Unreliable grid power: design the whole backup chain
A backup generator is one part of a system. Keeping a data centre operating through a supply failure also requires stored energy, reliable switching, cooling continuity, fuel, controls and people who know what to do. Start by describing the failures the site experiences. A short voltage dip, a four-hour outage and repeated failed returns of supply need different operating responses.
Build an outage record
Request utility interruption records and compare them with measurements at the site. Record duration, frequency, voltage quality and the time of day. Include planned outages, brownouts and brief interruptions that may not appear in headline availability figures. If there are no measurements yet, make that uncertainty explicit and arrange temporary monitoring before fixing protection and ride-through requirements.
Describe the business consequence of each interruption. Some computing workloads can pause; others cannot. Separate essential IT and the cooling needed to support it from loads that may be shed. Agree who can authorise shedding and how those loads return. This allows the backup design to support a stated service requirement instead of an undefined promise that everything will always remain on.
Follow the sequence from grid loss to stable operation
On grid loss, the UPS supports its connected load while generators start and reach acceptable voltage and frequency. Transfer equipment then connects the supported distribution. Cooling controls, pumps and fans must either ride through or restart quickly enough to prevent unacceptable inlet temperatures. Controls and communications need their own dependable supplies; a powered server with a failed cooling controller is not a successful outcome.
Write the sequence as timed events. Identify what happens if a generator does not start, a breaker fails to close or the grid returns briefly and fails again. Use return delays and load sequencing appropriate to the site. Do not repeatedly transfer a live facility between unstable sources simply because voltage has reappeared for a few seconds.
Size for the load that actually transfers
Generator sizing must consider the facility load carried during an outage, not IT alone. Include essential cooling, pumps, UPS losses and the intended recharge rate. Check motor starts, step loads, harmonics, ambient temperature and altitude against the equipment data. The set's continuous or prime duty suitability matters if long outages become normal operation.
As a simple capacity exercise, a 6 MW supported load requires three 2.5 MW units before spare capacity is added. N+1 would make four installed units under that simplified assumption. Actual selection can change after derating and transient analysis. A 2N arrangement requires each side to support the required load independently, with distribution and controls arranged to preserve that independence.
Design fuel as an operating system
Calculate fuel autonomy from the selected engines' consumption curves at expected load. Allow for unusable tank volume, testing, maintenance and the reserve policy. Check delivery access during the same event that may have caused the power outage. A fuel truck route that floods with the site is not a dependable replenishment plan.
Document storage, transfer pumps, day tanks, filtration, leak detection and fuel quality checks. Identify single failures in shared pipework or controls. Space and approvals for tanks, exhausts, noise control and vehicle access can determine whether a backup arrangement fits the site at all. Resolve them before placing generators on a drawing with no workable route for fuel or exhaust.
Prove the failure modes before handover
A factory run does not demonstrate the complete site sequence. Plan integrated tests with a safe load arrangement, clear stopping criteria and agreed roles. Test grid loss, failed start, loss of a duty unit, staged load acceptance, cooling restart and controlled return to normal. Record the actual event times and electrical behaviour, not only whether an alarm light appeared.
Repeat the exercise for maintenance states. Show how a unit is isolated, how the remaining plant carries load and who controls changes to settings. Keep test records, battery checks, fuel checks and operating instructions together. A backup system's capability depends on its condition over time as well as the original drawing. Train the site team using the sequences they will actually operate.
Questions
Is a UPS a replacement for generators?
It serves a different duration and operating role. The stored-energy requirement and the outage duration determine whether a battery-only arrangement is practical.
Does N+1 remove every single point of failure?
No. Shared switchgear, controls, fuel systems and distribution paths can still fail. Review the whole chain and its maintenance states.
How many hours of fuel should we keep?
Set the autonomy requirement from outage evidence, replenishment access and the service requirement. Then calculate storage using the selected equipment and reserve policy.
Start with your four inputs: site, IT load, available power and target date. The estimator gives a browser first pass; an engineer checks the design.
Try the estimator →Send your four inputs →