If you’ve ever tried to find how much airflow a new server requires, then you’ve experienced the frustration of burning precious time to no avail. Finding server airflow requirements should be as easy as finding its power requirements since they are just as important, yet IT equipment vendors leave us guessing on how to properly specify cooling.
The air conditioning fans cooling your IT equipment represent roughly 10% of your data center’s entire electric bill (including IT energy), or about 40% of your total cooling energy. This energy could be much lower if we didn’t oversize our computer room cooling unit airflow. Certainly, some oversizing occurs due to redundancy, but the oversizing I’m talking about happens because designers don’t know the IT equipment airflow or deltaT. Let me explain.
The following formula quantifies how much heat a server produces OR how much heat a cooling unit rejects.
Q, measured in kilowatts (kW), is the rate of heat produced by the server or heat rejected by the cooling unit.
Airflow, measured in cubic feet per minute (CFM), is the volume of air moving through the server or cooling unit . Data center cooling units typically have a fixed airflow (some are fixed at 100% and some set airflow as a % of maximum).
deltaT, measured in Fahrenheit, is the temperature difference between the server’s exhaust and inlet air or the cooling unit’s return and supply air. Server designers design to a specific deltaT across the server.
(specific heat of air) and (density of air) are constants for air at set conditions so these are known.
When I enter values for the constants along with a conversion to 60 seconds / minute (1.08), and conversion factor to kW (3,412) I get a more practical formula:
From this formula, you can see that for a given “Q” there are an infinite number of combinations of airflow and deltaT values. For example, an increase in airflow through a cooling unit, results in a decrease in deltaT across the cooling coil and vice-versa. What this means is that when designers size a cooling unit for a given IT “Q”, it WILL remove the IT heat but not at the same airflow and deltaT of the IT equipment. What happens when the cooling unit airflow doesn’t match the IT airflow? Let’s pretend that you are responsible for buying the cooling units to cool a 1,000 kW data center (i.e. 1,000 kW of IT load), but you don’t know the total airflow required by the IT equipment.
Here are some possible airflow scenarios you can choose from, to cool this hypothetical data center, assuming no redundancy or safety factors. In each scenario, there are enough cooling units to cool 1,000 kW of IT load. Which do you choose? If I told you that the total IT airflow requirement was 130,000 CFM or deltaT was 24F, you would choose Scenario D. But you don’t know. So, what happens if you choose Scenario A? In this case, the IT equipment will be starved of cool air and will ingest its own hot exhaust air causing significant hotspots. What happens if you choose Scenario G? In this case, the IT equipment will use only the cool air it needs, and the remaining cool air will bypass the IT equipment and return back to the cooling unit without performing any thermodynamic work. Ideally, no hot spots in this scenario but you use 69% more energy than scenario D.
Herein lies the dilemma for data center design managers. Without knowing the IT airflow requirement or deltaT, it’s understandable why over-specifying airflow occurs. But this means that I’m spending more capital for cooling units, and paying indefinitely for fan energy I don’t need. What this industry needs are for IT equipment vendors to provide the typical and worst-case airflow or deltaT requirements on their spec sheets. I realize that IT equipment fans increase or decrease airflow upon different events like high temperature or high CPU utilization. However, this is no reason for IT manufactures to forego publishing the typical and worst-case airflow requirements, so that cooling specifiers don’t oversize your budget and energy bill.
I would appreciate any comments on your experiences with sizing cooling units for airflow requirements. If you are a designer, how do you manage this issue today?
6 years ago
Good coverage of the airflow and delta-t
However, the problem is that IT airflow is dynamic..
6 years ago
Thanks for your comment. The dynamic nature of IT airflow is exactly why server vendors should provide both average and worst case airflow. However, ultimately I would like to see real-time airflow requirements from all IT equipment. With this information we as a cooling industry can be more proactive in saving energy.
6 years ago
Ultimately there are methods of controlling against this, and I have recently done a project where the fan speeds had low setpoint and then modulated against the effective deltaT between supply and return air – the end result was encouraging and did minimise energy usage.
Things to watch out for are particular items of kit that could be operating at low and high deltaTs effectively mixing the return condition.
5 years ago
This is why customers need the Active Flow Controller (AFC). Hands-free cooling adjustment to any CFM/kW.
4 years ago
Yes, it may be a dilemma, but not quite as disastrous as the scenario review suggests. Remember, we have fan laws working against us in the servers, but for us in our cooling units. For example, if we spec the space out for scenario G but reality turns out to be scenario D, we only need to run our cooling units at 60% capacity, resulting only 1.848kW fan power per unit, or 37kW fan energy for the 20 units as opposed to 101 kW to operate 12 units at capacity. Crazy investment, but amazingly it would pay for itself well inside the life of the data center.
3 years ago
Reading this again, and still agree with the dilemma. And maybe we can get away with estimating a cfm value, regardless of the room configuration. Just say we need x – y cfm on an overall average with the understanding that any room will not be at a thermal equilibrium. And I think most of us would be OK with that in a traditional raised floor space, regardless of what type of tiles are used, row widths, cabinet depths, etc.
The bigger issue is what happens when we use row based systems – APC InRow, Vertiv XD, Rittal LCP and others. Now, we are concentrating fan capacity and heat removal at a very local level and perhaps at a more efficient level. Do the numbers go down? Up? Do installation specifics – number of devices, RU of each appliance (1RU vs Blade), application become more important?
Still looking for answers