Two machines, one uptime promise
Servers don't tolerate a blink. Utility power fails for milliseconds routinely and for days occasionally, and a data room's answer is a division of labor: the UPS (uninterruptible power supply) rides through every disturbance instantly from batteries, and the standby generator takes over for anything longer than the batteries should carry. The UPS buys seconds to minutes; the generator buys hours to weeks. Neither replaces the other — the full comparison is in UPS vs generator.
The choreography during an outage: utility fails → UPS carries the load with zero interruption → the automatic transfer switch starts the generator and transfers within its rated window (10 seconds is the NFPA 110 Type 10 benchmark used for the most demanding systems) → the UPS recharges on generator power → on utility return, the chain reverses. The servers never notice.
N, N+1, 2N — redundancy in plain words
Data-center people describe backup capacity in N notation, and it's simpler than it looks. N is exactly the capacity the load needs:
- N — one generator sized for the load. If it fails to start, the batteries run out and the room goes down. Cheapest, and a single point of failure.
- N+1 — enough units for the load, plus one spare. Three 500 kW sets carrying a 1,000 kW load: any one machine can be down — failed, or out for maintenance — and the load is still covered.
- 2N — two complete, independent systems, each able to carry everything. Full duplication: any single system can fail entirely.
- 2N+1 and beyond — duplication plus spares, for facilities where downtime is measured in dollars per second.
Formal availability frameworks — the best known is the Uptime Institute's Tier Classification (I through IV) — map roughly onto this ladder: higher tiers require redundant components and, at the top, fault tolerance and concurrent maintainability, which in generator terms pushes toward N+1 and 2N designs. A small business server room rarely needs formal tiering; it needs the same logic scaled down — an honest answer to "what happens if the one generator doesn't start?"
Why data centers bias toward diesel
Walk large data centers and you'll mostly find diesel, for reasons that hold at smaller scale:
- Stored, sovereign fuel. Diesel in your tank answers to nobody. A pipeline-fed gas unit has effectively unlimited runtime but depends on gas infrastructure staying up during the same regional event that took the grid down — a dependency calculus every operator weighs. The trade is covered in natural gas vs diesel.
- Block-load acceptance. When the ATS closes, the generator takes a large step of load — UPS rectifiers plus cooling restarting. Diesel engines are strong at accepting large load steps quickly, which shortens the anxious seconds.
- Runtime is a design variable. Tank hours are chosen, not hoped for: sub-base tanks at 24/48/72 hours, bulk storage and refueling contracts beyond that. The fuel storage guide covers classes, containment and fuel quality.
Gas earns its place in smaller server rooms and edge sites where the load is modest, gas service is robust, and the no-tank, no-fuel-degradation ownership story wins — the calculus shifts with scale.
Ratings: why "standby" isn't always the right rating
Generator nameplates follow ISO 8528 rating classes, and data-center duty stresses the distinction:
- ESP (standby) — rated for emergency duty over the length of a utility outage, with annual-hour limits set by the manufacturer.
- PRP (prime) — rated for unlimited annual hours at variable load; the same machine typically carries a prime rating 15–20% below its standby number.
- COP (continuous) — constant load, unlimited hours; the most conservative rating.
A generator that exists to make outages irrelevant for a revenue-critical facility gets sized and specified more conservatively than a code-minimum life-safety set — many operators buy to the prime rating, and asking "is that kW figure ESP or PRP?" is one of the sharpest questions to put to suppliers. The full ratings story is in how long a standby generator can run.
Sizing a data room: the load is more than the servers
The IT load is the visible half. The generator also carries:
- Cooling — the big one. A data room that loses cooling cooks in minutes even with the servers powered, and compressor motors bring starting surge.
- The UPS itself — rectifier draw plus battery recharge on top of the pass-through load.
- Everything else on the backed-up panel — lighting, security, access control, the office space that shares the building.
For a small commercial server room, our sizing engine anchors the analysis around 40 kW before site adjustments — but honest sizing starts from your equipment list, not an average. Run it through the selector wizard, and pressure-test the budget with the cost estimator and the cost guide. Downtime math makes the business case: the downtime cost calculator turns your revenue-per-hour into the number the CFO needs.
The redundancy trap: underloaded engines
Here's the irony of resilient design: redundancy creates chronic underloading. Three N+1 machines sharing a load each run light; a 2N system's units run lighter still. Underloaded diesels wet-stack — accumulating unburned fuel that degrades the very reliability the redundancy bought. The fix is disciplined testing: monthly exercise at real load and annual load-bank tests for units that can't reach 30% of nameplate in normal rotation. Budget it as part of the design, not a surprise.
Frequently asked questions
What does N+1 mean for generators?
Enough generator capacity for the full load, plus one extra unit — so any single machine can fail or be serviced without losing coverage. It's the most common resilience level for facilities that are serious about uptime but can't justify full 2N duplication.
How long can a data center run on generator power?
As long as the fuel lasts: tank hours are a design choice (24/48/72-hour sub-base tanks are standard options), and refueling contracts extend runtime indefinitely — with an oil-service window on multi-day runs. Pipeline gas removes the tank limit but adds a utility dependency.
Does a small server room need a generator or is a UPS enough?
If outages at your site are rare and short and a graceful shutdown is acceptable, a UPS alone may be enough. The moment the requirement is "keep running through a multi-hour outage," batteries lose on cost and the answer is UPS + generator. See UPS vs generator.
Design the chain, not just the machine
Uptime comes from the whole chain — UPS bridge, transfer, generator, fuel, testing — designed against an honest failure model. Size your room with the wizard, check the ratings and runtime story, and post your project with your load and redundancy target so dealers propose the architecture, not just a nameplate. For code-covered systems and formal tier objectives, verify requirements with your AHJ and your engineer.