Data center airflow management: bypass, recirculation, and containment

What data center airflow management is

Data center airflow management is the work of getting cold supply air to every server inlet and the hot exhaust back to the cooling units without letting the two streams mix on the way. It is the air-distribution job that sits between the cooling plant and the chip. Pick the wrong architecture and you have a system problem. Get the air wrong in the room and you have an airflow problem, and the second one is what strands capacity on most floors.

Two neighbors own the pieces this guide leaves out. Whether the hall is air or liquid cooled, and whether the units are CRAC or CRAH, is a system-selection question covered in the cooling system types guide. The unit fan curve and the underfloor plenum static pressure that lifts air through the tiles are covered in the CRAC-CRAH static pressure guide. This guide is about what happens to the air in the room: where it goes, where it leaks, and how to keep cold and hot apart.

Cooling is judged at the server inlet, not at the unit discharge and not at the room average. A hall can read a comfortable average and still have a rack pulling inlet air ten degrees hotter than the units supplied, because the cold air mixed with hot exhaust before it reached the server. Every fault in this guide is a different way that mixing happens, and every fix is a way to stop it.

Is your data center short on cooling or short on airflow?

Most halls that run hot are not short on cooling. They are short on airflow management, and the cooling that was paid for is being lost to mixing before it reaches the load. This is the thesis the rest of the guide rests on. When a rack alarms, the first question is not whether to add a unit. It is where the air is going.

The pattern repeats from site to site. The chillers have capacity to spare, every unit is running, and a row still cooks because cold supply is short-circuiting back to the returns while hot exhaust loops into the inlets. Adding another unit pushes more air into the same leaky room and often makes the airflow worse, because more supply with nowhere good to go is more bypass.

Stranded capacity is the cost. A hall rated for a megawatt on paper can top out at sixty to seventy percent of it because the air never carried a full load of heat. The capacity was there. The commissioning and the housekeeping that make the air reach the load is what nobody finished. Fix the airflow first and the rated capacity usually comes back without a single new unit.

Hot-aisle/cold-aisle discipline as airflow management

Hot-aisle/cold-aisle is the layout that makes airflow management possible, and it is a discipline before it is a piece of equipment. Racks face each other so every cold aisle has server inlets on both sides and every hot aisle has exhausts on both sides. Cold supply enters the cold aisle, gets pulled through the gear, and dumps into the hot aisle as exhaust. Keep that orientation and the two streams stay separable. Break it and they mix.

The orientation breaks in ordinary ways. One rack installed backward pulls its cooling air out of the hot aisle and dumps exhaust into the cold aisle, poisoning the inlets of its neighbors. A rack with its inlet against a wall or another cabinet starves. Network gear with side-to-side airflow sits crosswise to the room and cooks unless it is baffled. Walk the floor and check that every rack breathes front to back, the same direction as the aisle it sits in.

The discipline is not optional once density climbs. At low density a sloppy room forgives itself because there is air to waste. At fifteen kilowatts a rack, one reversed cabinet or one open aisle end shows up as an alarm within the hour. Hot-aisle/cold-aisle is the frame the sealing, the tiles, and the containment all hang on. Get the orientation right first, because none of the later fixes work in a room where the racks face the wrong way.

What is bypass airflow?

Bypass airflow is cold supply that finds its way back to the cooling unit without ever passing through a server. It cost energy to cool and it did no work. The classic source is a perforated tile in the hot aisle, where cold air rises straight into the exhaust stream and rides the return back to the unit. The next is an oversized or unsealed cable cutout under a rack, where supply dumps into the return path through a hole nobody grommeted.

Bypass is quiet because it does not throw an alarm. The room stays cool, the units run, and the only tell is in the numbers. The return air comes back cold, the delta-T collapses, and the units move more and more air to reject the same heat. A floor can be drowning in bypass and look healthy on a walk-through. You find it by reading the return temperatures and by following the cold air with smoke.

The cure is sealing and restraint. Pull tiles out of the hot aisle and out of lightly loaded rows. Grommet the cutouts. Seal the gaps under and beside the units and at the edges of the floor. Resist the reflex to open more tiles when a rack runs hot, because more open area on a leaky floor usually feeds the bypass instead of the rack.

What is recirculation in a data center?

Recirculation is the opposite leak: hot exhaust that loops back into the cold aisle and reaches the server inlets before the unit can cool it. It raises inlet temperatures even when the room average looks fine, and it is what creates the top-of-rack hot spot, because hot air rises and finds the path over the top of a short rack or around an open row end. Recirculation is the fault that actually burns equipment.

The sources are gaps. A missing blanking panel lets exhaust pull straight through an empty rack slot to the inlet side. Gaps at the sides and bottom of a rack, between the mounting rails and the enclosure, let hot air sneak around. Short rows, open aisle ends, and the space above a rack with no containment all let exhaust roll back over the top. Every one of them is a path from the hot aisle to a cold inlet.

Recirculation and bypass both wreck delta-T, but they read differently. Bypass pulls the return temperature down because cold air dilutes it. Recirculation pushes the inlet temperature up because hot air reaches it. The Return Temperature Index separates them: above 100 percent is recirculation, below 100 percent is bypass. The table below is how to tell the two faults apart on the floor.

Blanking panels and rack-level sealing

Sealing the rack is the cheapest, highest-return move in airflow management, and it is almost always done halfway. Start with blanking panels in every empty rack slot. An open U is a window from the hot side of the rack to the cold inlet side, and a column of open Us turns the whole cabinet into a recirculation path. Snap-in panels cost a few dollars a slot and can drop inlet temperatures several degrees the same afternoon.

Then close the gaps the panels do not cover. The space between the mounting rails and the side of the enclosure lets exhaust curl around to the inlet. The gap under the rack to the raised floor or the slab lets it come up from below. The opening where a rack meets its neighbor or the end of a row lets it in from the side. Foam, brushes, and gap-seal kits handle these, and they belong to the rack build, not to an afterthought.

Under the floor, brush grommets seal every cable cutout so the only way out of the plenum is a tile you chose. A cutout you can seal with a grommet during fit-out becomes a permanent loss once a loaded rack is parked in front of it. The checklist below is the order to work it. None of this needs a chilled water valve or a fan curve. It is housekeeping, and it returns more capacity per dollar than anything else on the floor.

• Fit a blanking panel in every empty rack U, top to bottom, with no gaps left open.
• Seal the side gaps between the mounting rails and the enclosure so exhaust cannot curl around.
• Close the gap under the rack to the floor or slab with a base seal or brush.
• Brush-grommet every cable cutout in the raised floor, sizing the brush to the bundle.
• Seal the gaps between adjacent racks and at the ends of each row.
• Cap or seal unused tile openings and any abandoned floor penetrations.
• Baffle or snorkel the top of open racks where there is no aisle containment.

Containment as an airflow tool

Containment is the physical barrier, doors at the ends of an aisle and a roof or ducted ceiling over it, that closes the loop so cold supply cannot short-circuit into the hot return. Seen as an airflow tool, it does one thing: it kills bypass and recirculation at the same time by giving the supply air nowhere to go but through the racks. The delta-T comes up, the inlets even out end to end, and the airflow becomes a balance you can control instead of a guess.

Cold-aisle and hot-aisle containment differ in which aisle you seal, and at the airflow level the trade is about where the heat sits. Cold-aisle containment encloses the supply aisle and lets the rest of the room fill with warm exhaust, which is simpler to add to an existing hall but means the room runs hot and anything outside the aisle lives in return air. Hot-aisle containment encloses the exhaust and ducts it away, so the open room stays cool and habitable, at the cost of more ceiling and ductwork. Both raise delta-T the same way.

The number that shows the payoff is the difference between open and contained delta-T. An open hot-aisle/cold-aisle floor with good housekeeping might hold a delta-T in the low teens because some mixing always sneaks through. Seal the aisle and the same racks return air at the full design rise, because every cubic foot of supply now passes through a server before it can return. That is the whole case for containment in one measurement, and it costs panels and doors, not megawatts.

Delta-T, RTI, RCI, and the airflow metrics

Delta-T is the first airflow metric and the one to read every time. It is the temperature difference between the air returning to the units and the air they supply, and it tells you whether the air did its work. A delta-T near the design rise means the supply passed through the racks and came back loaded with heat. A low delta-T means the air came back barely warmer than it left, which is the signature of mixing.

Three named indices put numbers on the same picture, and they belong in the framework, not on a pedestal. The Return Temperature Index compares the actual rise to the design rise: above 100 percent is recirculation, below 100 percent is bypass. The Rack Cooling Index measures how well inlet temperatures sit inside the ASHRAE band, where 100 percent means nothing is over or under the limit. The Supply Heat Index measures how much heat the supply picks up before it reaches the inlet, so a rising SHI is recirculation heating the cold aisle.

Use the indices to point, not to grade. They are derived from the inlet, supply, and return temperatures you already measure, and they turn a wall of readings into a direction: this hall has recirculation, that one has bypass. Hedge them to the framework and the ASHRAE TC 9.9 envelope they reference, because the threshold that matters is still the inlet temperature at the worst rack, not a perfect score on an index.

Perforated tile placement and the over-provisioning trap

Perforated tiles and floor grilles are the airflow valves of a raised-floor hall, and where they go is a management job that changes as the racks fill in. They belong in the cold aisle, in front of the inlets, and their count and open area get matched to the heat in that row. A standard tile runs around 25 percent open; high-flow grilles reach 50 to 60 percent. The pressure that drives air through them is the plenum static covered in the CRAC-CRAH static pressure guide. The placement is the part this guide owns.

The over-provisioning trap is the most common floor mistake after the evenly spread tile. It feels safe to add tiles and crank fan speed until the hot rack quiets down, but past the point where supply matches demand, the extra air does not go through the racks. It returns through any opening it can find as bypass, drops the delta-T, and burns fan energy for nothing. More tiles is not more cooling. It is often more leak.

Not every hall has a floor to tune. Slab floors with overhead supply and ducted in-room distribution move the same airflow problem up to the ceiling: the question becomes where the diffusers throw and how the return is captured, but the discipline is identical. Match supply to demand at the inlet, keep the hot return from mixing back, and seal the leaks. The raised floor gives you more knobs and more ways to leak. Overhead trades tile-tuning for diffuser layout and stands or falls on the same delta-T.

Matching supply airflow to rack demand

Airflow balance is the core arithmetic of the whole guide: the supply delivered to a cold aisle has to match the air the racks in it pull, at the design temperature rise. Get them equal and the air passes cleanly through the load. Supply less than demand and the racks make up the difference by pulling hot air from the aisle, which is recirculation. Supply more than demand and the excess returns without doing work, which is bypass.

The number to carry is roughly 160 CFM per kilowatt at a 20 F rise. Push more heat and you need more air or a bigger rise; there is no third option. A 10 kW rack at design pulls about 1,600 CFM, so a cold aisle of ten such racks needs about 16,000 CFM of tile airflow to balance. Add up the rack demand, add up the tile delivery, and a gap between them is the airflow fault you are chasing.

This arithmetic is where stranded capacity hides. A hall can have plenty of installed cooling and still run out of usable airflow, because the supply was never matched to the demand row by row. The loaded rows starve while the light rows flood, the delta-T collapses across the whole floor, and the plant hits its airflow ceiling at a fraction of its thermal rating. Balance the air to the load and the rating comes back.

Cable and under-floor obstruction

The plenum under a raised floor is a supply duct, and anything in it that blocks the air is an airflow fault hiding out of sight. Cable trays run crosswise to the flow, piping, and especially abandoned cable left from years of moves and changes choke the cross-section and rob pressure from the far rows. A floor can pass every unit-level check and still starve a distant aisle because the air cannot physically get there.

Abandoned cable is the worst of it because nobody owns it. Each project pulls new cable and leaves the old in place, and over a decade the plenum fills with dead bundles that do nothing but block air. Some codes now require abandoned cable to be removed for fire reasons, and the airflow reason is just as real. Pulling dead cable out of a choked plenum can recover the pressure a far row needs without touching a single cooling unit.

The same logic applies overhead and inside the rack. A return path crammed with cable trays does not let the hot air leave. A rack stuffed front to back with slack cable blocks its own airflow and creates a hot spot the room cannot fix. Manage the cable as part of the airflow, route it out of the supply and return paths, and treat the plenum survey as a real commissioning item, not a glance under one tile.

Airflow commissioning and ongoing management

Airflow does not get commissioned once and stay commissioned, because the load moves every time a rack is added or pulled. The work is a measure-then-tune loop, run at acceptance and again whenever the floor changes. Measure the inlet temperatures across the hall top and bottom, measure the tile airflow with a flow hood, read the return delta-T per unit and per aisle, then tune the tiles and the sealing to close the gaps and measure again.

Airflow visualization is how you see what the numbers imply. A smoke pencil or a haze generator in the cold aisle shows supply escaping at a cutout or a rack gap, and in the hot aisle it shows cold air bleeding in. CFD modeling does the same thing in design and in troubleshooting: it predicts where the air will go before the racks land and tests a fix before you move a tile. Treat CFD as a tool for both the basis of design and the live-floor problem, not as a one-time drawing.

Ongoing management lives in the telemetry. DCIM and inlet-sensor networks trend the inlet temperatures, the delta-T, and the aisle pressures continuously, so the slow drift from a filling row or a new hot rack shows up as a trend instead of an alarm. The acceptance test sets the baseline; the telemetry holds the floor to it. A hall with no airflow baseline turns every warm aisle into a fresh investigation instead of a check against how it has always run.

The airflow-first remediation sequence

When a rack runs hot, work the air before the tonnage, and work it in order from cheapest to dearest. The sequence is the same on almost every floor, and skipping to the end is what wastes money on units that make the problem worse. Seal the rack, blank the gaps, fix the tiles, contain the aisle, and only then consider adding cooling. By the time you reach the last step, the alarm is usually gone.

The order is not arbitrary. Sealing the rack and fitting blanking panels kills the recirculation closest to the inlet for almost nothing. Grommeting cutouts and pulling stray tiles kills the bypass next. Re-tiling matches the supply to the loaded rows. Containment closes whatever mixing survives. Each step is cheaper and faster than the one after it, and each one buys back capacity that an extra cooling unit would not, because the extra unit does not fix mixing.

Adding cooling is the last resort, not the first call, and after the airflow work it is often unnecessary. If you do reach it, you are now adding capacity to a floor where the air actually carries its heat, so the new unit does real work instead of feeding the leak. The checklist below is the order to follow, and the discipline is to resist jumping to the bottom of it when the top steps cost a fraction as much.

• Seal the rack: blanking panels in every empty U, side and bottom gaps closed.
• Blank the floor leaks: brush-grommet cutouts and seal under-unit and edge gaps.
• Fix the tiles: pull tiles from light rows and the hot aisle, concentrate them on the loaded racks.
• Contain the aisle: close the ends and the top so supply has nowhere to go but through the racks.
• Re-measure the inlets and the delta-T, and only then size any added cooling against what remains.

Field example: capacity stranded by mixing

A 1.2 MW hall started throwing high-temperature alarms at the top of the racks at about 800 kW, well short of design. The chillers had capacity to spare and every unit was running, so the reflex was that the cooling was undersized. The airflow numbers said otherwise, and the fix cost panels and grommets, not a new unit.

Return air came back at a delta-T near 9 F against a 20 F design, and the Return Temperature Index sat below 100 percent, the bypass signature. A walk found the causes in three ordinary places. Roughly a third of the empty rack slots had no blanking panels, several cable cutouts under the dense rows were wide open, and perforated tiles were spread evenly instead of concentrated in front of the loaded racks. Cold air was leaking back to the units while hot air rolled over the open racks at the same time.

Fitting blanking panels, grommeting the cutouts, and moving tile airflow to the loaded rows brought the return delta-T up toward 17 F over the next shifts. With the air carrying its heat, the same units held the hall past 1.0 MW inside the envelope and the top-of-rack alarms stopped. No equipment was added. The capacity had been there the whole time, stranded by mixing the commissioning never closed out.

What to document

An airflow floor that was tuned but never recorded leaves operations with no baseline, so the first warm aisle six months later becomes an investigation instead of a check. The record is the as-balanced state of the air, and it is what the next engineer trends every airflow problem against. Capture it at acceptance and update it whenever the floor changes.

Capture the inlet temperature map top and bottom across the hall, the return delta-T per unit and per aisle, the tile layout and open area by row, the rack-level sealing state, the containment configuration and any aisle pressures, and the airflow metrics if you compute them. Note what you sealed and re-tiled, with before-and-after readings, because a change you cannot show is a change the next person will not trust.

Common mistakes

• Adding a cooling unit to a hot hall before checking where the air is going.
• Judging cooling by the room average instead of the temperature at each server inlet.
• Leaving empty rack slots without blanking panels, so exhaust recirculates to the inlet.
• Ignoring the side and bottom rack gaps that let hot air curl around to the cold side.
• Putting perforated tiles in the hot aisle, dumping cold supply straight into the exhaust.
• Spreading tiles evenly instead of matching tile airflow to the rack load in each row.
• Over-provisioning airflow with extra tiles and fan speed, feeding bypass and wasting fan energy.
• Leaving floor cutouts ungrommeted, so cold supply leaks back to the units as bypass.
• Letting abandoned cable choke the plenum so the far rows starve no matter the plant size.
• Tuning the floor once and never re-measuring as the racks fill in and the load moves.

Field checklist

This is the airflow-management walk, run at acceptance and again whenever the floor changes. It assumes the system type is already chosen and the plenum pressure is already set, both covered in the sibling guides.

Standards and references

The thermal target airflow management has to satisfy comes from ASHRAE Technical Committee 9.9 and its Thermal Guidelines for Data Processing Environments, which set the recommended and allowable server-inlet envelopes the inlet map is judged against. The airflow metrics in this guide, the Return Temperature Index, the Rack Cooling Index, and the Supply Heat Index, are industry air-management measures derived against that envelope, not ASHRAE mandates. Confirm the current edition and the actual numbers against the published guideline and the equipment manufacturer.

The airflow delivery itself is the province of the test-and-balance bodies, AABC and NEBB, whose procedures a witnessed balance follows, and the basis of design is the project document that sets the design delta-T, the supply temperature, and the containment scheme. The commissioning process framework is the ASHRAE commissioning guidance, commonly Guideline 0, and the commissioning agent is the one who witnesses the inlet map and the airflow balance. Where a facility chases a tier, the Uptime Institute Tier standards drive the witnessed demonstrations, and TIA-942 is the broader data center infrastructure standard.

Two siblings carry the pieces deliberately left out here. The air-versus-liquid families and the CRAC-versus-CRAH choice live in the cooling system types guide. The unit fan curve and the underfloor plenum static pressure that this guide treats as a given live in the CRAC-CRAH static pressure guide. Cite the body that owns the point, hedge the metric to the framework, and let the basis of design and the equipment data override any rule of thumb.

Units, terms, and acronyms

Airflow management carries vocabulary from HVAC, from test and balance, and from the IT side, and the same idea reads differently across a TAB report, a DCIM dashboard, and a commissioning script. The terms below travel across the airflow-management scope.

Airflow management

Getting cold supply to every inlet and hot exhaust back to the units without the two streams mixing

Bypass airflow

Cold supply that returns to the unit without passing through a server, doing no cooling work

Recirculation

Hot exhaust that loops back into the cold aisle and raises the server inlet temperature

Delta-T

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

CFM

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

Blanking panel

A filler in an empty rack U that blocks the recirculation path from exhaust to inlet

Brush grommet

A sealed bristle insert for a floor cable cutout that passes cable while stopping bypass air

Containment

Doors and a roof that seal a cold or hot aisle so the two air streams cannot mix

RTI

Return Temperature Index; above 100 percent is recirculation, below 100 percent is bypass

RCI

Rack Cooling Index; how well inlet temperatures sit inside the ASHRAE envelope, 100 percent is clean

SHI

Supply Heat Index; the fraction of heat the supply picks up before reaching the inlet

Tile open area

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

Related tools

Calculators and readiness checks for this work

Compare your options

FAQ

People also ask

• How are data centers cooled?
• What underfloor static pressure should a data center plenum hold?
• What is a blanking panel in a server rack?
• What is the difference between hot-aisle and cold-aisle containment?
• What is rack readiness in a data center?