Datacenter
Hot-aisle and cold-aisle containment field guide for commissioning
Where the cold air actually goes: leakage paths, blanking panels, the differential pressure that proves the seal, the inlet map, and the fire code you cannot skip.
Direct answer
Hot-aisle/cold-aisle containment is a physical barrier of doors, roof panels, or curtains that stops cold supply air from mixing with hot exhaust. Separating the two air streams is the biggest air-side efficiency move in a data center, because it lets you raise the supply temperature and cut fan energy. The IT equipment class and the AHJ control the limits.
Key takeaways
- Hot-aisle/cold-aisle containment is a physical barrier of doors, roofs, or curtains that stops cold supply air from mixing with hot exhaust.
- Differential pressure is the primary containment proof: a slightly positive cold aisle, read with a sensitive manometer in hundredths of an inch water column.
- Fit a blanking panel in every empty rack slot; open U-spaces let hot exhaust recirculate to the inlet above and are the most common leak.
- Cold aisles should hold the server inlet inside the ASHRAE TC 9.9 recommended band, commonly 18 to 27 C (64.4 to 80.6 F), with equipment class controlling the limit.
- NFPA 75 and the adopted NFPA 13 edition govern fire protection; suppression must cover the contained space, so get AHJ signoff in writing before install.
Aisle containment, and why it is the biggest air-side move
Hot-aisle/cold-aisle containment is a physical barrier, doors at the ends of the aisle and a roof or ducted ceiling over it, that keeps cold supply air and hot exhaust from mixing. Racks face each other so cold air enters one aisle and hot exhaust dumps into the next. Containment closes the gaps so the cold supply cannot short-circuit into the hot return before it has passed through a server.
The reason it is the biggest air-side move you can make has nothing to do with adding cooling. It is about not wasting the cooling you already have. When cold and hot air mix, two things go wrong at once. Cold supply that reaches the return without passing through a server cost energy and did no work, and hot exhaust that loops back into the cold aisle raises the server inlet temperature and makes the hot spots. Seal the aisle and both stop.
The payoff is money. With the mixing gone, the cold aisle stays cold end to end, so you can raise the supply temperature toward the top of the ASHRAE recommended band and the chiller plant does less work. The number that matters is still the temperature at the server inlet, not the room average. Containment is what makes the inlet match the supply.
What is the difference between hot-aisle and cold-aisle containment?
Both keep the two air streams apart. They differ in which aisle you wrap. Cold-aisle containment (CAC) encloses the cold supply aisle, so the contained space is cold and the rest of the room becomes a warm return plenum. Hot-aisle containment (HAC) encloses the hot exhaust aisle and ducts it back to the return, so the open room is held at the cool supply condition.
The tradeoff is where you want the room to sit. With CAC the open room runs hot, which is fine for the IT gear but rough on anyone working outside the contained aisle and on any non-contained equipment sitting in the room. With HAC the room stays comfortable and only the sealed hot aisle gets hot, which protects standalone gear and the people, but the hot aisle itself can run well past 100 F and that drives the fire and detection requirements inside it.
For a retrofit, CAC is usually the easier add, because you cap the existing cold aisle over a raised floor without re-ducting the return. For a new build, HAC paired with a ducted ceiling return is often the cleaner design and holds the room at a single benign temperature. Neither is wrong. The room you want to live in and the return path you already have usually decide it.
The leakage paths that defeat containment
Containment is only as good as its leaks, and the leaks are physical and findable. Walk the aisle and they show up in the same places every time.
Missing blanking panels in the rack are the most common. An open rack-unit slot lets hot exhaust pull straight through to the inlet side. Open U-spaces above and below mounted gear do the same. Gaps under and beside the racks let air sneak around the row instead of through it. Unsealed floor cutouts and cable openings, the brush grommets that were never fitted, dump cold supply straight into the plenum return or let hot air up into the cold aisle, which is why the raised-floor acceptance work seals them before the racks land.
Then there is the containment shell itself. Gaps at the ends of the row where there is no door, gaps between the rack tops and the roof panels, and the seal to the ceiling or return plenum that was never finished. The ends of rows are the worst offenders, because they are the largest single opening, and a row left open at the end leaks more than every missing blanking panel combined. Find the big openings first, then chase the small ones.
Why do I need blanking panels?
A blanking panel is a flat filler that closes an empty rack-unit slot so air cannot pass through the front of the rack except through equipment. Without it, an open U is a direct hole between the hot side and the cold side of the rack. Hot exhaust at the back finds that hole and pulls forward into the cold aisle, right at the inlet of the gear mounted above it.
That is the mechanism behind the recirculation hot spot. The fan in the server above the open slot pulls the easiest air it can get, and the easiest air is the hot exhaust coming through the gap below it, not the cold supply out in the aisle. So the device runs hot while the aisle reads cold. People chase it with colder supply when the fix is a panel that costs a few dollars.
Blanking panels are the cheapest move in the whole hall and the one most often skipped. Fit every empty slot, top to bottom, and seal the gaps at the sides of the mounting rails too, because air goes around the equipment as readily as through an open U. The audit is simple. Stand in the cold aisle and look for daylight or feel for warm air coming forward. Every gap you find is a panel you owe.
The containment structure: doors, roofs, and seals
The shell is doors at the ends of the aisle, a roof or ceiling over it, and the seals that tie it all to the racks. Each piece has a failure mode worth knowing.
End-of-aisle doors close the largest opening. They can be hinged or sliding, and they have to actually close and stay closed, because a door propped open during a maintenance shift undoes the containment for the whole row. Sliding doors that bind get left open. Self-closing hinged doors get props wedged under them. Watch for both during the walk.
The roof is either solid panels or curtains. Solid drop-out or drop-away panels seal tighter and are common on hot-aisle systems and over higher-density rows. Flexible strip curtains and clear vinyl panels are cheaper and easier to retrofit but seal less completely and sag over time. The seal to the top of the racks matters as much as the panels. A row of racks with mismatched heights leaves a gap along the top that the roof never closes unless someone fills it. And on a ducted-return HAC, the seal to the ceiling or the return plenum is where the hot air actually leaves. If that connection leaks, the hot aisle pressurizes and pushes exhaust back into the room. Check the panels, then check every seam where one material meets another.
How do you measure containment performance?
Three measurements tell you whether containment works: the differential pressure across the barrier, the rack-inlet temperature uniformity top to bottom, and the indicators of bypass and recirculation. None of them is the room thermostat.
Differential pressure (dP) is the primary one. You read the pressure of the cold aisle against the hot aisle, or the contained aisle against the room, with a sensitive manometer, because the numbers are small, often a few hundredths of an inch of water column. A cold aisle held slightly positive means no hot air is leaking in. A hot aisle held slightly negative means no cold air is bypassing through. A dP that swings or sits at zero means the seal is leaking or the fans are not matched to the load.
The inlet map is the second. You measure the inlet temperature at the bottom, middle, and top of representative racks across the row, not one reading at one rack. A working contained aisle reads nearly uniform top to bottom. A spread of several degrees from bottom to top is the signature of recirculation over the rack or through open slots. The third set is the tells. A low return delta-T at the cooling units says cold air is bypassing the racks, and warm air felt at the top of a rack in a cold aisle says hot air is getting back in. Measure all three. One alone can lie.
Fan control and the containment dP setpoint
Containment changes how the fans should run, and the control scheme is a commissioning item people forget is theirs. In a contained aisle the supply fans do not chase a room temperature anymore. They hold a pressure. The control loop modulates fan speed or tile dampers to keep the aisle dP at a target, commonly a small positive cold-aisle pressure, so the supply just matches what the racks draw.
Get the setpoint wrong and you trade one waste for another. Hold too much positive pressure and you push cold air out through every gap and balloon the aisle, wasting fan energy. Hold too little and the racks at the far end or the top can starve and pull hot air back in under load. The sensor location matters too. A dP sensor dropped in a dead spot reads a pressure the racks never see.
This is why the dP sensor, its placement, and its setpoint are acceptance items, not just design intent. Verify the control actually modulates the fans to hold the target as load changes, not just that the number looks right at one moment. A containment that holds dP at 20 percent load and starves at 80 percent passed the wrong test.
The top-of-rack hot spot and the vertical gradient
The top RU runs hottest in an uncontained aisle, and the reason is geometry. Hot exhaust rises. In an open hot-aisle/cold-aisle layout with no roof, the hot air coming off the back of the racks rises and rolls over the top of the row into the cold aisle, landing on the inlets of the gear mounted highest. So the bottom of the rack gets clean cold supply and the top gets supply diluted with its own recirculated exhaust.
That is the vertical gradient: cool at the floor, warm at the top, sometimes 10 F or more of spread across the height of one rack. It is why the top third of a rack throttles or alarms while the bottom is fine, and why the device that fails is so often the one mounted highest.
Containment fixes it by closing the path over the top. With a sealed roof and blanking panels, the hot exhaust has nowhere to roll except into the contained hot side, so the cold aisle stays the same temperature from the floor to the top of the rack. When you map inlets top to bottom and the gradient is gone, the containment is doing its job. When the gradient is still there with a roof installed, you have a leak at the rack tops or missing blanking panels, not a cooling problem.
Does aisle containment affect the sprinklers?
Yes, and it is the part most likely to fail an inspection or get value-engineered out and bite later. Containment puts a roof and walls between the ceiling sprinklers or gaseous suppression and the equipment they protect. A roof panel over the aisle can block water or agent from reaching a fire inside the contained space. The fire marshal will look at this before they sign off, so bring them in early.
NFPA 75, the standard for the fire protection of information technology equipment, and the locally adopted edition of NFPA 13 for sprinkler installation are the references that govern. The general principle in recent editions is that the suppression has to cover the contained space at all times. In practice that means one of a few approaches: sprinkler or detection heads inside the containment, smoke detection inside the aisle, or roof panels that open or drop away automatically on a fire signal so the ceiling system can reach in. A hot aisle is often treated as its own compartment, which can pull in in-aisle heads and detection, and any detection or suppression inside a hot aisle has to be rated for the high temperature in there.
The older approach of fusible-link panels that melt and drop away has fallen out of favor, because the fire has to grow large to trip the link, which is exactly the wrong time. Many designs now use active removal tied to the detection system instead. The exact provisions, section numbers, and what your jurisdiction accepts change between editions and between AHJs, so do not design the fire interaction from a rule of thumb. Confirm the adopted editions of NFPA 75 and NFPA 13 and get the AHJ signoff in writing before the containment goes in.
What temperature should the cold aisle hold?
The cold aisle should hold the server inlet inside the ASHRAE TC 9.9 recommended envelope, commonly given as 18 to 27 C (64.4 to 80.6 F) dry-bulb, but the IT equipment's rated class and the current edition control the real limit. The cooling-and-airflow pillar covers the envelope and the A1 through A4 allowable classes in full. The point for containment is narrower.
Containment is what lets you run at the top of that band safely. In an uncontained hall, operators drop the supply temperature to satisfy the worst rack, which over-cools everything else and burns chiller capacity for no reliability gain. Seal the aisle and the worst rack reads close to the best, so you can raise the supply toward 24 or 25 C and let the economizer carry more hours of the year.
The trap is reading one ASHRAE number and ignoring the gear. A row of older storage or a vendor with a tight warranty can be the binding constraint, not the guideline. Set the cold-aisle target to the actual equipment class in that row and the edition the design referenced, then prove the inlet holds it top to bottom. The envelope is judged at the inlet, which is the whole reason containment exists.
Matching supply airflow to the rack draw
Inside a contained aisle the supply airflow has to match, or slightly exceed, the total air the racks draw, or the containment turns against you. The racks are fans. They pull a fixed volume of air through themselves regardless of what the floor delivers. If the contained supply is less than that draw, the racks make up the difference by pulling air from wherever they can, which means hot exhaust back through any gap, and the aisle goes negative and starves.
Push too much and you pressurize the aisle, force cold air out through every seam, and waste fan energy moving air that never does work. The target is a slight surplus, enough to hold the cold aisle a touch positive without ballooning it.
The way you set it is tile and damper placement matched to the load in each rack, not spread evenly down the row. A 15 kW rack needs roughly 160 CFM per kW at a 20 F rise, so it wants far more tile area in front of it than a 3 kW rack two positions down. Even tile spacing is the lazy default that starves the dense racks and floods the empty ones. Match the air to the heat, rack by rack, then confirm with the inlet map.
Retrofit gotchas: partial rows, short rows, and mixed depths
Retrofits rarely give you the clean full row the containment kit was drawn for, and the gaps are where the leaks live. The half-full row is the classic. You contain a row that is only half populated, and every empty rack position and every open slot is a leak unless it gets a filler panel or a blanking rack. A contained row with holes in it can perform worse than an open row, because you have built a plenum and then punched it full of holes.
Short rows and partial rows leave the ends and the transitions open. A containment system sized for a 10-rack row dropped onto a 6-rack row leaves a gap at the end that nobody ordered a panel for. Mixed rack depths and heights break the seal along the top and the sides, so the roof and the end doors do not land flush and the gaps run the length of the row.
The fixes are filler panels, end-of-row panels for the open positions, and brush or foam at the height and depth transitions. None of it is exotic, but it has to be on the order before the containment goes up, not improvised after. Walk the row against the rack elevation drawing first, count the gaps, and order the fillers. The retrofit that fails acceptance is almost always the one where nobody counted the holes.
What if the contained aisle still runs hot?
A contained aisle that still runs hot is telling you the seal leaks, the air is short, or the heat moved. Work it in that order, because the cheapest causes are the most common.
Leaks first. Walk the row for missing blanking panels, open floor cutouts, gaps at the rack tops and ends, and a door left propped. Most still-hot aisles are a containment with holes in it, not a cooling shortfall. The inlet map points you at the leak. A hot spot at the top of one rack is a missing panel or a top-of-rack gap right there.
Air second. If the seal is tight and the aisle still runs hot, the supply is not matching the draw. Check the dP. A cold aisle sitting negative means the racks are pulling more than the floor delivers, so add tile area in front of the loaded racks or raise the fan output. A low return delta-T at the units confirms bypass is stealing the air.
Heat last. If the seal is tight and the air balances and a rack still runs hot, the rack itself may have outgrown air cooling. A rack that climbed past 30 or 40 kW during a refresh is a density problem, and the answer is rear-door cooling or direct-to-chip, not more cold air. Do not keep dropping the whole hall's supply temperature to chase one dense rack. That is the move containment was supposed to end.
Commissioning and the containment acceptance test
The acceptance test for containment is four things done in order: the blanking-panel and seal audit, the leakage walk, the differential-pressure verification, and the inlet temperature map under load. Skip any one and you have accepted a number, not a sealed aisle.
Start with the static audit. Every empty rack slot has a blanking panel, every floor cutout is grommeted, every end has a door, and the roof seals to the rack tops. This is a walk with a drawing and a checklist, before anything is energized. The leakage walk follows. With the supply running, you feel the seams, the rack tops, the ends, and the floor for air going where it should not, and you find the daylight gaps by eye.
Then the numbers. Verify the cold-aisle-to-hot-aisle dP holds the target, and confirm the fan control actually modulates to hold it as load changes, not just at one setpoint. Map the inlet temperatures top, middle, and bottom across representative racks and confirm the vertical gradient is gone and every inlet sits inside the envelope. Run it under real or simulated load, because a containment that holds at 20 percent load and fails at 80 is a containment that failed. Record all of it against the design targets. That record is the acceptance.
Field example: a contained row that still alarmed at the top
A newly contained cold-aisle row in a colo hall kept throwing high-inlet alarms at the top of two racks even though the roof and end doors were installed and the supply temperature was already down to 20 C. The reflex was to drop the supply further. The inlet map said do not.
Bottom-of-rack inlets read about 20 C as supplied. The middle read 23 C. The top read 31 C, well over the recommended band and climbing under load. That vertical spread, clean at the floor and hot at the top, is the signature of recirculation over the rack tops, not a cooling shortfall. A walk found it. The two hot racks were shorter than their neighbors, so the roof panels left a gap along the top of those positions, and several empty slots in both had no blanking panels.
Filler panels along the top transition and blanking panels in the open slots closed the path. Over the next shift the top-of-rack inlet fell from 31 C toward 22 C and the gradient flattened to about 2 C top to bottom. The supply temperature was then raised back to 23 C with the aisle still inside the envelope, recovering the chiller energy the over-cooling had been spending. No cooling was added. The aisle had been contained on paper and leaking at the top.
| Measurement | As found (alarming) | After sealing the top |
|---|---|---|
| Top-of-rack inlet | about 31 C | about 22 C |
| Bottom-of-rack inlet | about 20 C | about 20 C |
| Vertical gradient top to bottom | about 11 C | about 2 C |
| Supply temperature | 20 C, dropped to chase it | 23 C, raised back |
| Cooling added | none | none |
What to document
A containment that was sealed but never documented leaves operations with no baseline. The record is what tells the next engineer whether a warm rack is new or how the aisle has always run, and what the containment was accepted at. Capture it per aisle, not as a single room number.
For each contained aisle, record the cold-aisle-to-hot-aisle differential pressure and the setpoint, the inlet temperatures at top, middle, and bottom of the representative racks, the result of the blanking-panel audit, the gaps found and sealed during the leakage walk, the load at which it was tested, and the open punch items. Tie the fire-protection approach and the AHJ signoff to the package too, because that is the record that proves the containment was accepted with its suppression, not in spite of it.
| Field to record per aisle | Why it matters |
|---|---|
| Aisle ID and type (HAC or CAC) | Identifies the contained space and how it is sealed |
| Cold-to-hot dP and setpoint | Proves the seal holds and the fans match the load |
| Inlet temps top / mid / bottom | Proves the gradient is gone and the inlet is in envelope |
| Blanking-panel audit result | Empty slots are the most common recirculation path |
| Gaps found and sealed | The leakage walk record, where the air was escaping |
| Test load and date | A pass at low load is not a pass at full load |
| Fire-protection approach and AHJ signoff | Proves suppression covers the contained space |
| Open punch items | What is left before the aisle is truly accepted |
Field checklist
Want this checklist to run itself on every job — with photo proof and a signed record crews can hand the customer? That's FieldOS.
Common mistakes
- Leaving empty rack slots without blanking panels, so hot exhaust recirculates to the inlet above them.
- Containing a half-full row and punching the new plenum full of open rack positions.
- Leaving floor cutouts and cable openings unsealed, so cold supply bypasses the racks into the return.
- Running the row with no end-of-aisle doors, or propping the doors open during maintenance.
- Spreading perforated tiles evenly instead of matching tile airflow to the load in each rack.
- Ignoring the fire code and blocking the sprinklers or suppression from reaching the contained space.
- Sizing the supply below the rack draw, so the contained aisle goes negative and pulls hot air back in.
- Judging the aisle by the room average instead of mapping the inlet top to bottom under load.
- Dropping the whole hall's supply temperature to chase one hot rack the containment should have fixed.
- Accepting the containment at low load and never testing it at the load it will actually carry.
Standards and references
The temperature side of containment lives in 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 aisle is held to. The energy side is ASHRAE Standard 90.4 for data centers. The cooling-and-airflow pillar carries those in full.
The fire interaction is governed by NFPA 75, the standard for the fire protection of information technology equipment, and the locally adopted edition of NFPA 13 for sprinkler installation. The principle in recent editions is that the suppression has to cover the contained space at all times, which drives in-aisle heads, in-aisle detection, or automatic panel removal, with any device inside a hot aisle rated for the temperature there. The fusible-link drop-away approach has fallen out of favor. The specific provisions and section numbers move between editions, so confirm the adopted editions and the AHJ acceptance rather than designing from a rule of thumb.
Where a facility is chasing a tier, the Uptime Institute Tier standards drive the witnessed demonstrations, and TIA-942 is the broader telecommunications infrastructure standard for data centers, including environmental and redundancy provisions. The test-and-balance work behind the airflow is performed to the AABC or NEBB procedures. Confirm every edition and value against the published document and the IT equipment manufacturer before citing it on a submittal.
Units, terms, and acronyms
Aisle containment borrows vocabulary from HVAC, from the IT side, and from fire protection, so the same idea reads differently across a TAB report, a rack elevation, and a fire submittal. The terms below travel across the whole containment scope.
- HAC
- Hot-aisle containment, which encloses the hot exhaust aisle and ducts it to the return, keeping the room cool
- CAC
- Cold-aisle containment, which encloses the cold supply aisle and lets the surrounding room run warm
- dP
- Differential pressure across the containment barrier, the primary number that proves the seal holds
- Blanking panel
- A filler that closes an empty rack-unit slot so air passes only through equipment, not around it
- Bypass
- Cold supply air that reaches the return without passing through a server, energy spent for no work
- Recirculation
- Hot exhaust that loops back into the cold aisle and raises the server inlet temperature
- RU / U
- Rack unit, 1.75 in of vertical mounting height; an open U is a recirculation path without a blanking panel
- Delta-T
- The temperature rise across the rack; a high, stable delta-T means the air carried its full heat load
- CFM
- Cubic feet per minute of airflow; roughly 160 CFM per kW at a 20 F rise on the air side
- AHJ
- Authority having jurisdiction, the official who interprets and enforces the fire and building code locally
FAQ
What is the difference between hot-aisle and cold-aisle containment?
Cold-aisle containment encloses the cold supply aisle and lets the room run warm; hot-aisle containment encloses the hot exhaust and ducts it to the return, keeping the room cool. Cold-aisle is the easier retrofit over a raised floor; hot-aisle suits new builds and protects standalone gear and people in the room.
Why do I need blanking panels in a server rack?
Blanking panels close empty rack-unit slots so hot exhaust cannot pull forward through the rack into the cold aisle. An open slot lets the server above it draw its own hot exhaust instead of cold supply, creating a recirculation hot spot. They are the cheapest air-management fix and the one most often skipped.
Does aisle containment affect the sprinklers?
Yes. A containment roof can block ceiling sprinklers or gaseous suppression from reaching a fire inside the aisle. NFPA 75 and the adopted NFPA 13 edition require coverage of the contained space, usually through in-aisle heads, in-aisle detection, or automatic panel removal. Confirm the approach with the AHJ before installing.
What if the contained aisle still runs hot?
A contained aisle that still runs hot usually has a leak, not a cooling shortfall. Check for missing blanking panels, open floor cutouts, gaps at the rack tops, and a propped door first. Then check the dP and tile airflow. Only a rack past roughly 30 to 40 kW has truly outgrown air cooling.
How do you measure containment performance?
Measure three things: the differential pressure between the cold and hot aisle, the rack-inlet temperature at the top, middle, and bottom of representative racks, and the return delta-T at the cooling units. A held dP, a flat vertical gradient, and a high delta-T mean the containment is sealing. The room thermostat tells you nothing.
What temperature should a contained cold aisle run at?
A contained cold aisle should hold the server inlet inside the ASHRAE TC 9.9 recommended band, commonly 18 to 27 C (64.4 to 80.6 F), with containment letting you run near the top safely. The actual equipment class and the current edition control the limit, so set the target to the gear in that row.
How much airflow does a contained aisle need?
A contained aisle needs supply airflow that matches or slightly exceeds the total rack draw, roughly 160 CFM per kW at a 20 F rise. Too little and the aisle goes negative and pulls hot air back in; too much pressurizes it and wastes fan energy. Match tile airflow to each rack's load, not evenly.
Is cold-aisle or hot-aisle containment better for a retrofit?
For a retrofit, cold-aisle containment is usually easier because you cap the existing cold aisle over a raised floor without re-ducting the return. Hot-aisle containment needs a ducted ceiling return and suits new builds. Neither is wrong; the return path you already have and the room temperature you want decide it.
Do fusible-link drop-away roof panels still meet fire code?
Fusible-link panels that melt and drop away have fallen out of favor, because the fire has to grow large before the link trips, which is the wrong time. Recent NFPA guidance leans toward active removal tied to the detection system or coverage designed into the containment. Confirm what your AHJ and the adopted editions accept.
Why does the top of the rack run hottest without containment?
Without a sealed roof, hot exhaust rises off the back of the racks and rolls over the top into the cold aisle, landing on the inlets of the highest-mounted gear. That creates a vertical gradient, cool at the floor and hot at the top, often 10 F or more. Containment closes the path and flattens it.
People also ask
Codes cited in this guide
This guide is written and reviewed against the published standards below. Always confirm the current adopted edition with the authority having jurisdiction.