Data center noise and acoustics control field guide

Data center noise is a two-front problem

Data center noise is the sound produced by the cooling, power, and IT equipment, and it has to be controlled in two directions at the same time. Inside, the data hall and the mechanical and generator rooms get loud enough that workers need hearing protection. Outside, the chillers, cooling towers, dry coolers, and the generators on test push noise across the property line toward whoever lives next door.

Most teams treat these as one job and end up failing both. The two fronts have different limits, different measurement points, and different fixes. The worker side is governed by OSHA hearing rules and measured at the person's ear over a shift. The community side is governed by the local noise ordinance and measured at the property line, usually with a tighter limit at night. A wall that protects the neighbor does nothing for the technician standing between two rows of GPU racks.

The expensive version of this is the one that shows up after the building is running. The neighbors file complaints, the county pulls out a sound ordinance written for block parties, and now the owner is retrofitting barriers and silencers onto equipment that was sited at the fence line. Design for noise up front and you spend a fraction of that. The generator and the CRAH fans are two of the loudest sources, and both are covered in their own guides.

How loud is a data center?

A data center is loud enough to require hearing protection in the data hall and the mechanical rooms. Published field measurements put average levels in server areas in the range of 85 to 96 dBA, with a single high-density rack producing 75 to 80 dBA on its own and a row of them in an enclosed aisle adding up well past that. Chillers can reach about 100 dBA at the machine, and large generators on a run approach the same.

Those numbers sit at or above the level where OSHA hearing conservation kicks in, which is the whole point. The data hall is not a quiet office with a hum. It is an industrial space, and the people who underestimate it are the ones doing a quick task without protection because they will only be in there a minute.

Outside is a different scale. By the time the noise reaches the property line a few hundred feet away, distance and barriers have knocked it down, and the limit you are checking against is the ordinance figure, commonly somewhere in the 45 to 65 dBA range depending on the jurisdiction and the time of day. The contrast is the issue: the same plant that needs ear protection inside has to read like a quiet suburb at the fence.

Decibels, dBA, and the logarithmic scale

Sound level is measured in decibels, and the decibel is a logarithmic ratio, which is why the math does not behave the way people expect. Doubling the sound power adds only 3 dB, a change you can barely hear. To sound roughly twice as loud to a person takes about a 10 dB increase, which is ten times the power. So two generators running together are 3 dB louder than one, not double, and that surprises people every time.

The A-weighting is the other piece. A sound level meter set to dBA applies a filter that rolls off the very low and very high frequencies to approximate how the human ear actually responds. dBA is the unit OSHA and most noise ordinances are written in, because it tracks perceived loudness and hearing risk better than a flat measurement.

The hedge worth carrying: dBA can understate low-frequency noise, the deep hum that travels far and rattles a neighbor's window. That is why some data center ordinances now add octave-band or C-weighted limits on top of the dBA limit. If the complaint is a low rumble and your dBA reading looks fine, the A-weighting is hiding the problem, and you measure the low end before you argue you are in compliance.

dB

Decibel, a logarithmic ratio of sound power or pressure, where 3 dB is a doubling of power

dBA

A-weighted decibel, filtered to approximate human hearing, used by OSHA and most noise ordinances

dBC

C-weighted decibel, flatter at low frequency, used to catch the deep hum dBA misses

What makes the inside loud

The dominant noise source inside the white space is the IT equipment itself. Every server has a bank of small high-RPM fans, and a full rack carries dozens of them spinning hard to pull air through tightly packed components. Multiply that by a row of racks and the data hall hum becomes a roar that does not let up, because the load never stops.

The cooling units add the next layer. CRAC and CRAH units move large volumes of air with big fans, and when those fans ramp to chase a rising inlet temperature they get louder fast. The CRAH setup and fan control are covered in the airflow guide, and fan noise is one more reason to run them on the right control signal rather than wide open. Pumps for chilled water and the power gear, the PDUs and the UPS, add a steady mechanical and electrical hum underneath everything.

What makes the white space hard is that the noise is broadband and constant, and the room is full of hard reflective surfaces, raised floor, equipment, and bare walls, that let it build. There is no quiet moment to recover hearing in. That is why the data hall, not the brief loud event, is the exposure that adds up over a shift.

The AI density problem is making data halls louder

High-density racks built for AI and GPU workloads are the new noise story, and it is a real shift, not a marketing line. Higher power per rack needs more aggressive cooling, and more aggressive air cooling means more fans turning faster. Fan noise rises sharply with speed, so as rack density climbs to feed AI compute, the data hall gets meaningfully louder than the same floor space did a few years ago.

This matters most on the worker side. A floor that used to read in the mid 80s dBA can climb into the 90s as density goes up, which moves more workers past the OSHA action level and shortens how long anyone can be in there without protection. The hearing conservation program that was borderline becomes mandatory.

Liquid cooling is part of the answer, and part of why it is being adopted is exactly this. Direct-to-chip and immersion cooling move heat with far less air, which cuts the fan count and the fan speed, so the acoustic load drops along with the thermal one. A hall going liquid for thermal reasons gets quieter as a side benefit. But the hybrid floors, still air-cooled in places, are where the noise jump is sharpest, and that is the floor to measure rather than assume.

What makes the outside loud

Outside, the loud equipment lives in the yard and on the roof, and it is what the neighbors hear. The generators are the loudest source when they run, and they run on a schedule, which is its own problem covered below. The cooling plant is the constant one: chillers, cooling towers, dry coolers, and air-cooled condensers all reject the building's heat to the outdoors, and they do it with large fans and moving water.

Cooling towers add fan noise and the broadband splash of water cascading over fill. Air-cooled chillers and dry coolers are banks of fans running whenever the building is hot, which in a data center is always. Large rooftop air handlers can read 85 to 100 dBA at the unit depending on size. Even the intake and discharge louvers on the building skin radiate noise from the air moving through them.

The community side is driven by the equipment that runs continuously, not the generator that runs an hour a week. A neighbor adapts to a brief weekly test more easily than to a chiller plant that hums at 2 a.m. every night of the year. So the cooling yard, sited and treated correctly, usually decides whether the property line passes, with the generator as the loud event you also have to control.

What noise limits protect the workers?

OSHA sets two numbers that drive everything on the worker side, and they are easy to confuse. The action level is 85 dBA as an 8-hour time-weighted average. Cross it and the employer owes a hearing conservation program: noise monitoring, baseline and annual audiometric testing, hearing protection made available, training, and recordkeeping. The permissible exposure limit, the PEL, is 90 dBA as an 8-hour average, and at that level the employer is required to use feasible engineering and administrative controls.

The exchange rate is the part people miss. OSHA uses a 5 dB exchange rate, so the allowable time halves for every 5 dB over 90. That means 8 hours at 90 dBA, 4 hours at 95, 2 hours at 100. In a data hall running in the 90s, the allowable unprotected time is short, and a tech who spends a full shift on the floor is over the limit well before lunch.

Be blunt here, because this is the hazard. If the room is over 85 dBA and a person is in it for the shift, they wear hearing protection, full stop. Post the loud rooms with signage, keep plugs and muffs at the door, and run the program. Confirm the current OSHA noise standard and any state-plan rules, because the adopted requirements control, but the action level and PEL are long-standing. Noise-induced hearing loss is permanent and it does not hurt while it happens.

What noise limits apply at the property line?

The community limit is set by the local noise ordinance or the zoning conditions, and it is almost always a dBA limit measured at the property line, with a lower limit at night. The exact figures vary by jurisdiction, but a common pattern is something in the range of 55 to 65 dBA in the daytime and 45 to 55 dBA at night, with the nighttime number being the one that usually controls the design because mechanical noise carries further when the background is quiet.

There is no single national number, so this hedges hard to the local ordinance and any conditions attached to the special-use permit. Many counties have rewritten or are rewriting their ordinances specifically because data centers exposed how the old rules, written for barking dogs and parties, did not handle a continuous industrial hum. Some now require a pre-construction sound study, a post-construction verification, equipment screening, and minimum setbacks, with tighter limits if the setbacks are not met.

The conflict is real and it is local. A data center next to a residential neighborhood is one of the most common noise disputes in the industry right now. The low-frequency hum is the usual culprit, because it travels far and the standard dBA limit can read compliant while the neighbor is still kept awake. Find the ordinance, find whether it has a low-frequency or C-weighted clause, and design to the night limit at the nearest receptor, not the average.

Generator noise is the loudest event

The standby generators are the loudest single source at most data centers, and they do not run quietly in the background. They run on a schedule, the weekly or monthly no-load or load-bank exercise required to keep the plant proven, and they run hard during an actual utility outage. A large diesel set can read close to 100 dBA, and a site can have many of them. The acceptance and load-bank testing that drives those runs is its own guide, and it is worth coordinating the noisy test windows with the community expectations.

Generator noise comes from three places, and you treat each separately. The engine casing and the cooling fan radiate noise from the set itself. The radiator discharge throws air and noise out one face. And the exhaust, the loudest path, carries the combustion pulse out the stack. Quieting a generator means handling all three, because silencing the exhaust does nothing about a screaming radiator fan.

The timing makes it a community-relations problem as much as an engineering one. A monthly test at 2 p.m. on a weekday lands differently than the same test at 6 a.m. on a Sunday. Pick the test window with the neighbors in mind, and treat an extended outage run, when the sets may run for hours, as the worst case the silencing has to cover.

Generator enclosures and exhaust silencers

Two controls do most of the work on a generator: the enclosure and the exhaust silencer. A sound-attenuated walk-in enclosure wraps the set in a steel housing lined with sound-absorbing material, with baffled intake and discharge scoops that let air through while redirecting noise. A bare steel enclosure gives at least about 10 dBA, and a purpose-built sound-attenuated enclosure does considerably more depending on its rating.

The exhaust silencer, the muffler, is graded by how much it cuts the exhaust noise, and the grade is a number you specify, not an accident. The common grades run industrial at roughly 15 to 20 dBA, residential at 20 to 25, critical at 25 to 32, super-critical at 30 to 38, and hospital at 35 to 42 dBA of reduction. For a data center near homes, a critical-grade silencer is a common minimum, and the tighter the property-line limit, the higher up that ladder you go.

The mistake that bites is silencing one path and forgetting the others. A critical-grade exhaust silencer on a set with an unenclosed radiator fan still throws a wall of fan noise out the discharge. Match the enclosure rating, the silencer grade, and any discharge treatment to the same target, and verify the combination against the property-line number, not the silencer's standalone rating.

Cooling plant noise

The cooling plant is the continuous outdoor source, so it usually sets the steady property-line level. Chillers carry compressors that can read up to about 100 dBA at the machine. Cooling towers combine fan noise with the broadband sound of water falling through the fill. Dry coolers and air-cooled condensers are banks of fans that run whenever the building rejects heat, which for a data center is around the clock.

Because it runs continuously, the cooling plant is where low-noise equipment choices pay off the most. Low-noise or low-sound fan options, often larger-diameter fans turning slower for the same airflow, cut the noise at the source before any barrier sees it. Variable-speed fan control helps twice: the plant rarely runs at full design load, and at part load the fans slow down and quiet down, especially at night when the limit is tightest and the ambient is coolest.

Specify the sound power rating of the cooling equipment up front and make it part of the selection, the same way you weigh efficiency. Manufacturers publish sound ratings under AHRI and ASHRAE test methods, so you can compare units before they are on a purchase order. Picking the quieter chiller or the low-noise tower at selection is far cheaper than wrapping a loud one in barriers after it is set on the pad.

Mitigation: barriers and sound walls

A sound barrier wall blocks the direct path from the equipment to the receptor. It works by breaking the line of sight: if you can see the source over or around the wall, sound takes the same path, so the barrier has to be tall enough and long enough to interrupt that straight line between the noisy equipment and the nearest property line or neighbor.

Two properties make a barrier effective. It has to block the line of sight, and it has to have enough mass that sound does not simply pass through it. The mass law is the rule of thumb: roughly every doubling of the wall's surface weight buys about 6 dB more transmission loss, so a heavy solid wall outperforms a light fence even at the same height. Gaps kill it. A barrier with a hole at the bottom or an open joint leaks sound through the gap and the rated performance never shows up at the receptor.

Barriers are most effective close to the source or close to the receptor, and they do more for higher frequencies than for the deep low-frequency hum, which tends to bend over and around the top. A barrier is one tool in the stack, not the whole answer. Pair it with quieter equipment and distance rather than expecting a wall to carry the property line by itself.

Mitigation: enclosures, lagging, and acoustic treatment

Where a barrier is not enough, you wrap the source. An acoustic enclosure boxes the equipment in an absorptive, sealed housing, which is the generator approach applied to pumps, compressors, and other point sources in the yard. The catch with any enclosure on running equipment is airflow and heat: the enclosure has to breathe, and the intake and discharge openings are the weak point unless they carry their own silencing.

Lagging handles the noise that radiates off pipes and ducts. Acoustic pipe and duct lagging is a layered wrap, a limp mass barrier over a soft decoupling layer, that adds mass and damping to a surface that would otherwise act like a speaker. Chilled-water piping, large ductwork, and exhaust runs all radiate noise along their length, and lagging the run is often the difference between a treated source and a quiet-looking enclosure that leaks through its own connected pipes.

Inside, absorptive treatment on hard surfaces takes the edge off the reverberant buildup, though in a data hall packed with equipment the room is dominated by the source noise, not the room acoustics. Treatment earns its place more in the NOC, the offices, and the corridors than in the white space itself.

Mitigation: silencers and duct attenuators

Silencers and attenuators quiet the air paths without blocking the air. Any opening that lets air in or out, an exhaust stack, a ventilation intake, a generator-room louver, a fan discharge, is an open path for noise unless something in the path absorbs it. A silencer is a length of lined duct, sometimes with baffles, that the air passes through while the sound is absorbed against the lining.

Duct attenuators do the same job on the building's air systems, and louver silencers handle the wall openings on mechanical and generator rooms so the room's noise does not pour straight out the vent. A well-designed silencer or attenuator in a duct or exhaust path can give in the range of 10 to 30 dB depending on its length and how it is sized. The trade-off is always pressure drop: a silencer that is too restrictive starves the equipment of air, so it is sized for both the insertion loss and the allowable static loss together.

Silencer performance is tested by a standard method, commonly ASTM E477 for insertion loss and pressure drop, so the published numbers are comparable between products. Specify the insertion loss you need across the frequencies that matter and the maximum pressure drop the equipment can tolerate, and let those two numbers size the silencer rather than picking by length alone.

Mitigation: distance and siting

Distance is the cheapest control you have, and it is free if you plan for it. Sound from a point source spreads out, so the level drops about 6 dB for every doubling of distance in the open. Move the loud equipment from 100 ft to 200 ft from the property line and you have bought roughly 6 dB before any barrier or silencer is involved.

That makes siting the decision that pays back the most on the project, and it is made on the site plan long before any equipment is selected. Put the generators, chillers, and cooling towers on the side of the building away from the nearest homes, use the building mass itself to shield the receptor, and hold the setback the ordinance asks for, or more. A data center that puts its quiet loading and office side toward the neighborhood and its loud equipment yard toward a highway or an industrial parcel has solved most of the community problem with geometry.

The 6 dB figure is a free-field rule of thumb. Reflections off the building, the ground, and hard surfaces fill it in, so do not bank the full theoretical drop. But the direction is reliable: every foot of separation you design in is a foot you do not have to buy back later with a taller barrier or a higher silencer grade.

Vibration isolation

Not all of the noise travels through the air. Generators, pumps, chillers, and fans shake the structure they sit on, and that vibration travels through the slab, the steel, and the walls and re-radiates as sound somewhere else in the building, often as a low hum in a room nowhere near the equipment. This is structure-borne noise, and a barrier or a silencer does nothing for it because it never went through the air.

The control is vibration isolation: you set the equipment on isolators that decouple it from the structure so the shaking is absorbed instead of transmitted. Spring isolators handle low-frequency, heavy machinery like generators and large chillers, while rubber or neoprene mounts and pads suit lighter or higher-speed gear. Isolators are selected for the equipment's operating speed and weight so their natural frequency sits well below the disturbing frequency, which is what lets a properly chosen mount reach 90 percent or better isolation.

The piping and conduit are the bypass everyone forgets. You can isolate a chiller beautifully and then hard-pipe it to the building with rigid chilled-water lines, and the vibration walks straight out through the pipe into the structure. Flexible connectors at the equipment and resilient pipe supports keep the isolation from being short-circuited by its own connections.

The noise study and acoustic model

The noise study is the design step that predicts the property-line level before anything is built. An acoustic consultant takes the sound power ratings of the planned equipment, the site layout, the distances, the barriers, and the ground and building geometry, and runs an acoustic model that estimates the dBA at the nearest receptors. The output is a predicted property-line level you compare against the ordinance limit while there is still time to change the design.

This is increasingly not optional. Many jurisdictions now require the study as a condition of the permit or the special-use approval, and the study often has to show compliance at the day and night limits at the nearest homes. If the model says the night limit fails, you fix it on paper, with siting, quieter equipment, barriers, or higher silencer grades, which is where the cost difference between design-in and retrofit is decided.

Treat the model as a planning tool with a margin, not a guarantee. Models make assumptions about ground absorption, weather, and equipment that the real site will not match exactly, so a design that predicts right at the limit has no room for the assumptions to be optimistic. Build in a few dB of headroom so the post-construction measurement does not put you over a number you cleared on paper.

How do you measure data center noise?

You measure noise on both fronts with a sound level meter, but the where and the how differ for each. For the worker side, you measure the personal exposure at the technician's ear over the work shift, typically with a noise dosimeter that integrates the 8-hour time-weighted average, because what matters is the dose a person actually accumulates, not a spot reading. For the community side, you measure the level at the property line or the nearest receptor with a sound level meter, day and night, and compare it to the ordinance.

Calibrate the meter before and after a session and use the weighting the limit is written in, dBA for most, with octave-band or C-weighted readings added when the concern is the low-frequency hum. Measure with the equipment running at the condition you care about: full cooling load for the steady community level, and an actual generator run for the worst case. A property-line measurement taken on a cool night with the chillers idle proves nothing about the hot afternoon.

Background noise is the trap on the community side. If the ambient without your equipment is already near the limit, you have to separate the equipment's contribution from the background, which the standard measurement methods account for. A reading taken next to a busy road tells you about the road, not the data center, so the measurement protocol and the time of day are part of getting an honest number.

Interior acoustics: the NOC, offices, and walls

Inside the building, the acoustic job splits between the loud spaces and the spaces that have to stay quiet. The data hall is loud and reverberant by nature, and you accept that with hearing protection rather than try to make it quiet. The network operations center, the offices, and the conference rooms are the opposite: people work there for the full shift and the data hall noise next door cannot follow them in.

That comes down to the wall and its rating. The Sound Transmission Class, the STC, rates how well a wall assembly blocks airborne sound between spaces, and a higher number blocks more. A NOC sharing a wall with a data hall needs an assembly with enough STC, and sealed penetrations, to keep the hall's roar from bleeding through. The wall is only as good as its weakest path, so a high-STC wall undone by an unsealed cable penetration or a shared ceiling plenum performs like the gap, not the rating.

The detail that gets missed is the path around the wall. Cable trays, conduit penetrations, and a continuous ceiling plenum over the top of a partition all carry sound past a wall that looks solid on the drawing. Seal the penetrations and break the plenum, or the STC on the spec sheet never shows up in the quiet room.

Design it in, do not retrofit it

The single biggest lesson on data center noise is that designing for it up front costs a fraction of fixing it after complaints. Siting the loud equipment away from the neighbors, selecting quieter chillers and towers, specifying the right silencer grade, and modeling the property line are decisions that cost almost nothing extra at design and a great deal to add later.

The retrofit version is the nightmare case. The plant is built, the neighbors complain, the county enforces an ordinance, and now the owner is craning barriers into a finished yard, swapping silencers on running generators, and lagging pipe in occupied mechanical rooms, all while the equipment cannot come offline because the load is live. The work costs more, performs worse because it is fitted around what already exists, and comes with a community already angry.

Catch it at design and the cost lands where it belongs, in the model and the equipment selection. Skip it and you own the retrofit, the bad press, and a property line you may never quite bring into compliance.

Commissioning the noise controls

The noise controls get commissioned like any other system, by verifying they perform at turnover rather than assuming the spec sheet came true. The acceptance check confirms the actual installed condition: the property-line level under real load, a generator run measured at the fence, the worker exposure on the floor, and the physical controls in place and intact.

Verify the things that quietly go wrong in construction. The silencer that got value-engineered to a lower grade than specified. The enclosure with a panel left off or a door gasket missing. The vibration isolators shipped with their shipping blocks still bolted in, which hard-connects the equipment to the slab and defeats the isolation entirely. The barrier with a gap at grade. Each of those reads fine on the submittal and fails in the field.

Tie the acceptance to the predicted numbers from the noise study and to the OSHA exposure limits, and document the measured results against both. A noise control program that was modeled, specified, and installed but never verified is one undiscovered gap away from a property-line failure or an over-exposed worker, and the time to find the gap is at turnover, not when the complaint arrives.

What to document

The noise record answers the two questions that come up later: are the workers protected, and is the property line in compliance. For each source, capture the concern, the control applied, and the measured result, so the file shows the chain from a loud machine to a number that passes.

Keep the worker side and the community side as parallel records. On the worker side, the exposure monitoring, the hearing conservation program status, and the signage in the loud rooms. On the community side, the noise study prediction, the post-construction property-line measurements at day and night, and the permit conditions you are holding to. When a complaint or an audit comes, the record is what you have.

Common mistakes

• No noise study at design, so the community limit is ignored until the complaints arrive.
• Treating the worker side and the community side as one job with one fix.
• No hearing protection or signage in the data hall and mechanical rooms running over 85 dBA.
• Specifying a generator without a critical-grade exhaust silencer or with the enclosure rating too low for the property line.
• Silencing the generator exhaust but leaving the radiator fan and discharge untreated.
• Siting the loud equipment yard at the property line nearest the neighbors instead of using distance and the building as a shield.
• Skipping vibration isolation, or leaving the shipping blocks in the isolators, so the hum travels through the structure.
• Hard-piping isolated equipment with rigid connections that short-circuit the isolation.
• Reading dBA only and missing the low-frequency hum that the A-weighting hides.
• Ignoring the AI-density noise jump and assuming a high-density floor reads like the old one.
• Designing right at the ordinance limit with no headroom for the model's assumptions.
• Retrofitting barriers and silencers after the build instead of designing them in.

Field checklist

Standards and references

The worker side runs on the OSHA occupational noise exposure standard. The 85 dBA action level triggers the hearing conservation program, and the 90 dBA 8-hour permissible exposure limit triggers engineering and administrative controls, with a 5 dB exchange rate that halves the allowed time per 5 dB. Confirm the current OSHA standard and any state-plan amendments, because the adopted requirements control, though these thresholds are long-standing.

The community side runs on the local noise ordinance and the zoning or special-use conditions, which set the property-line dBA limits and the day and night periods. There is no single national limit, so this hedges entirely to the jurisdiction. Many counties have revised their ordinances specifically for data centers and now require sound studies, setbacks, and screening, sometimes with octave-band or C-weighted limits for low-frequency noise.

On the engineering side, equipment sound ratings come from AHRI and ASHRAE test methods, so chillers, towers, and air handlers can be compared on published sound power before selection. The ASHRAE Handbook chapter on noise and vibration control is the standard reference for the design methods. Silencer insertion loss and pressure drop are tested by methods such as ASTM E477. An acoustic consultant ties these together into the noise study, and bringing one in early is what separates a plant that passes from one that gets retrofitted.

Units, terms, and conversions

Noise work uses a small set of units and terms that show up across the ordinance, the OSHA standard, and the equipment data, often in slightly different forms.

Sound level is in decibels (dB), almost always A-weighted (dBA) for limits and sometimes C-weighted (dBC) for low-frequency. Sound power level (Lw) describes the source itself and is what manufacturers publish; sound pressure level (Lp) is what a meter reads at a location, and the two are different numbers for the same machine. Worker exposure is an 8-hour time-weighted average (TWA), and a noise dose is the percentage of the allowed exposure used. Octave bands break the level into frequency ranges to find where the noise lives.

dBA / dBC

A-weighted decibel for most limits; C-weighted decibel to capture low-frequency noise dBA misses

Sound power (Lw) vs sound pressure (Lp)

Lw is the source's emission, what manufacturers rate; Lp is the level measured at a location

TWA

Time-weighted average, the noise exposure averaged over an 8-hour shift for OSHA

Exchange rate

How allowed exposure time changes with level; OSHA uses 5 dB, halving the time per 5 dB

Insertion loss

The noise reduction a silencer or barrier provides, the difference with and without it

STC

Sound Transmission Class, how well a wall assembly blocks airborne sound between rooms

Related tools

Calculators and readiness checks for this work

Compare your options

FAQ

People also ask

• What is an NFPA 110 acceptance test?
• What underfloor static pressure should a data center plenum hold?
• What humidity should a data center be?
• How do data centers control physical access?
• What are zinc whiskers?