Redundancy design for cleanrooms and data centers: N, N+1, 2N, and split-fan AHUs

What this covers

• What N, N+1, 2N, and 2(N+1) each survive, and how the Uptime Institute tiers relate to them.
• How much airflow one fan of a two-fan air handler delivers alone, for fans rated from 50 to 100 percent.
• Why a cleanroom can sometimes ride through on reduced airflow and a data room cannot.
• How separate power feeds, controls, and running redundant units at part speed change the design.

What changes the result

• What a single failure costs the room: product, containment, or IT equipment.
• The shape of the fan curve and the system curve, which set a surviving fan's airflow.
• Regulated minimum air change rates that reduced airflow would drop below.
• Shared power sources, controllers, and sensors that let redundant equipment fail together.

Start from what a failure costs

Redundancy is bought against a consequence, so name the consequence first. A cleanroom that loses airflow loses its classification and any product exposed at the time. A containment room that loses exhaust can release what it holds. A data room that loses cooling heats its IT equipment toward shutdown limits, faster as rack density rises. Each consequence tolerates a different amount of reduced capacity for a different length of time.

A single failure also includes planned work. A fan that has to be stopped for a bearing replacement removes the same capacity as one that fails, so a design that survives failures but not maintenance forces the room offline on a schedule instead of by surprise.

N, N+1, and 2N defined by what survives

N is the number of units that exactly meets the design load. N+1 installs one more, so any single unit can fail or be serviced and the rest still meet the load. 2N installs two complete systems, each able to carry the load alone on its own path, so a whole system can be lost. 2(N+1) makes each of those two systems N+1 in itself.

The Uptime Institute tier system describes outcomes rather than counts. Tier III is concurrently maintainable: every component and distribution path can be removed for planned work without affecting the critical load. Tier IV is fault tolerant: the facility withstands a single unplanned failure, with independent, physically separated power and cooling paths. N+1 units on one shared pipe loop or one power path can still leave a component whose maintenance stops the room, so a unit count alone does not establish a tier.

Split-fan air handlers: sizing the remaining fan

A split-fan air handler places two fans in parallel, each with its own isolation damper, so one can stop without taking the unit down. What the surviving fan delivers depends on where its own curve meets the system curve. System pressure falls with the square of airflow, so as airflow drops the resistance drops faster, and the single fan settles at a higher airflow and lower pressure than its rated point.

With a fan curve whose shutoff pressure is 1.5 times design pressure, a fan rated for 50 percent of design airflow at design pressure delivers about 71 percent of design airflow alone, at half the design pressure. Fans rated at 70 percent each deliver about 86 percent. Replace the simplified curve with the manufacturer's curve for a real selection, because a flatter or steeper fan curve moves the answer.

Two conditions make the result real. The failed fan's isolation damper has to close, or air recirculates backward through the stopped fan and the survivor pushes much of its airflow in a loop. And the surviving fan's motor has to carry the run-out point, where airflow is above its rating; National Institutes of Health guidance also asks designers to confirm that arrays operate in a stable region of the fan curve in the reduced condition.

What reduced airflow does to a cleanroom and to a data room

A cleanroom at 86 percent of design airflow still dilutes particles, just less, and recovery takes about 16 percent longer. Pressure direction can be held if return and exhaust are reduced in step. Whether that is acceptable depends on the margin in the design: a room designed at exactly the USP 797 minimum of 30 air changes per hour falls to about 26 and is out of compliance, so a regulated room at its minimum needs full N+1 airflow, not a split-fan compromise.

A data room cannot trade airflow for time in the same way. Servers draw the air their own fans demand, and if the cooling units supply 30 percent less, the shortfall is made up by hot exhaust air recirculating to server inlets. Inlet temperatures rise, server fans speed up and demand more air, and the room moves toward thermal limits. Data room cooling is sized N+1 or better at full airflow for that reason, as the data center cooling calculation shows with nine units covering an eight-unit load.

Fan arrays and N+1 inside one air handler

A fan array replaces one large fan with several small direct-drive fans in parallel. The NIH Office of Research Facilities cites its design requirements for multiple air handlers or exhaust fans to give N+1 redundancy in laboratories and critical facilities, and recommends arrays that still meet design airflow with one fan failed, a separate variable frequency drive for each fan, identical fans and motors, and an isolation damper on each fan. NIH arrays hold no more than eight fans, beyond which the bulletin reports diminishing returns in cost, redundancy, and maintenance.

Running every fan, including the redundant one, saves energy. A four-fan array sized so three fans carry design airflow can run all four at 75 percent speed. By the ideal fan laws, power falls with the cube of speed, so each fan draws 0.42 of its full-speed power and the four together draw 1.69 fan-units of power against 3 for three fans at full speed, 44 percent less. Real savings are smaller once motor and drive efficiency at part load and any fixed static pressure setpoint are counted, but the direction holds, and the standby fan is proven to work every day it runs.

Cooling plant and exhaust redundancy

Air handler redundancy is wasted if the chilled water behind it has none. Carry the same level through chillers, pumps, cooling towers or dry coolers, and the valves and pipe paths between them, and look for the single component everything shares: one header, one expansion tank, one control valve on a common bypass.

Containment exhaust gets the most conservative treatment, because exhaust is the airflow whose loss releases material. A standby exhaust fan with automatic changeover, or an array sized to carry design airflow with one fan out, keeps the room negative through a failure. The DOE handbook describes parallel filter housings with inlet and outlet isolation dampers, so one housing can stand by or be isolated for maintenance and testing while the other carries the exhaust. The containment exhaust system design page covers those housings in detail.

Separate power feeds and shared controls

Redundant fans on the same electrical panel fail together when that panel does. Where the failure being guarded against includes a power path, feed the paired fans from separate distribution, one from each of two buses or transfer switches, and put the fans that protect containment or IT equipment on standby power. NFPA 110 classifies standby systems by the seconds between power loss and restored power, and a Type 10 system restores power within 10 seconds, a gap the restart sequence must still manage.

Controls become the hidden single point of failure. Two fans with one controller, one airflow sensor, or one power supply for their damper actuators are one failure away from both stopping. Give each redundant unit its own drive, its own proof of airflow, and controls that keep running on uninterruptible power. The failure sequences page writes out the ordered restart that separate feeds make necessary, since the two fans may no longer return at the same moment.

Comparing the options without invented prices

Prices change by project and year, so compare options by what they add. With N at four units, N+1 adds 25 percent more units, 2N adds 100 percent, and 2(N+1) adds 150 percent, along with the floor space, pipe, duct, and electrical capacity each unit needs. A split-fan unit adds a second fan, drive, and damper inside one casing rather than a second air handler.

Set that against energy and consequence. Running N+1 units together at part speed can cut fan power well below running N at full speed, as the fan array arithmetic shows, and nine data room cooling units sharing an eight-unit load each run near 89 percent speed and draw about 21 percent less fan power in total. The remaining difference is priced by what one outage of the room would cost, which only the owner can state.

Questions people ask about this