Replacing UPS batteries: when is it time?

Why UPS batteries age

Most UPS units contain sealed lead-acid batteries (VRLA). They age even if the UPS never switches to battery: the constant float charge slowly corrodes the plates. The datasheet states a design life according to Eurobat, such as “3–5 years” or “10–12 years”. That value applies at 20 °C. A battery is usually considered at the end of its life when it delivers only 80 % of its rated capacity.

Temperature has the biggest influence. As a rule of thumb, life roughly halves for every 10 °C above 20 to 25 °C:

Deep discharges and frequent short transfers, for example on an unstable grid, add to this. Lithium-ion UPS units last considerably longer but need monitoring just the same.

How to tell that the battery needs replacing

The most dangerous warning sign is the one nobody sees: the UPS shows “replace battery” on its display in the server room, but nobody looks. That message belongs in an alert.

What the self-test shows and what it does not

1. Self-test — the UPS switches to battery for a few seconds. That finds defective blocks and connection faults, but not how much capacity is left. An aged battery often still passes the test but lasts only a fraction of the expected time in a real outage.
2. Runtime calibration — the UPS largely discharges the battery under real load and measures the actual runtime. Only afterwards is the displayed runtime accurate again. Calibration stresses the battery and leaves it half empty for hours. Only run it in a maintenance window with a fully charged battery, and no more than once or twice a year.

A regular self-test, for example monthly, still makes sense: it fails before the next power outage does. How runtime relates to load and age is covered in the guide Calculating UPS runtime.

Planning the battery replacement

1. Same specification — capacity, form factor and design life as stated in the UPS manual, original battery packs or approved replacement types.
2. All blocks of a string at once — a new block next to old ones gets dragged down by them.
3. Choose the time — many rack UPS units allow hot swapping. During the swap, however, the UPS does not protect the load. No swap when a storm is forecast, and only in a maintenance window for UPS units without hot swap.
4. After the swap — let the battery charge fully, enter the installation date in the network card, run a self-test and calibrate the runtime later.
5. Check the thresholds — with a new battery the runtime is accurate again. Adjust the shutdown thresholds to the measured runtime.
6. Disposal — old batteries do not belong in household waste. Resellers and manufacturers take them back.

Monitoring battery health with the UPS Hyper-V Shutdown Monitor

The UPS Hyper-V Shutdown Monitor polls the UPS regularly anyway and evaluates the battery state along the way:

1. “Replace battery” alert — when the UPS reports that the battery needs replacing or a self-test failed, an alert goes out over the configured channels, by email and SNMP trap in every edition.
2. UPS alarms on the dashboard — active alarms such as “battery bad” from the UPS alarm table, via SNMP and NUT.
3. Battery health trend — in the Reports tab, compares the discharge rate of the older and the more recent half of the period and shows the self-test history. A simple, traceable heuristic, not a prediction.
4. Scheduled self-test — manually or weekly or monthly, via SNMP or NUT. The test only starts on mains power, without bypass and with sufficient battery charge.

Which values a UPS provides depends on the vendor and the network card. The software evaluates the “replace battery” indicator for APC, for Eaton cards of the MGE series and via NUT.

UPS battery checklist

1. Installation date known — entered in the network card and noted in the maintenance plan.
2. Temperature in view — the UPS does not sit in the warmest corner of the room.
3. Alert set up — “replace battery” and “self-test failed” reach a person or the ticket system.
4. Regular self-test — about monthly, with the result evaluated.
5. Runtime measured — the thresholds match the actual runtime, not the datasheet value.
6. Replacement planned — well before the end of the design life, not after the first outage.

Frequently asked questions