4:47 PM and Then Nothing
It was a Thursday afternoon in March 2024, 4:47 PM. I was in the office with less than 24 hours until a client software deployment deadline when the lights went out.
Not a flicker. Not a quick reset. A transformer down the street had failed — overloaded, according to the utility crew — and the entire block was dark. Estimated restoration: six hours.
I run a small IT services company. We handle server deployments and infrastructure for about 20 clients. In five years, I've coordinated 300+ rush jobs — midnight crashes, last-minute migrations, calls that come in with six hours to deadline. You'd think I'd have seen enough to be ready for this.
My Unraid server hummed for a second, then went quiet as it switched over to the CyberPower UPS — a 1500VA tower unit I'd bought about five months earlier to protect the server rack. It held. The server stayed up.
I breathed.
For about 90 seconds.
Then the Screaming Started
The UPS started screaming.
Not the polite beep every 30 seconds — the "hey, maybe check your battery at some point" kind. This was high-pitched, continuous, and relentless. Loud enough that someone from the next office came in to check what was happening. The kind of noise that makes your teeth vibrate and your dog start pacing.
"High-pitched noise" sounds almost pleasant on a search result. In person, it's alarming as hell.
I searched online. The results were a mess. Some said overheating. Some said overload. Some said battery failure. A few people mentioned that CyberPower units use a "pure sinewave" alert tone during input transitions — like that was supposed to make me feel better.
I didn't want to guess. I grabbed a multimeter.
If you ever hear a high-pitched noise coming from a UPS, there are three things to check: input voltage, battery health, and load level. A multimeter knocks out two of those in about five minutes.
How to Test a Battery With a Multimeter (The Thing I Should've Done Months Ago)
I set the multimeter to DC 12V and — with the UPS under load — placed the probes on the battery terminals.
A healthy 12V sealed lead-acid battery should read around 12.6V at rest, fully charged. Under load it dips a little. But on a good battery, you shouldn't see it drop below 11V.
I read 10.8V. And it was falling. Fast.
That's a problem. 10.8V means the battery is deeply discharged — likely with a dead cell. And the screaming, I realized, was the UPS's low-battery alarm because the battery couldn't provide the current the load was demanding.
Normal runtime on that UPS should have been about 15 minutes. The battery gave me maybe two.
I hung up on the support line I was about to call. I already knew what they were going to say. What I didn't know was what to do in the next 90 seconds.
The Decision
I had two options: let the server keep running on a dying battery and risk a hard power loss — which, for an Unraid array, is one of the worst-case scenarios because parity corruption can be catastrophic — or do a clean shutdown and miss the client's delivery window.
I did the clean shutdown.
And then I called the client. The call I never want to make.
"Hey, I'm not going to hit the deadline tonight. We lost power and the UPS battery is dead — just found out. I can have the deployment ready by noon tomorrow if power comes back."
Silence. Then: "Is the data safe?"
"Yes. Clean shutdown."
"Okay. Noon."
I spent the next ten minutes sitting in the dark office, holding a multimeter, staring at a dead server, and being furious at myself.
What Actually Went Wrong
Power was back by morning. The deployment finished at 11:40 AM. But the real cost wasn't the hours of delay. It was the fact that I could've caught this problem before it caught me.
The battery was inside its normal aging window, but it was failing — and nothing warned me. More to the point: I'd never tested it. Because it "seemed fine," I assumed it was fine.
Same week, a different problem: a power surge fried the control panel on my Samsung dryer at home. I mention this because it's the same lesson in a different room. The "it should be fine" assumption.
And just so you know — if you ever search "samsung dryer control panel replacement," you'll find this isn't a rare failure. Surges kill those boards. Replacement runs $200–$350 depending on the model. That's more than a decent surge protector.
What I Did About It
After that night, my power protection routine changed:
- I now test the UPS battery every quarter. Not "when it feels like it." A multimeter test takes five minutes. The cost of missing it once was a client relationship.
- I added surge protection in front of the server rack. Cheap compared to replacing anything on the other side of it.
- I installed a low-voltage transfer switch for a portable generator — the kind you can pick up at Lowe's. The one I bought ran about $200, but it depends on the amperage you need. Now an extended outage doesn't leave me waiting for the utility company to show up.
The Lowe's transfer switch setup made sense after the dryer incident. I'd assumed a UPS was enough. It isn't. A UPS buys you minutes. A generator with a transfer switch buys you hours.
If you're using a UPS just to bridge to a clean shutdown, battery health is the single most important variable. A three-to-five-year-old battery may perform at a fraction of its rated capacity under load. It won't tell you until it decides to scream.
The Part That Stings the Most
The outage itself wasn't the painful part. Outages happen. That's why UPS systems exist.
The sting came from knowing I could have avoided it. Five minutes of routine testing. I do this for clients — asset checks, scheduled replacements, monitoring dashboards. But on my own office power system, I never once thought to test the battery.
Batteries have a calendar life. UPS units don't warn you in advance — usually they just start screaming.
What did I buy after that?
A replacement battery. A backup replacement battery. A new multimeter (the old one is still with my neighbor). A transfer switch. And a stack of sticky notes with test dates written on them.
Not the most glamorous purchase list. But probably the most useful one I've made in years.
Sarah Okonkwo
Sarah Okonkwo is an electrical test and measurement analyst specializing in multimeters, clamp meters, installation testers, sensors, and power-quality monitors. She uses IEC 61010-1 and IEC 61010-2-032 safety requirements for test and current-clamp equipment, then applies IEC 61000-4-30 methods to power-quality parameters while examining category ratings, accuracy, resolution, bandwidth, harmonics, voltage events, and uncertainty. She helps technicians and buyers select instruments, set up safe measurements, interpret readings, and diagnose faults within the intended circuit environment.