Commercial HVAC & refrigeration for Data Centers, Server Rooms & Telecom

Who this page is for, and who it is not for

Data center is a broad word, and the honest first step is to sort the facilities that suit a general commercial mechanical contractor from the ones that do not. Hyperscale and large wholesale campuses need specialists. A purpose-built campus with megawatts of critical load, engineered chilled-water plants, adiabatic or free-cooling systems, and tightly written commissioning programs is a different trade from the one most commercial contractors practice. Those owners generally work with design-build teams, controls integrators, and service firms that live inside that world every day. If that describes your site, we would rather tell you so now than waste your time.

Where Davinci Mechanical fits is the large middle of the market, which is where most of the actual computing and switching in Southern California sits:

• Enterprise server rooms inside office buildings, hospitals, schools, city halls, courthouses, and corporate campuses, usually cooled by one to four precision or split-system units.
• Edge sites and telecom switching spaces such as small regional facilities, cell and fiber huts, remote equipment shelters, and central-office style rooms where uptime commitments are high but the physical footprint is small.
• Small and mid-sized colocation suites where the cooling is a handful of CRAC or CRAH units, in-row coolers, or a modest chilled-water loop.
• MDF and IDF closets that were never designed as equipment rooms but now hold the switches, firewalls, and storage that a whole building depends on.
• Network operations centers (NOCs) and dispatch or command spaces with heavy equipment and people in the same room.
• Support cooling and mechanical infrastructure around larger facilities: chilled-water pumps, glycol loops, condenser and heat-rejection equipment, make-up air, and the building HVAC that surrounds a computer room.

Many of the owners who contact us are not IT people at all. They are facility managers who inherited a server room when a department moved, and who are now responsible for cooling they did not design. If that is you, the sections below are meant to give you a working vocabulary. For the broader approach we take to finding the cause of a problem rather than swapping parts, see our root-cause diagnostics guide.

CRAC and CRAH units in plain language

The two acronyms describe the two traditional kinds of computer room air handler, and the difference is simply where the cold comes from. A CRAC (computer room air conditioner) carries its own refrigeration circuit: a compressor, an evaporator coil, an expansion device, and a connection to a condenser that rejects heat outdoors, or to a glycol or water-cooled heat exchanger. A CRAH (computer room air handler) has no compressor. It is a large fan and coil, and the coil is fed chilled water from a central chiller plant.

That distinction drives everything that follows. A CRAC is self-contained, so one unit can serve one room anywhere you can get a refrigerant line or a glycol loop to. A CRAH depends on the plant behind it, so a chilled-water problem upstream affects every CRAH at once. CRACs fail through compressors, contactors, refrigerant leaks, and condenser problems. CRAHs fail through valves, actuators, pumps, strainers, and plant control issues. The troubleshooting tree is different, and the person who arrives needs to know which tree they are standing in.

Both types come in downflow and upflow arrangements. Downflow units sit on a raised floor and push cold air into the plenum below, where perforated tiles deliver it to the racks. Upflow units discharge from the top, often into ducting or a ceiling plenum, and suit slab-floor rooms. Fans, filters, humidifiers, reheat elements, and condensate pumps are the support components that most often generate alarms, and they are also the parts that are quietly neglected in a room that has run without incident for years.

Plenty of existing CRACs in Southern California are old enough to be running legacy refrigerants and obsolete control boards. When a unit reaches that stage, a repair plan and a replacement plan have to be considered side by side. Our commercial repair service starts with documentation and measurement so that decision is made on evidence, not on the last emergency.

In-row cooling and containment

Room-level cooling has a weakness: cold air and hot air travel across the whole room, mixing along the way. Most of the engineering effort in modern computer rooms goes into stopping that mixing, and two ideas do most of the work.

In-row and close-coupled cooling

An in-row cooler is a narrow unit that sits between racks, in the same line as the equipment it serves. It draws hot air from the hot aisle, cools it, and discharges cold air toward the cold aisle, a few feet from where the load is. Because the air only travels a short distance, there is less opportunity for mixing, less fan energy, and better control of hot spots in dense rows. Variants sit above racks or attach to the back door of a rack. In-row units suit slab-floor rooms, retrofits where a raised floor does not exist, and rooms with a few high-density rows that the room-level system cannot serve evenly.

Hot-aisle and cold-aisle containment

Containment is the physical separation of supply and return air. In a cold-aisle arrangement, the aisle the servers draw from is enclosed with doors at the ends and a roof overhead, so cold supply air can only travel through the equipment. In a hot-aisle arrangement, the exhaust aisle is enclosed and ducted back to the cooling units. Either approach stops hot exhaust air from circling back into the intakes, which lets the cooling run at a higher supply temperature and often lowers fan power.

Containment does not fix a bad room, though, and it can make one worse. If blanking panels are missing, if cable cut-outs are open, or if a tile in a contained aisle is blocked, air short-circuits through the gap and the containment hides the cause. When we are called in for a room that runs hot despite adequate nameplate cooling, airflow is the first thing we measure, not the last. A few simple practices carry most of the benefit:

• Fit blanking panels in every unused rack space, and seal floor and cable penetrations.
• Keep return paths clear so containment is not starving a unit of air.
• Check rack intake temperatures at the top, middle, and bottom of the rack, not just at the thermostat.
• Confirm fire detection and suppression have been coordinated with any containment added to the room, since enclosures can change how detection and discharge behave. Your fire protection contractor and local authority should review this.

Split systems for IT closets and small equipment rooms

The most common failure scenario we see is not in a data center at all. It is a closet. A building was designed with a small telecom room and an ordinary office air handler. Over time the closet filled with switches, a firewall, a storage array, a UPS, and a patch panel that someone added on a Friday. The room now produces several kilowatts of heat, the building cooling is shut off at night and on weekends, and the room quietly overheats every time the office is empty.

The answer is almost always a dedicated cooling source that runs all hours, independent of the building system. Ductless and commercial split systems are the usual choice: an indoor unit in or above the room, and an outdoor condenser connected by a refrigerant line set. Several details matter more than the size of the unit:

• Year-round operation. Many standard split systems are rated for cooling only down to a certain outdoor temperature. A closet needs cooling in January as well as August, so low-ambient operation has to be addressed in the selection.
• Condensate. A drain pan that overflows over a rack is a worse outcome than a warm room. Condensate pumps, safety switches, and drain routing deserve attention.
• Controls that do not hand the room back to the office schedule. The thermostat for the closet should not follow the building occupancy program.
• Alarm notification. Even a small room should have a temperature sensor that reaches a person, not just a display on the wall.
• Redundancy where it matters. Two smaller units that share the load can be a more resilient answer than one larger unit, with the second able to hold the room alone.

Our VRF and split systems service covers ductless and commercial split equipment, and our building controls team can set up schedules and alarms so a closet unit is no longer tied to an office program that turns it off at 6 p.m.

Redundancy: N, N+1, and 2N in plain language

Redundancy is simply how many extra units you have beyond the number needed to carry the load. The letter N stands for the number of units required to keep the room within its design conditions on the hottest expected day. The notation after it says how much spare capacity exists.

Three honest observations about redundancy that rarely appear in brochures. First, redundancy on paper is not redundancy in practice. If four units are installed to serve a load that needs three, but the room's hot spots are only served by two of them, then losing the wrong unit still creates an outage. Capacity has to be checked by location, not just by total tons. Second, redundant units must actually run. A standby unit that has not been started in years may fail on the day it is needed. Rotating the lead unit and testing the standby as part of routine service avoids unpleasant surprises. Third, shared dependencies defeat redundancy. Two CRACs on one power feed, one condensate drain, one glycol pump, or one control board are really one system.

For small rooms, N+1 is often the sensible target: it lets us take a unit offline for maintenance without putting the room at risk, and it gives a failure a place to land. For sites with strict commitments, such as telecom switching or public-safety dispatch, a more thorough review of dependencies, power paths, and controls is worth doing before choosing equipment.

Humidity and the ASHRAE TC 9.9 envelope

ASHRAE Technical Committee 9.9 publishes thermal guidelines for data processing environments. The committee is made up of equipment makers, designers, and operators, and its guidance is the reference most designers and auditors use. Without quoting a specific table, the structure is worth understanding.

• A recommended envelope describes the temperature and moisture range in which equipment should run reliably for its full service life. It is the range most operators aim to stay inside.
• Allowable ranges are wider and are grouped by equipment class. They describe conditions that equipment can tolerate for limited periods, and they are what a design leans on for free cooling and for failure ride-through.
• Moisture limits are expressed with dew point alongside relative humidity. Dew point is the better measure, because relative humidity changes whenever the air temperature changes.

The guidance has loosened over the years. Older rule-of-thumb targets such as a cold room held tightly at a single temperature and a narrow humidity band have given way to broader envelopes, partly because modern equipment tolerates more and partly because over-cooling wastes energy. Confirm the current edition and the class that applies to your equipment, and check the manufacturer's own requirements, since those take priority for warranty and support.

Why does humidity matter at all? Air that is too dry encourages static discharge, which can damage components. Air that is too humid risks condensation and corrosion, particularly on cold surfaces and in rooms with outside air leaks. In Southern California, the practical problems are different by location. Coastal rooms see marine moisture and salt. Inland rooms see heat, dust, and very dry air in some seasons. A room with a leaky vapor barrier or a door that opens to a loading dock can swing moisture quickly.

A humidifier inside a CRAC is a common maintenance hot spot: cylinders scale up with mineral deposits, water supply valves leak, and drain lines clog. Many operators now accept wider humidity ranges and run humidifiers less. That is a design decision worth discussing with your IT and facilities leads before changing setpoints, not something to alter during a service call.

Monitoring and alarms that reach a person

A cooling failure that is noticed in two minutes is an inconvenience. The same failure noticed in two hours is an outage. The gap between those two outcomes is nearly always monitoring, and the most common weakness we find is an alarm that exists but goes nowhere useful.

A practical monitoring setup for a server room or edge site has four layers:

1. Unit alarms. The CRAC or CRAH reports its own faults: high head pressure, low pressure, fan failure, water leak, filter, humidifier. These tell you why a unit stopped.
2. Room sensors. Temperature and humidity at rack intake height, in the cold aisle and, ideally, at the top of the rack where heat collects. These tell you what the equipment actually experiences, independent of what the cooling unit says.
3. Water and leak detection. Sensors under raised floors, around CRAC bases, and near drain pans and chilled-water piping.
4. Notification. Alarms routed to people by text, email, or an on-call system, with a defined escalation path when nobody acknowledges. A beacon in an empty room is not a notification.

Alarm thresholds deserve as much attention as the sensors. If a high-temperature alarm is set so late that equipment is already shutting down when it fires, it records the failure instead of preventing it. Set a warning level that gives a technician time to arrive, and a critical level for immediate action. Test the whole chain on a schedule: trip a sensor on purpose and confirm that the right person gets the message.

Our DaVinci Portal records readings and service history for each unit, which gives you a baseline. A discharge temperature that has crept upward over six visits is a trend a single emergency call would never reveal.

Failure response and temperature-rise timelines

People ask how long a room can stay up after the cooling stops. The honest answer is that it depends on the room, and anyone who gives you a single number without seeing the space is guessing. What we can offer is how to think about it.

When cooling stops, the equipment keeps producing heat at its full rate. The only thing slowing the temperature rise is the thermal mass of the room: the air, the racks, the structure, and the raised floor. A dense row of equipment in a small, well-sealed room heats up in minutes. A lightly loaded room with a large volume of air and a lot of surrounding mass may take considerably longer. In general terms:

• Small closets with a few kilowatts of load can climb out of the recommended range within a few minutes to a few tens of minutes after the cooling stops, particularly if the door is closed and the room is well insulated.
• Rooms with a high load per square foot are the least forgiving, and containment can shorten ride-through further, since the hot air has fewer places to go.
• Large, lightly loaded rooms give you more time, but the rise accelerates as the temperature increases, because equipment fans speed up and add their own heat.
• Equipment protects itself. Servers and switches often throttle or power down when internal temperature thresholds are reached, which can create a different kind of outage from the one you were trying to prevent.

Because the window is short, the response plan matters more than the response time of a contractor who is thirty minutes away. A sensible plan assigns each of these before the incident:

1. Who receives the alarm, and who is the backup when they do not answer.
2. Who is authorized to shed non-critical load, and which equipment that includes.
3. What temporary measures are approved: opening the door, a portable cooling unit, a fan. Each has side effects, including the risk of drawing humid or dusty air into the room.
4. How a mechanical contractor is reached, what access they need, and who escorts them.
5. What gets documented afterward, so the root cause is found and not just the symptom.

When a contractor arrives, the first question is whether the failure is isolated, such as a tripped breaker or a failed contactor, or systemic, such as a refrigerant leak, a plant problem, or a power quality issue. Quick fixes that restore cooling are worthwhile, but we also want to know why it happened. If you would like a second set of eyes on a unit that has been repaired more than once, our second-opinion diagnostic is built for that.

Maintenance windows and refrigerants

Working inside change control

Servicing a live computer room is a coordination exercise. Most facilities with serious uptime obligations operate under change-control rules: a written description of the work, an approval, a window, a rollback plan, and sometimes a freeze period around quarter-end, product launches, or holidays. A good maintenance plan for precision cooling is built around those rules instead of fighting them.

• Stagger the units. With N+1 or better, service one unit at a time and confirm the room is stable before touching the next.
• Choose the window for the cooling, not just the IT schedule. The hottest part of an afternoon is the worst time to take a unit offline; early morning is usually easier on the room.
• Walk the scope in advance. Access, escorts, badging, parking, tools, and any lock-out and tag-out steps should be settled before the day.
• Define what a stop condition looks like. If room temperature reaches an agreed point, the work pauses and the unit goes back on line.
• Close out with a record. Readings before and after, parts replaced, and observations about the room go into the service history.

Our maintenance plans are structured around scheduled visits with a defined checklist: filters, belts and fans, coil condition, refrigerant circuit readings, condensate and humidifier components, control and alarm verification, and a review of what the readings are saying. See the preventive maintenance program guide for the general approach.

Refrigerants

Refrigerant rules have been tightening for years, and precision cooling equipment is not exempt. In general terms, higher global-warming-potential refrigerants are being phased down under federal and California rules, newer equipment is moving to lower-GWP alternatives, and some of those alternatives are classified as mildly flammable, which affects handling, charge limits, and equipment design. Leak-detection, repair, and recordkeeping requirements can also apply to larger systems. Rules and dates change, so confirm current requirements for your equipment.

The practical consequences for an owner: know what refrigerant each unit uses, keep leak records, understand that an aging unit on a hard-to-source refrigerant may be a replacement candidate sooner than its mechanical condition suggests, and use a technician who holds the proper federal certification. Our refrigerant compliance guide for California covers the topic in more depth.

Power and electrical coordination

Cooling and power are two halves of the same system. Every kilowatt that a UPS delivers to a rack comes back out as heat that the cooling must remove, and a cooling system without protected power can become the weak link in an otherwise resilient room. A mechanical contractor does not design the electrical system, and we are careful to stay in our lane, but the interfaces deserve explicit attention.

• Are the cooling units on protected power? A UPS keeps servers running through a utility dip, but if the CRAC is not on a protected or generator-backed circuit, the servers will outlast the cooling. Fans may need power through a transfer; compressors often need to be delayed on restart.
• What happens on restart? After a power event, CRAC compressors and fans should restart in a controlled sequence. Simultaneous starts can trip breakers or overload a generator. Check what your units do on return of power and how long they take to recover.
• Single feed or dual feed? Redundant cooling units on one panel are not redundant. The same question applies to controls, condensate pumps, and the outdoor equipment.
• Generator and transfer-switch testing. Cooling should be included in the load-bank and transfer tests, not just IT equipment, so an actual outage does not reveal a problem for the first time.
• Heat from the electrical gear. UPS units, batteries, and power distribution equipment sit in the same thermal envelope. Battery life in particular is sensitive to temperature, so a room that runs warm can shorten the life of the UPS batteries.
• Permits and licensed trades. Electrical work belongs to licensed electricians. We coordinate with your electrical contractor on disconnects, circuit sizing, and shutdown sequences so that each trade's work is done once and done right.

Bringing mechanical and electrical teams together early is the single cheapest improvement most small facilities can make. A one-hour walk with the electrician, the IT lead, and a mechanical technician often surfaces more risk than a month of reading drawings. If your project involves a bid package, see our bid-project services for how we approach scope, sequencing, and documentation, and the commercial project delivery guide for the general process.