Herding Air Molecule Cats in the Data Center10 min read
What is the number one most important airflow management discipline in the data center? Let’s take an imaginative leap that your boss has paid a couple thousand bucks for you to spend a few days sitting in conference halls at an ultra-posh hotel in some downtown mega-metropolis – no, better yet, you’re in Orlando or San Antonio or San Diego. You’re sitting there armed with notepad, pen and Styrofoam cup of coffee and you see me up on the stage and I hit you with this question about the most important airflow management discipline. Imagine that this is the perfect time to be attending this conference—it’s the first day and you are wide-awake and eager to participate. Now, this really happened recently and for the first time in my nearly forty years of speaking at conferences I got that prime slot on the first day right before coffee break and my question actually produced answers:
- Use blanking panels
- Use floor grommets
- Control supply temperature
- Hot aisle – cold aisle
- Control temperatures at the server inlet
- Hot and cold aisle separation
- Reduce fan speeds
- Raise temperatures
- Active DCIM
- Variable speed controls on everything
- Pressure differential feedback
- Turn the job over to a good colo provider
You are on a roll and these are all really good answers, and they are all really wrong answers. The single most important airflow management discipline in the data center is for all the ICT equipment to breathe front-to-rear.
I don’t think it would be an unfair stretch to assert that everybody who spends any kind of time in a data center knows the basics of hot aisle – cold aisle separation and the associated value of rack-mounted equipment that breathes from front-to-rear. After all, the granddaddy of data center standards clearly and simply asserts:
Equipment should be placed in cabinets and racks with “cold” air intake at the front of the cabinet or rack, and “hot” air exhaust out the back. Reversing equipment in the rack will disrupt the proper functioning of “hot” and “cold” aisles. Equipment that uses the front-to-rear cooling scheme should be used so that it does not disrupt the functioning of hot and cold aisles.1
In fact, every relevant standard and industry consortium stresses the efficacy of front-to-rear breathing rack-mounted ICT equipment in data centers. The BICSI standard addresses this recommendation even more directly than TIA-942:
Equipment should be installed in cabinets with the air intake oriented toward the front of the rack or cabinet and the air exhaust oriented toward the rear of the rack or cabinet.2
The inspiration for the early data center standards was likely the set of Telcordia (via Bellcore) standards for central offices, which pre-dated the transition from “computer rooms” to “data centers.” The ASHRAE recommended airflow protocols for data center equipment of front-to-rear, front-to-top and front-to-rear & top closely align with the telecom equipment recommendations in Telcordia GR-3028-CORE.3 While ASHRAE recommends that rack-mounted equipment should follow the front-to-rear protocol, they suggest that cabinet-installed equipment can follow any of those three recommended airflow protocols; however, my experience has been that top exhaust will still compromise the integrity of hot aisle – cold aisle separation, as illustrated in the CFD model graphic in Figure 1. This compromise is exacerbated by cabinet top fans, but will still be problematic unless captured by a cabinet chimney or hot aisle containment boundary and directed into an isolated return plenum.
The telecom standard actually requires rather than recommends rear exhaust and then allows exceptions to the front aisle air inlet recommendation if the protocol is effectively converted to front-to-rear by some mechanism of baffles or air deflectors.
These best practices for data center energy efficiency further assert that any non-compliant equipment needs to be located apart from the main area of the data center with the hot and cold aisle separation. My assertion, however, is that no such compromise is either needed or justified.
Clearly, the front-to-rear airflow protocol is on everybody’s lists of required airflow management practices and equipment definitions. That being said, what makes it the number one most important factor? Consider the other behaviors and disciplines:
Blanking panels without exclusive front-to-rear breathing equipment could prevent bypass air from entering the rear of the cabinet or the sides of the cabinet to mix with exhaust air and thereby create a more tenable input air temperature.
Floor grommets without exclusive front-to-rear breathing equipment could prevent bypass air from entering the rear of the cabinet or the sides of the cabinet to mix with exhaust air and thereby create a more tenable input air temperature.
Hot aisle – cold aisle separation, by definition, would not exist without exclusive front-to-rear breathing equipment.
Controlling supply temperatures by feedback from the equipment air intake points would not be practical with 15-20˚F differences in input temperatures, depending on the airflow protocol direction.
Changing from a return thermostat temperature control to a supply control temperature will not be effective when some equipment will be ingesting input air mostly from the return side of the data center row layout.
Reducing fan speeds may not be practical when some amount of bypass air needs to be produced to drop the temperature in the “hot” aisle to support non-standard breathing equipment.
Raising supply air temperatures may be disastrous when some equipment may be ingesting air that is input temperature plus equipment temperature rise (ΔT), or even exhaust temperature plus ΔT.
I am pretty confident that all of our recommended airflow management best practice behaviors and disciplines are going to be ineffective, unnecessarily expensive, and likely counter-productive if not deployed in conjunction with a data center fully implemented with front-to-rear breathing equipment. That observation would seem like it should be the conclusion of this discussion were it not for the fact that there are over 300 different models of ICT equipment (mostly switches) that move air in some other airflow protocol than front-to-rear, and some of these are so ubiquitous that they are essentially “Kleenex” specifications for most data centers.
I have been writing and speaking about data center cooling efficiency and effectiveness for many years and once this field actually became a discipline, all the vast rewards for disciplined behavior have depended on filling the data center with equipment that breathed front-to-rear. Isn’t it nice to know that no matter what competing objectives may produce in terms of apparent non-compliant airflow protocol, there is absolutely no reason to compromise all the benefits derived from airflow management best practices?
1. TIA-942, Telecommunications Infrastructure Standard for Data Centers, p42
2. ANSI/BICSI 002-2011 Data Center Design and Implementation Best Practices, paragraph 184.108.40.206
3. Thermal Guidelines for Data Processing Environments, 4th Edition, ASHRAE TC9.9, p.59
Data Center Consultant
Let’s keep in touch!