hot collar) containment. With a sixteen feet pitch from the center of one cold aisle to the center of the next cold aisle, fire suppression on eight feet centers will run parallel to barriers and easily avoid any sprinkler clearance issues. Therefore, in a new space, simple planning removes any barriers to containment.
Installing a data center into a non-purpose-built space or into a space acquired from a non-planner or facing one’s own space that was designed and built over ten years ago when containment was still new and not a mandatory part of everyone’s toolkit, will likely present the data center owner with the challenge of working around fire suppression deployed on ten feet centers. If cold air containment is being considered, which is typically recommended as an easier retrofit to existing spaces, then, per code, that contained aisle is a separate space requiring separate fire suppression, so the layout of the pipe grid in the larger room is a non-factor offering no interference issues. For hot air containment, hot collars can be added on typically three rows of cabinets before NFPA-13 sprinkler clearance issues kick in and sprinkler heads need to be relocated.

Hot aisle containment does not necessarily need to be ruled out in a retrofit project with sprinkler fire suppression installed on a ten feet grid. Partial containment can live below the spray zone radius boundaries defined by NFPA 13 and still provide both density and efficiency benefits. Partial containment could take the form of end-of-row doors or partial containment walls that do not completely bridge from the server cabinets to the ceiling. If the return air stays in the room and CRAHs are located at the ends of hot aisles, top-of-cabinet barriers will help. If the cabinet rows are short and there are only CRAHs on one wall, the end-of-row doors should be considered for the ends of rows away from the CRAHs. If the CRAHs are located on walls parallel to the length of the server cabinets (never a good idea without full containment), the top of cabinet barriers can still keep most of the return air high enough to minimize re-circulation in the cold aisles. Return extensions on the CRAH units can help with keeping the return path higher in the room. On the other hand, if a suspended ceiling space is being used for the return air, then the partial top-of-cabinet barriers and end-of-row doors work together to maintain a much better segregation of supply air and return air.

A very useful metric for driving continuous efficiency improvement in the data center is:

1. No server inlet temperature exceeds whatever the local specification or SLA defines
2. The supply temperature should be as close to the maximum measured inlet temperature as possible
3. The airflow volume should be as low as possible to meet #1 and #2.

PUE after all, is an after-the-fact metric and provides no integral direction on what to do to improve it. The rack face ΔT and supply air volume metrics, however, can be manipulated to drive chiller efficiency, free cooling access and cooling fan energy efficiency. Modest reductions in rack face ΔT can produce significant energy savings because the lower ΔT is achieved by higher supply temperatures, driving both chiller efficiency and increased access to free cooling. The reduced air flow volume delivers logarithmic efficiency gains because of fan laws cube effect.

Fire suppression systems should never be compromised for the sake of efficiency. Fortunately, there are a variety of paths for improving airflow management without compromising fire protection and life safety.