Key takeaways
- The UPS is used to shut down IT equipment cleanly without corrupting data volumes.
- Actual runtime is obtained by measuring the watts drawn, not catalogue nominal ratings.
- The service life of sealed lead-acid batteries is halved for every ten-degree increase.
- The documented annual test under real load remains the only way to validate hold-up time.
Calculating useful load to ensure graceful server shutdown
Attempting to keep an IT rack operational for several hours during a mains power failure is a common design mistake. In the absence of a standby generator, the purpose of an uninterruptible power supply (UPS) is strictly to provide the time required to settle transactions, sync disks, and execute an orderly shutdown of services. A useful runtime of ten to fifteen minutes is largely sufficient to absorb mains distribution micro-outages and, if the outage persists, trigger shutdown scripts for virtual machines, databases, and network-attached storage units.
- Orderly shutdown (graceful shutdown)
- An automated sequence in which management software or a network agent sends a staggered shutdown command to servers, cleanly closing application services and unmounting file systems before power is cut.
The discrepancy between paper-calculated runtime and actual runtime often stems from confusing apparent power in volt-amperes (VA) with active power in watts (W). A server's nameplate indicates the maximum rating of its power supply unit, often calibrated at 500 W for an operating consumption that does not exceed 180 W. If nominal values are summed, the UPS is oversized, runs at low efficiency, and skews the discharge curve. Conversely, overlooking the power factor of network equipment or forgetting PoE switches supplying access points and cameras leads to premature battery collapse.
| Rack-mounted equipment | Measured average consumption (W) | Power factor (cos φ) | Useful apparent power (VA) |
|---|---|---|---|
| 2U hosting server | 220 | 0.95 | 232 |
| 4-bay NAS storage system | 65 | 0.90 | 72 |
| 24-port core switch | 45 | 0.85 | 53 |
| PoE switch (cameras + Wi-Fi access points) | 280 | 0.95 | 295 |
| Access router and gateway | 30 | 0.80 | 38 |
To size the equipment, the total active load must be increased by an operating reserve of between 20 and 30%. This margin prevents the UPS from tripping into a fault state during CPU load spikes or inrush current from a second redundant server. Communication between the UPS and the rack is established using a dedicated network management card inserted into the UPS accessory slot. The SNMP protocol or local agents installed on hypervisors receive on-battery failover alerts and initiate system shutdowns according to a defined priority: business applications first, storage next, and network equipment last.
Controlling battery ageing and scheduling load testing
The vast majority of rack-mounted UPS units use valve-regulated lead-acid batteries with gas recombination (VRLA, AGM type). These storage batteries have a theoretical service life of three to five years under ideal laboratory conditions. In a poorly ventilated technical room or inside an enclosed rack, site reality is completely different. Heat is the battery's primary enemy. An empirical rule of electrochemistry shows that every 10 °C increment above an ambient temperature of 20 °C halves cell longevity through accelerated grid corrosion and electrolyte dry-out.
The built-in self-test in modern UPS units only tests the battery for a few seconds. It detects a fully open-circuit cell or a dead short, but remains incapable of evaluating residual capacity in ampere-hours (Ah) under sustained demand. In practice, a four-year-old battery may display a 100% state of charge on the software interface, yet collapse after forty seconds as soon as mains power fails, causing servers to crash before disk journals can even close.
- Protocol preparationInform users of a maintenance window. Back up active configurations and ensure that previous-day cold backups are valid before handling the installation.
- Measuring open-circuit voltagesCheck the overall voltage of the battery string in floating regime. Record the ambient temperature of the rack at battery tray level using a contact probe.
- Disconnecting the upstream sourceOpen the supply protective device of the UPS to force failover to battery mode. Do not unplug the network patch lead in order to maintain communication with the SNMP card.
- Timing the discharge curveMonitor the voltage drop under the rack's real load for 5 to 10 minutes. Confirm that pre-alarm signals are triggered on the network management system.
- Restoration and loggingClose the UPS supply circuit breaker before the critical shutdown threshold. Note the voltage drop, hold-up time, and inspection date in the rack's technical binder.
The periodic annual test under operating load is the only reliable method for validating the standby power chain. For the telematics installer, maintaining these checks properly and adhering to trade rules fall under the duty-of-care obligations of art. 364 CO. As soon as measured runtime drops below 80% of its initial value, preventive replacement of the battery cartridges must be proposed to the client in the intervention report.
Shutdown scheduling and critical load management
An IT rack UPS is not a standby power station intended to maintain operations during a prolonged mains outage. Its sole function lies in absorbing transient micro-outages and providing a calibrated reserve of energy to initiate, sequence, and complete the safe shutdown of operating systems before storage batteries are completely depleted. Attempting to prolong operation down to the last battery percent exposes the infrastructure to an abrupt drop under residual voltage, resulting in relational database corruption, storage metadata loss, and locked virtualised volumes.
- Detection and initial time delayThe UPS network management card detects mains loss. A 120-second time delay prevents initiating disruptive shutdowns during network switching operations or upstream auto-reclosers.
- Alerting and broadcasting shutdown ordersThe software daemon or network agent broadcasts a shutdown order via SNMP or secure channels to hypervisors, physical servers, and domain controllers hosted in the rack.
- Cascading shutdown of software loadsVirtual machines and application services are suspended or shut down according to a strict hierarchy, followed by physical hypervisor hosts and then SAN or NAS storage arrays to flush write cache buffers.
- Selective power strip disconnectionThe UPS or managed power distribution unit (PDU) cuts power to servers that are already shut down to instantly offload the battery and preserve residual runtime for active network devices and home automation gateways.
| Elapsed time after outage | Target equipment | Action executed | Technical objective |
|---|---|---|---|
| T + 2 min | Hypervisors and servers | Broadcasting network shutdown signals | Validation of persistent mains fault |
| T + 4 min | Databases and VMs | Closing transactions and stopping services | Preventing disk write corruption |
| T + 8 min | SAN/NAS storage arrays | Flushing memory caches to disk | Securing file system integrity |
| T + 10 min | Physical servers | Complete hardware power-off | Reducing load on UPS batteries |
| T + 12 min | Switches and gateways | Minimal hold-up then final output cut-off | Isolating the UPS before deep discharge |
Load test procedure and battery maintenance
The theoretical runtime shown on a new unit's digital display collapses over time. Valve-regulated lead-acid (VRLA) batteries undergo permanent chemical degradation, accelerated by every degree above the reference ambient temperature. In unconditioned technical rooms or closed enclosures without forced extraction, the electrolyte dries out and the internal resistance of the cells increases. The automated internal diagnostics provided by UPS manufacturers, run without a representative load or for only a few seconds, cannot detect a block whose voltage plummets abruptly under high inrush current.
To guarantee the durability of the entrusted work, the telematics installer must formalise a periodic test protocol. According to art. 364 CO, the contractor is liable for the proper and faithful execution of their work. Under data infrastructure maintenance contracts, the operator must schedule a discharge test under real load or on a resistive load bank at least once a year. This operation measures the gap between observed residual runtime and the initial theoretical duration, enabling the battery string replacement to be scheduled before a critical failure occurs.
- Visual inspection of each cartridge: checking for mechanical deformation of plastic cases, signs of whitish sulphation on terminal posts, or acid leaks.
- Monoblock surface temperature check using an infrared thermometer to identify the onset of thermal runaway on a short-circuited cell.
- Internal resistance measurement of blocks using an impedance meter for stationary batteries prior to any discharge test.
- Switching to operating load by simulating the opening of the upstream supply during a service window agreed with IT services.
- Continuous recording of string voltage and individual block voltage until reaching the 50% depth of discharge of nominal capacity.
- Systematic logging of measured values, achieved runtime, and date in the maintenance logbook associated with the IT rack.
What to do in the event of failure: sudden shutdown, degraded batteries, and disaster recovery
When the rack shuts down abruptly as soon as the electrical mains drops out, the diagnosis is clear: the storage battery could not withstand the power surge, or the graceful shutdown management failed. In a corporate IT rack, such an outage causes unscheduled downtime of hypervisors, corruption of RAID storage volumes, and desynchronisation of KNX home automation gateways. The telematics installer must react methodically to isolate the exact root cause of the incident, protect hardware, and restore service without risking a second collapse.
- Isolating the critical loadUnplug the C13 and C14 cords supplying host servers and storage arrays before re-energising the UPS. This disconnection prevents destructive overload caused by the simultaneous inrush currents of power supply units upon mains return.
- Physical inspection of the battery packWithdraw the pure lead valve-regulated (VRLA) battery modules to check their mechanical condition. Any swelling of the plastic casing, wall bulging, electrolyte weeping, or sulphation on terminal lugs requires discarding the entire pack without attempting recharge.
- Sequential restoration and software checkRestart the network core first (distribution switches, router, IP-KNX gateway). Then power up servers one by one, and check the event logs of the SNMP network card to understand why shutdown commands were not executed in time.
To prevent these defects, documented periodic testing must be included in the commissioning and maintenance protocol. Under art. 364 CO, the contractor is liable for the proper and faithful execution of their work. When the telematics installer logs the actual discharge curve, battery age, and successful validation of shutdown scripts in the rack's maintenance logbook, they provide technical proof of the installation's proper functioning and allow the client to anticipate the triennial renewal of the battery pool.
Frequently asked questions
Why does the UPS cut out after two minutes when the data sheet stated fifteen minutes?
The data sheet calculates runtime based on a constant resistive load with brand new batteries at 20 °C. Under real operating conditions, server switch-mode power supplies cause a degraded power factor, whilst the temperature at the top of the rack often exceeds 30 °C. This thermal profile halves useful capacity. Furthermore, a two-year-old battery loses its resilience under high current inrush.
What is the technical difference between a dry-contact shutdown signal and an SNMP network card?
The dry contact provides a simple binary mains failure signal. It is suitable for triggering the shutdown of a KNX automation controller or alerting a telephone dialler. The SNMP card communicates over IP protocol with software agents installed on hypervisors. It transmits the percentage of remaining runtime, controls the graceful shutdown of virtual machines followed by physical servers, and finally commands a clean UPS power-off to preserve battery reserve.
How often should lead-acid batteries in a rack-mounted UPS be replaced?
The nominal five-year service life advertised by VRLA battery manufacturers is only valid at a stabilised 20 °C. Inside an enclosed network rack, every additional 10 °C increment halves chemical longevity. Preventive replacement is essential every three to four years at most in a climate-controlled technical room, and from two years onwards if the rack is not air-conditioned or if batteries exhibit any mechanical deformation.
How should the annual runtime test be carried out without risking data corruption?
The annual test is carried out during off-peak hours after prior validation of full IT backups. The operator does not unplug the upstream plug to maintain the equipotential earth bond. They launch the calibrated self-test from the network administration console or switch the UPS to controlled discharge via its rack input circuit breaker. They time the voltage drop under the actual production load and abort the test before reaching the servers' automatic shutdown threshold.
Texts and standards cited
- CO art. 364obligations of the contractor: liability governed by the same rules as that of the employee
