CRAC and CRAH airflow and static pressure field guide for data centers

What CRAC/CRAH airflow setup proves

CRAC/CRAH airflow setup is the work of getting the cooling units and the air path they feed to deliver the design airflow to every server inlet at a temperature inside the equipment's envelope, with the plenum or room holding enough pressure that cold air actually reaches the racks instead of mixing on the way. When it is done right, the proof is simple. Every rack inlet sits inside the ASHRAE band at full load, the return air comes back warm, and a single unit dropping out does not put a row over the line.

The unit is judged at the rack inlet, not at its own discharge. A CRAH can blow a perfect 65 F off its coil and still leave a rack at the end of the row pulling 85 F air, because the cold supply short-circuited back to the return before it ever climbed through a tile. Cold air supplied is not cold air delivered. The whole job is closing that gap.

Three things have to be true at once. The unit moves its rated air against the real static of the floor or the duct. The plenum or the overhead distribution holds pressure across the whole space, not just near the units. And the tiles, grilles, or diffusers put that air where the heat is. Miss any one and you get the same symptom from a different cause: a hot rack with plenty of cooling capacity sitting idle in the room.

CRAC or CRAH, downflow or upflow: which do you have?

A CRAC is a computer room air conditioner with its own direct-expansion refrigerant circuit, a compressor and a condenser, that makes its own cold. A CRAH is a computer room air handler with no compressor: a chilled water coil fed from a central plant, with a control valve modulating water flow to hold supply temperature. From the airflow side the difference is the capacity-control behavior. A DX unit stages or modulates compressors, so its sensible cooling steps; a chilled water unit trims a valve, so it modulates smoothly. Read the cut sheet. If there is a compressor in the cabinet it is a CRAC, if there is a water coil it is a CRAH, and the industry uses CRAC loosely for both, which fouls submittals.

Configuration matters more than the cold source for how you set up the air. Downflow units sit in the white space and blow down into a raised-floor plenum, which is the classic arrangement and the one the underfloor pressure sections below address. Upflow units draw return low and discharge up, into the room or into overhead ductwork, and show up in rooms with no raised floor and in smaller edge halls.

The cooling-and-airflow pillar covers the CRAC-versus-CRAH choice at the architecture level. This guide is the airflow commissioning of whichever you have: setting the plenum, placing the tiles, balancing to the inlet, and proving the fail-over.

What underfloor static pressure should a data center plenum hold?

A raised-floor plenum is commonly held around 0.05 in. wg, with most halls living somewhere between roughly 0.03 and 0.07 in. wg depending on tile open area, floor height, and how leaky the floor is. There is no single right number. The plenum is a pressurized supply duct, and the target is whatever holds enough lift through the perforated tiles at the far racks while not over-pressurizing the floor near the units. Confirm the figure against the project's mechanical basis of design, because it is set by the tile selection and the load, not by a rule of thumb.

Too high and the floor tells on itself. Tiles whistle, lightweight tiles lift and rock in their stringers, and the units burn fan energy pushing against a floor that is choked or sealed too tight. Too low and the far rows starve: the plenum cannot lift air through the distant tiles, so the racks at the end of the run pull their makeup air out of the hot aisle and run hot while the room average looks fine.

Manufacturers rate tile flow at 0.10 in. wg on the brochure, which almost no real floor runs at. Do not size the air distribution off the catalog pressure. Map the actual plenum pressure across the floor with a manometer, the same instrument and the same hundredths-of-an-inch resolution the external static pressure guide uses on a building air handler, because this is that same physics turned on its side. The number that matters is the pressure at the worst tile, not the average.

Where do perforated tiles go, and how much open area?

Perforated tiles and floor grilles go in the cold aisle, in front of the rack inlets, and the count and open area get matched to the heat in that row, not spread evenly across the floor. A standard perforated tile runs around 25 percent open area and passes a few hundred CFM at typical plenum pressure; high-flow grilles with dampers reach 50 to 60 percent open and pass well over a thousand CFM each. The brochure numbers assume 0.10 in. wg, so derate to your real plenum pressure when you plan the layout.

The single most common floor mistake is the evenly spread tile. It feels fair and it is wrong. A row of 5 kW racks and a row of 15 kW racks do not need the same air, and equal tiles starve the dense row while flooding the light one. Concentrate tiles in front of the loaded racks, pull them from the empty and the lightly loaded rows, and never put a perforated tile in a hot aisle. A tile in the hot aisle dumps cold supply straight into the exhaust and shows up as a low delta-T and a cooling bill for air that did no work.

Tiles too close to a downflow unit are their own trap. Right at the unit the plenum air is moving fast and horizontal, and the static pressure dips, so a tile there can pull less air than one farther out, or even draw room air down into the floor. Keep the first row of tiles back from the units and let the plenum pressure recover. As loads change, the tile layout is not a set-and-forget item. It is a balancing knob you turn as racks fill in.

Measuring airflow: flow hood, traverse, and the unit CFM

You measure airflow at three places and they have to reconcile: at each perforated tile, at the cold-aisle face of the rack, and at the unit itself. At the tile, a balometer or flow hood capped over the tile reads the CFM passing through it directly, which is the cleanest way to confirm the floor is delivering what the layout intended. Walk the floor with the hood and you get the actual per-tile distribution, not the design assumption.

At the unit, the supply or return airflow is read by a pitot traverse across the duct or the unit face, or off the fan's published curve using measured static and speed, the way you would plot a building blower. EC-fan units often report airflow from the fan controller, which is convenient and worth a spot check against a traverse, because a reported number drifts from a measured one when a filter loads or a damper moves.

The discipline that separates a real balance from a paper one: balance airflow to the load, and balance to the inlet, not the return. The sum of tile airflow in a contained cold aisle should track the racks' airflow demand, roughly 160 CFM per kW at a 20 F rise, which the cooling pillar derives. Set the tiles so the cold aisle holds slightly positive against the racks, then confirm with inlet temperatures. A floor balanced to make the return air look right while a corner rack cooks is a floor that was balanced to the wrong number.

Delta-T across the unit and the room

Delta-T is the temperature rise between the air a unit supplies and the air that returns to it, and it is the number that tells you whether the air is doing work. A healthy air-cooled hall returns warm air at a delta-T near the design rise, commonly around 20 F, which means the supply climbed through the racks, picked up its full load of heat, and came back. A low delta-T, return air barely warmer than supply, is the signature of mixing, and it is the quiet capacity killer in a data center.

Low-delta-T syndrome comes from two leaks of the same kind. Bypass is cold supply that reaches the return without passing through a server, through an unsealed cable cutout, a missing blanking panel, or a tile in the wrong aisle. Recirculation is hot exhaust looping back over the top or around the end of a row into the cold aisle, raising the inlet temperature even though the room average looks fine. Both pull the return temperature down, and the units answer by moving more air to reject the same heat. The plant runs out of airflow long before it runs out of cooling.

The fix is almost never another unit. It is sealing the raised-floor cutouts with brush grommets, fitting blanking panels in every empty rack slot, closing the gaps at the ends of containment, and moving tile airflow to the loaded rows. The cooling pillar walks the mechanism in full, and the raised-floor cutout sealing is its own acceptance scope, because a cutout you can seal with a grommet during fit-out is a permanent loss once a rack is parked in front of it. Chase a low delta-T as a leak, not as a cooling shortage.

EC and VFD fans, and the affinity laws that govern them

Modern CRAC and CRAH units run EC (electronically commutated) plug fans or VFD-driven fans instead of fixed-speed belt-drive blowers, and the reason is the cube. Fan power follows the cube of speed, so trimming fan speed to match the real load instead of running flat out is the largest single fan-energy lever in the hall. The affinity laws are worth carrying in your head, because they tell you what a speed change costs and buys before you touch the controller.

Airflow rises and falls in direct proportion to fan speed. Static pressure rises with the square of speed. Power rises with the cube. Drop a fan to 80 percent speed and you get 80 percent of the airflow, about 64 percent of the static, and roughly half the power. That is why a hall full of EC fans running at 70 to 80 percent on a well-contained floor uses a fraction of the energy of the same fans pinned at 100 percent fighting a leaky room.

What you control the fans on is the next section's question, but the setup point is this: with EC fans, a contained hall, and the fans trimmed to hold a plenum-pressure or inlet-temperature target, the airflow tracks the load automatically and the energy follows the cube down. The catch is that the fan curve still rules. Trim too far and the plenum pressure collapses at the far tiles, and the cube that saved you energy starves a row. Set the minimum speed to hold the worst tile, not the average.

EC fan

Electronically commutated plug fan; a DC-driven, variable-speed fan that trims to a control signal

VFD

Variable frequency drive, which varies an AC motor's speed by changing supply frequency

Affinity laws

Airflow scales with speed, static with speed squared, power with speed cubed

Return-air, supply-air, or cold-aisle control: which wins?

Three control strategies set the unit's capacity and fan speed, and they are not equal. Return-air control trims the unit to hold a target temperature at its own return, which is the old default and the weakest, because the return temperature is an average of everything that happened in the room and says nothing about the rack that is starving. Supply-air control holds a target off the coil, which is better because it fixes what the unit actually delivers. Cold-aisle or rack-inlet control holds a target at the server inlets, which is what you are actually trying to protect.

The modern approach favors supply-air or cold-aisle control, paired with fan speed modulated on plenum pressure or on the worst inlet temperature. The reason is the same reason cooling is judged at the inlet: return-air control will happily over-cool the whole hall to satisfy an average while a corner rack sits over the line, and it cannot tell bypass air from real load. A floor with containment and inlet-based control raises its supply temperature toward the top of the ASHRAE recommended band and lets the fans trim down, which is where the energy savings live.

Fighting controls are the failure mode to watch when units share a hall. Several units on return-air control, each reacting to a slightly different average, will hunt against each other, one ramping while its neighbor backs off. Coordinate the units on a common reference, group control or a shared inlet sensor network, so they share the load instead of chasing each other. Confirm the sequence of operations is what got installed, because the controls intent on the drawings and the controls behavior in the field part ways more often than not.

Humidity at the unit and the fighting-units problem

Humidity control at a CRAC or CRAH is a smaller job than it used to be, because the ASHRAE allowable envelopes widened and the industry moved to controlling dew point over a wide band instead of pinning relative humidity. Set up the units to a common dew-point target and let relative humidity float, and most of the old grief disappears. The grief that remains is the fighting-units problem, and it is almost always a controls and coordination failure, not a hardware one.

Here is how it happens. Each unit reads humidity at its own return and acts independently. A cold coil pulls moisture out of the air passing through it, so a unit running hard on sensible load is also dehumidifying whether you asked it to or not. Its return reads dry, so it calls for humidification. The unit across the hall, reading a slightly different return, calls for dehumidification at the same time. Now one unit is boiling water into the air while another is condensing it out, both burning energy, and the net humidity barely moves. A single hall can waste a startling amount of energy this way and never show a fault.

The fix is coordination and a wide band. Put the units on a shared humidity reference or a group controller, widen the dead band so small swings do not trigger action, and control to dew point. During commissioning, force a humidity excursion and watch the whole fleet respond, because the only way to catch two units fighting is to look at all of them at once. If one is adding moisture while another removes it at steady state, the setup is wrong.

Filter and coil pressure drop at the unit

Every CRAC and CRAH fan fights internal static before it ever sees the floor: the filter bank and the coil are the two biggest drops inside the cabinet, and they climb over time. A clean filter and a dry coil might cost a few tenths of an inch of water column together; a loaded filter and a wet, dirty coil cost far more, and on a fixed-speed unit that lost static comes straight off the airflow. On an EC-fan unit the fan speeds up to hold its airflow target, so the loss shows up as watts and noise instead, until the fan runs out of headroom and the airflow falls anyway.

This is the same static-pressure budget the external static pressure guide lays out for a building air handler, and the physics does not change because the box says CRAH. The rated airflow assumes a clean filter and a clean coil at a stated internal static. Let either foul and the real delivered airflow drifts below the rating while the controller still believes it is hitting target.

Set up the unit with the filters specified, not whatever fit, and trend the filter differential pressure if the unit reports it. The field tell is a leaving-air temperature that creeps up while the valve or compressor sits wide open: the coil is fouled or starved, the air is moving too slow over it, and no amount of control authority recovers what a cleaning would. Pull the panel and look before you chase it in the controls.

What happens to airflow when a CRAC unit fails?

When a unit fails, the survivors have to pick up its airflow without leaving any rack starved, and proving that is the whole point of N+1. N is the airflow the load needs with nothing to spare. N+1 adds one more unit so any single one can drop out, for a fault or for service, and the row stays inside the envelope. The number that catches people is not the cooling capacity. It is the airflow and the plenum pressure, because a failed downflow unit is a hole in the floor pressure as much as a loss of tons.

The local failure is the one that bites. A unit that fails takes its slice of plenum pressure with it, and the racks nearest that unit lose lift through their tiles first, even if the hall has plenty of total redundant capacity sitting at the far end. Redundancy that is real on a capacity spreadsheet can fail locally if the surviving units cannot push pressure into the zone the dead unit was feeding. This is why fail-over has to be tested, not assumed, and tested with the floor loaded.

A failed unit also has to stop being a leak. When a downflow unit's fans stop, the plenum can back-drive air up through the dead unit, dumping pressure out the top of the cabinet. Backdraft dampers on the units exist for exactly this, and they are a commissioning item: prove they close when the unit stops, because a stuck-open damper on a failed unit bleeds the plenum and pulls the neighbors down with it.

The commissioning sequence and functional tests

Airflow commissioning runs from static checks up to the loaded fail-over, and nothing functional starts until the units are verified installed and ready: fans turning the right way, filters and coils clean, condensate and chilled water connected and proven, controls points checked. A downflow fan running backward moves air and looks alive while delivering almost nothing into the plenum, so confirm rotation and discharge before you trust a single reading.

Then the air-side sequence has four field tests that actually prove the setup. First, airflow verification: traverse or plot each unit to its rated CFM, and flow-hood the tiles to confirm the floor delivers the layout. Second, the plenum pressure map: read the static across the whole floor, corner to corner, and confirm the worst tile holds enough lift. Third, the fail-over test: drop each unit in turn with the floor under load and watch the inlet temperatures and the plenum pressure hold while the survivors ramp. Fourth, the rack-inlet temperature map: read inlet temperatures across the hall, top and bottom of the racks, and confirm every one sits inside the ASHRAE envelope at full load.

These are test-and-balance and functional-performance tests, performed to the AABC or NEBB procedures by the TAB contractor and witnessed by the commissioning agent. The inlet map is the acceptance artifact the whole job is built around, because it is the only test that proves the air reached the load. The cooling pillar covers the full commissioning level structure up through the integrated systems test; the four tests here are the air-side core of it.

How containment changes the airflow setup

Containment, the doors and roof that seal the cold aisle from the hot aisle, changes the airflow setup from a guessing game into a controllable balance, and it is the single biggest air-side efficiency move on the floor. With the two air streams sealed apart, the supply air has nowhere to go but through the racks, so the plenum holds a steadier pressure, the delta-T comes up, and you can raise the supply temperature toward the top of the ASHRAE recommended band and let the EC fans trim down.

The setup interaction worth knowing: containment makes plenum pressure control behave. Without it, opening a tile in one aisle changes the pressure everywhere as air finds the path of least resistance, and balancing is whack-a-mole. With a sealed cold aisle, the tiles in that aisle feed that aisle, and fan-on-pressure control has a stable target to hold. The cold aisle held slightly positive against the racks means no hot air recirculates in; a contained hot aisle held slightly negative means no cold bypass leaks through.

Containment also raises a safety and controls item that lives in the commissioning scope: a sealed aisle with the cooling running becomes a sealed box, and the fire and cooling controls have to coordinate so a suppression event or a cooling loss does not trap heat. The containment work is its own topic, but on the airflow side the rule is simple. Seal the aisle, control the fans on pressure or inlet temperature, and the floor balances instead of fighting itself.

Field example: a starved far row on a balanced-looking floor

A new white space passed its unit-level airflow checks, every CRAH plotting close to its rated CFM, but the rack-inlet map flagged the last two racks of the most distant row pulling inlet air near 30 C against a 24 C target while the room average read a comfortable 22 C. The reflex call was to add a unit. The plenum pressure map said otherwise.

Static under the floor read about 0.06 in. wg near the units and fell to roughly 0.02 in. wg at the far row, well short of the lift those tiles needed. Two causes stacked up. A run of abandoned cable tray was choking the plenum cross-section ahead of the far row, and the tiles had been laid out evenly, so the light racks near the units were flooded with air the far row never got. Nothing was undersized. The pressure simply was not reaching the end of the floor.

Clearing the abandoned tray opened the plenum, and pulling tiles from the lightly loaded near rows pushed the far-row static up toward 0.045 in. wg. The far-row inlets came down inside the envelope over the next shift, the delta-T at the units rose, and no equipment was added. The capacity had been there the whole time, stranded by a blocked plenum and a tile layout balanced to the room average instead of to the inlet.

What to document

A floor that was balanced and never recorded leaves operations with no baseline, so the first warm aisle six months out becomes a fresh investigation instead of a check against how the hall has always run. The record is the as-balanced state of the air system, and it is the artifact the next engineer trends every airflow problem against.

Capture, per unit and per zone: the unit make and configuration, the measured and rated CFM, the supply and return temperatures and the delta-T, the plenum differential pressure across the floor, the perforated tile layout and open area by row, and the rack-inlet temperatures top and bottom across the hall. Record the fail-over test results, which unit was dropped and what the survivors held, because that is the proof the redundancy is real. If you moved tiles or cleared a plenum, write the before and after, because a change you cannot show is a change the next person will not trust.

Common mistakes

• Setting plenum pressure too high, so tiles whistle and lift and the fans waste energy.
• Setting plenum pressure too low, so the far rows starve while the room average looks fine.
• Spreading perforated tiles evenly instead of matching tile airflow to the rack load.
• Putting a perforated tile in the hot aisle, dumping cold supply straight into the exhaust.
• Controlling the units on return air, so the fleet over-cools to an average and misses the hot rack.
• Running units on independent humidity control, so one humidifies while another dehumidifies.
• Balancing airflow to make the return look right instead of balancing to the rack inlet.
• Skipping the loaded fail-over test, so a dropped unit starves a row nobody proved was covered.
• Trusting the catalog 0.10 in. wg tile rating instead of mapping the real plenum pressure.
• Leaving cable cutouts unsealed and empty rack slots without blanking panels, feeding low delta-T.

Field checklist

Standards and references

The thermal target the airflow has to satisfy comes from ASHRAE Technical Committee 9.9 and its Thermal Guidelines for Data Processing Environments, which set the recommended and allowable inlet envelopes and the A1 through A4 equipment classes. The energy side is ASHRAE Standard 90.4, the energy standard for data centers, which bounds the mechanical overhead by climate zone through a mechanical load component rather than regulating a single airflow number. Edition numbers and the envelope values move between cycles, so confirm the current edition and the actual numbers against the published guideline and the IT equipment manufacturer.

The air balance itself is the province of the test-and-balance bodies, AABC and NEBB, who certify the procedures and technicians that verify a system delivers its design airflow, and their procedures are what a witnessed balance follows. Duct and plenum construction and leakage classes come from SMACNA. Fan performance ratings, the curves you plot a unit against, trace back to AMCA test methods. Where a facility is chasing a tier, the Uptime Institute Tier standards drive the witnessed redundancy and the loaded fail-over demonstrations.

Underneath all of it sit the unit manufacturer's data and the project specification. The plenum pressure target, the tile selection, the supply temperature, and the redundancy level are project decisions captured in the basis of design, and the manufacturer's fan and coil data set what a given unit can actually deliver. Cite the body that owns the point, hedge to the softer claim when a value is project-dependent, and let the spec override any rule of thumb when it is stricter.

Units, terms, and conversions

Data center airflow carries vocabulary from HVAC, from test and balance, and from the IT side, and the same idea reads differently across a TAB report, a unit submittal, and a controls sequence. The terms below travel across the airflow commissioning scope.

Static pressure is written in. wg (inches water gauge), in. w.c., or in. H2O, all the same thing; 1 in. wg is about 249 pascals, so a 0.05 in. wg plenum is roughly 12 Pa. Airflow is CFM in the field and liters per second or cubic meters per hour in metric sources. The temperature rise across the racks is delta-T, in degrees F or C. Open area is the fraction of a tile that actually passes air, the number that sets its flow at a given pressure.

CRAC

Computer room air conditioner; a direct-expansion unit with its own compressor and refrigerant circuit

CRAH

Computer room air handler; a chilled water coil unit with no compressor, fed from a central plant

in. wg / in. w.c.

Inches of water column, the field unit for static pressure; 1 in. wg is about 249 Pa

Plenum pressure

The static under the raised floor that lifts air through the tiles; commonly near 0.05 in. wg

CFM

Cubic feet per minute, the airflow that carries the heat; near 160 CFM per kW at a 20 F rise

Delta-T

The temperature rise between supply and return air across the racks; the air-side efficiency number

Tile open area

The fraction of a perforated tile that passes air; commonly 25 percent, up to 50 to 60 percent for grilles

EC fan

Electronically commutated plug fan; variable-speed, trims to a pressure or temperature signal

Affinity laws

Airflow scales with fan speed, static with speed squared, power with speed cubed

Related tools

Calculators and readiness checks for this work

Compare your options

FAQ

People also ask

• What is bypass airflow in a data center?
• How do I measure total external static pressure?
• What is delta-T in a data center?
• What is a good temperature split for AC?
• What is an as-built drawing?