
A generator that starts every Tuesday at 10 a.m., runs for half an hour with nothing connected and shuts itself down has proved one thing: it can start on a mild Tuesday morning with a technician standing next to it. It says little about whether the building stays up when the utility drops at 2 a.m. in an ice storm, with the fuel truck stuck behind a closed bridge and nobody on site who has touched the transfer switch.
Backup power is a chain: UPS, transfer switch, generator, fuel and the pumps that move it, cooling, and the people who look after all of it. A test of one link tells you about that link. Superstorm Sandy made the point in Lower Manhattan in 2012. Plenty of generators were fine; the pumps that fed them sat in basements that flooded. NYU Langone and Bellevue hospitals evacuated patients, and at one data center building, staff carried diesel up the stairwell in buckets to keep an upper-floor generator running. Our history of the disasters that shaped continuity planning covers Sandy's wider lessons; this piece is about testing every link first.
The first fifteen minutes, link by link
| Time | What should happen | What commonly goes wrong |
|---|---|---|
| 0 s | Utility fails; UPS carries the critical load; lighting, compressors and elevators drop | Weak batteries; the UPS itself faults |
| 1 to 3 s | Transfer switch confirms the loss, after a short delay to ignore flickers, and starts the generator | Controller left in "off" or "manual" after maintenance; flat starting batteries |
| Within about 10 s | Generator reaches stable voltage and frequency; the switch moves load onto it | Engine fails to start or trips; switch fails to transfer |
| 10 s to minutes | Loads step on, motors restart in sequence, the UPS recharges from generator power | Simultaneous motor starts trip the generator; the chiller is slow to restart |
| Hours | Transfer pumps refill the day tank from the main tank | Pump not on emergency power, or under water; bad fuel |
| Utility returns | Switch waits for a stable supply, retransfers, and the generator runs on to cool down | Retransfer onto an unstable supply; a second blip on non-UPS loads |
The ten seconds is not arbitrary: the National Electrical Code requires life-safety emergency systems, such as egress lighting, to have power within 10 seconds. Your IT load probably sits on an optional standby system with looser rules, but usually on the same engine and switchgear.
Transfer switches: where good generators fail
The automatic transfer switch causes a disproportionate share of failed transfers. Check five things:
- Mode after maintenance. The classic failure is human: a technician leaves the generator controller in "off," the switch in test-inhibit or an output breaker open. Make a post-maintenance walk-down mandatory: controller in AUTO, switch in normal, breakers closed, no alarms, photographed and signed.
- Sensing. Confirm the switch monitors all three utility phases. Missing the loss of one phase can leave three-phase motors running on two, which overheats them.
- Settings. Time delays and pickup points drift from design over years of service visits. Get current settings in writing and check them against design.
- Wear. Contacts pit and mechanisms stiffen in switches that rarely move. An annual infrared scan of switchgear under load finds hot connections early.
- Retransfer and redundancy. Most switches are open-transition, so returning to utility power briefly cuts the load; the UPS hides it from IT, but lighting and cooling blink again. If one switch feeds the whole critical load, a bypass-isolation switch lets you service it without dropping power.
The UPS: runtime is a measurement
Datasheet runtime assumes a load, battery age and temperature your room does not match. Runtime does not scale neatly with load either: at half load a string runs well over twice its full-load time. Plan on a figure you measured at your actual load.
The UPS has two jobs: bridging the seconds until the generator picks up and, if it never does, buying time for an orderly shutdown. Test the second job separately. Does the shutdown software still talk to the UPS after the last firmware update, and does it start early enough to finish before the batteries run flat?
Batteries are a consumable. Valve-regulated lead-acid (VRLA) batteries, still the most common type, are often sold with a design life of five years or more and replaced at three to five in practice. Heat shortens that sharply: the usual rule of thumb is that VRLA life roughly halves for every 8°C (about 15°F) of sustained operation above 25°C (77°F). Our extreme heat and grid stress playbook covers the summer side, including generator derating.
- Trend, do not wait. Track per-block voltage, internal resistance and temperature, and replace when resistance climbs.
- Replace strings, not single blocks. A string performs only as well as its weakest blocks.
- Date-stamp everything and budget replacement like a lease payment.
- Watch load creep. Every new rack shortens measured runtime, whatever the batteries' age.
Load bank or building load?
No single test covers the whole chain:
| Test | What it proves | What it misses | Risk to operations |
|---|---|---|---|
| No-load exercise | Engine starts and runs | Everything under load; light running in diesels causes wet stacking (unburned fuel and carbon in the exhaust) | Low |
| Building-load transfer, utility still available | Switch operation, real load sequencing, UPS behavior on generator | Full rated capacity, if the building load is light | Moderate |
| Load bank, resistive or resistive-reactive | Engine and alternator at known loads; clears wet stacking | The transfer switch, building sequencing, real motor inrush | Low |
| Utility-fail simulation (opening the main breaker) | The whole chain, including controls, cooling restart and people | Long-duration fuel logistics | Highest; needs a change window and a back-out plan |
| Long-duration run | Transfer pumps, day-tank refill, actual burn rate, generator room heat | Refueling during a regional event | Moderate |
Two notes on load banks. A purely resistive bank runs at unity power factor, while a real building full of motors runs lower, which works the alternator and voltage regulator harder; a resistive-reactive bank tests at rated power factor. And NFPA 110 expects diesel units whose monthly exercise falls short of about 30 percent of nameplate load (or the manufacturer's minimum exhaust temperature) to get a supplemental load-bank run once a year. Check the edition your jurisdiction has adopted, and treat it as a floor.
A reasonable program for a critical site: monthly building-load transfers, an annual load-bank test and a yearly utility-fail simulation with the whole response team on site.
Fuel: the part you buy from someone else
Take your real burn rate from a long-duration run, not the spec sheet. Divide usable tank capacity (no tank draws down to empty) by that rate to get hours, and set a reorder trigger that allows for delivery times during a regional emergency. Then ask harder questions than fuel contracts usually get:
- "Priority delivery" relative to whom? How many other customers hold the same promise, how many trucks and drivers does the supplier run, and which terminal does it load from?
- Does the supplier have power? Terminals need electricity too. After Sandy, fuel distribution in New York and New Jersey broke down badly enough that gasoline rationing followed. Use more than one supplier, ideally loading from different terminals.
- Can the truck reach the fill point? Check its location and flood elevation, street access, who can accept a delivery at 3 a.m., and whether refueling while running is permitted.
- Is the fuel still good? Diesel degrades, and water and microbial growth clog filters at the worst moment. Sample annually and polish when results call for it.
- Plan for cold. Diesel gels in severe cold. Winter Storm Uri in 2021 brought days of hard freeze deep into Texas, where few sites plan for it.
FEMA's Ready Business program publishes a power outage toolkit that covers the non-engineering side of the plan.
Cooling: the load everyone forgets
IT equipment rides through the transfer on UPS. The cooling usually does not. Compressors, and often air handlers, drop at the moment of the outage, and chillers can take several minutes to restart on generator power. In a dense room, temperature climbs quickly in that gap.
Options include putting air-handler fans and chilled-water pumps on UPS, adding thermal storage, specifying fast-restart chillers and sequencing motor starts. Confirm the generator can start the chiller with the IT load already on it. During every building-load test, log room inlet temperatures each minute until cooling is stable; that curve is the number your plan needs.
If your equipment lives in someone else's building
Questions for a colocation provider:
- When did you last transfer the whole site to generator under real load, and may we see the log?
- How many hours of fuel are on site at current load, and how many suppliers deliver it?
- Do our A and B feeds come from separate UPS systems?
- Are the cooling plant and pumps on generator and on UPS, and what temperatures did the last test record?
- Are fuel pumps and fill points above flood elevation and on emergency power?
- Will you notify us when the site transfers to generator, rather than afterward?
On your side, dual-corded equipment protects you only if either circuit, on its own, can carry the full load of everything plugged into the pair.
A test log you can copy
Record every test. The log is your evidence for examiners (our briefing on surviving a BCM audit or regulatory examination covers what they ask for) and your source for real burn rates and transfer times.
| Field | Record |
|---|---|
| Date, start and end time | |
| Test type | Exercise / building load / load bank / utility-fail / long run |
| Participants and roles | |
| Pre-checks | Controller in AUTO, switch normal, fuel level, starting battery voltage |
| How the outage was initiated | |
| Seconds to generator ready; seconds to transfer | |
| Load carried (kW, percent of nameplate, power factor) | |
| UPS time on battery; lowest runtime estimate shown | |
| Peak room inlet temperature; chiller restart time | |
| Fuel used and calculated burn rate | |
| Retransfer behavior, alarms and anomalies | |
| Post-checks | Controller back in AUTO, switch normal, breakers closed, fuel reordered |
| Corrective actions, owner, due date |
The decision rule: if the log cannot tell you how long the last transfer took, how many hours of fuel you hold at the load you actually carry, and how hot the room got before cooling recovered, the next test on the calendar is the one that will.



