The uninterruptible power supply is the most ignored device in the server closet — a box that hums for years, asked to do nothing, until the half-second that justifies its existence: the power blinks, and the servers don’t. That’s the theory. The practice, discovered at businesses everywhere during the Northwest’s ice storms and outage seasons: the power fails, the UPS beeps once, and dies — because inside every UPS is a battery on a 3-to-5-year clock that’s been running since the install nobody remembers. A dead-battery UPS is worse than none: it’s confidence without coverage. Here’s the small maintenance discipline that keeps the box honest.

What the UPS is actually for (it’s not “running through the outage”)

Right-sizing starts with the honest mission. A UPS’s jobs, in order: ride through the blinks — the sub-second sags and one-minute flickers that are most power events, which would otherwise mean crashed servers and corrupted databases several times a year · condition the power — smoothing the surges and sags that quietly age electronics · and buy time for a graceful shutdown when the outage is real — minutes, not hours. Running the business through a long outage is a generator’s job (or the continuity plan’s); asking batteries to do it produces heroic-sounding specs and dead servers at minute forty. The corollary: what plugs into the battery side is a decision — servers, network core, the firewall and internet gear (connectivity dying with the power defeats the cellular failover), the phone system’s brain. Not the laser printer, whose startup surge can stagger a UPS all by itself, and not the space heater — both found, routinely, on real audits.

The graceful-shutdown piece almost everyone skips

Runtime without automation is a countdown to the same crash, just later: when the outage outlasts the battery, the UPS dies and takes the servers down hard anyway — unless the communication link is set up. Every business-grade UPS can signal connected servers (USB, network card, or agent) to shut down cleanly at a battery threshold — the feature that converts “20 minutes of runtime” into “the servers are always down gracefully, whatever happens.” Configuring it takes an hour, testing it takes ten minutes, and its absence is the most common finding in UPS audits: the hardware present, the handshake never made. If your servers can’t name what tells them to shut down when the power fails, that’s this week’s ticket.

The maintenance rhythm: small, calendared, honest

  • Quarterly — the self-test and a glance. Modern UPSes self-test on schedule and report battery health; someone reads the report — or the management card emails it — and a battery flagged “replace soon” gets replaced soon, not noted. Also: the audible-alarm settings verified (a UPS beeping in an empty closet is an alert without an owner — network-connected models should page a human).
  • Annually — the honest load test. On a scheduled maintenance window (change-managed like anything else): pull the wall plug and watch. Does everything hold? How many minutes does the battery actually deliver against its rated runtime? Does the graceful shutdown fire at threshold? Fifteen minutes, once a year, and the answer to “will it work?” becomes a stopwatch reading instead of a hope — the same philosophy as restore testing, applied to power.
  • Every 3–5 years — batteries, on the calendar, not on failure. Battery packs are consumables: inventory their install dates, budget replacement on the lifecycle like everything else, and swap proactively — the pack costs a fraction of the unit and a rounding error against one hard-crash recovery. Units themselves last ~7–10 years; when one goes, size the replacement against today’s load, not 2018’s.

The bottom line

A UPS is a promise with a battery-shaped expiration date. Size it for blinks and graceful shutdowns rather than heroics, wire the shutdown handshake that most installs skip, and keep the small rhythm — quarterly reports read, one annual pull-the-plug test, batteries replaced on the calendar. It’s an hour or two a year, and it’s the difference between the outage that becomes a story about the lights flickering and the one that becomes a recovery project. In this region, the test that battery will actually face is well documented: the winter ice-storm checklist is where this maintenance rhythm pays off.