Four probes, not one
Most rooms are judged on one number: the return temperature at the cooling unit. It is the number on the display, it is the one that trends, and it is almost useless on its own. A hot return can mean the room is working properly — hot air is getting back to the unit instead of leaking round the racks — or that the units are starved of air. A cold return usually means bypass, which is the failure everybody reads as success.
The four readings this page asks for are the discharge at the unit, the return at the unit, the average rack inlet and the average rack exhaust. With the load, they split into two independent statements. The unit rise and the rack rise give two airflows, and the ratio between them says whether the room delivers more or less air than the servers pull. The inlet reading sits between the discharge and the exhaust, and where it sits says what fraction of a rack intake is its own hot air.
What the defaults are describing
138 kW of heat, 470,876 Btu per hour, 39.2 tons. The racks pull 17,233 CFM to carry 120 kW of it on a 22 degree rise — 143.6 CFM for every kilowatt of IT load. The units are moving 25,647 CFM to carry the whole 138 kW on a 17 degree rise. That is 1.49 times as much air as the servers pull, and roughly a third of what the units move never passes through a rack at all.
And yet the inlets read 74 degrees against a 65 degree discharge. Nine degrees of the thirty-one degree spread between supply and exhaust have already been lost before the air reaches a server, which is 29 percent of the intake arriving as recirculated exhaust. Airflow explains none of it — there is a surplus, not a shortage. All 29 points are leakage: gaps over the cabinets, open U spaces without blanking panels, cable cutouts without brushes, and the ends of an aisle that nobody closed off.
Why adding cooling makes this worse
The instinct when inlets rise is to add a unit. In a room short of air that works. In this room it does not, and the reason is worth sitting with: extra airflow follows the path of least resistance, and the leakage path has almost no resistance compared with going through a server. So the extra air mostly goes round, the return temperature drops further, the unit rise narrows further, and the inlets barely move. The room now costs more to run and reads worse on every metric except the one on the display.
The 9 kW of fan power in the default is the direct cost of that surplus. Moving only what the servers pull would be 2.73 kW on the cube rule, and the 6.27 kW difference is not just electricity — it is also 6.3 kW of heat that the same units then have to remove. The cube rule is a fan law and holds while the system curve stays put, so treat it as the size of the prize rather than a cheque.
The averages are hiding the problem
One rack inlet temperature for a room is a fiction. Inlet temperature varies over the height of a single cabinet by more than it varies across a well behaved room, because the top of a rack is nearest the ceiling return and the recirculation path over the top of the cabinet is short. The reading that matters is the worst one, and it is usually at the top of the last cabinet in the row, next to a gap where a cabinet used to be.
Run the page twice: once with the floor average, and once with the worst inlet you measured. If the recirculation share jumps from a fifth to a half between those two runs, the room does not have a cooling problem at all. It has four blanking panels missing and an aisle that is open at one end.
Questions people ask
How many CFM do I need per kW in a data centre?
It falls straight out of the temperature rise you allow across the equipment, and nothing else. At the default 22 degree rise and standard air properties it is 143.6 CFM per kW. Allow a 30 degree rise and it drops to 105; hold to 15 degrees and it climbs to 211. There is no universal figure, only the figure that follows from the rise your equipment actually produces, which is why the form asks for two measured temperatures rather than offering a rule of thumb.
What does it mean when the return air is colder than expected?
Usually that supply air is finding its way back to the unit without passing through a rack. The unit is then producing a narrower temperature difference than its rating assumes, so it is doing less work per unit of airflow, and the fans are moving air for nothing. A cold return with hot rack inlets is the signature of that, and it is the one case where the display on the cooling unit reads best exactly when the room is behaving worst.
How do I measure containment effectiveness?
With four thermometers and no model. Read the supply air at the unit discharge, the exhaust behind a rack, and the inlet at the rack face. The inlet sits somewhere between the other two, and where it sits gives the mixture: inlet minus supply, divided by exhaust minus supply, is the share of the intake that is recirculated exhaust. Subtract from one hundred percent for the share that is genuinely supply air. It needs no airflow measurement and no assumption about the units.
Does this tell me whether my cooling is adequate?
No, and it is not a near miss. It compares numbers you typed: the heat you told it about against the rated capacity you told it about. Real capacity moves with entering conditions, with the condenser or the chilled water temperature, with coil fouling and with what happens when one unit is out. Adequacy is settled by an engineered design and by loading the room and watching it, not by a subtraction.
Should the other heat in the room include UPS losses?
If the UPS sits inside the walls being cooled, yes, and that is the item most often left out. Losses in the UPS, in the PDU transformers and in the fan motors that sit in the airstream all end up as heat in the same room, and together they can be a tenth of the IT load or more. If the UPS lives in a separate battery room on its own cooling, it belongs in that room instead. The field takes your figure either way.