It was 6:47 p.m. on a Tuesday in March 2024. I was done for the day. Keys in hand, laptop bag over my shoulder. Then the phone rang.
“We've got a network issue.”
Five words. That's all it takes to turn an evening into an emergency.
The client had a product launch the following Thursday. A Cisco switch in their production environment was intermittently dropping connections. They had already swapped in a spare. Same problem. So the switch wasn't the problem. That sounds obvious after the fact, but when you're under pressure, the easy target is the big, expensive box. The real culprit is often something much smaller and less glamorous.
In this case, it was a relay from Kemet Electronics Corp.
Kemet Relays: The Invisible Workhorse
Kemet Electronics Corp makes components you rarely think about until they stop working. Ceramic capacitors, tantalum capacitors, relays, connectors, MLCCs. They're not the kind of parts that get a keynote slot at a product launch. But they're the reason the product launch can happen at all.
Kemet relays, in particular, are all over industrial and networking equipment. They switch signals, isolate circuits, and do it thousands of times without complaint. If you've ever had a relay fail at the wrong moment, you know that sinking feeling. It's not a dramatic explosion. It's a subtle “wait, why did that stop?”
The “Switches vs Cisco” Red Herring
Look, I'm not going to settle the “switches vs Cisco” argument today. That's a conversation about port densities, licensing models, and whether you need modular uplinks. All valid stuff. But when a network misbehaves, the “switches vs Cisco” comparison won't help you find a bad relay.
The switch was a Cisco, but the fault was not a Cisco switch problem. The fault was a small Kemet relay on the power supply board. It's tempting to think that if the switch is from a reputable brand, all the components inside it must be bulletproof. That's a nice thought. It's also a simplification. Components are rated for a certain number of operations, temperatures, and current loads. Sometimes one of them dies early. That's not a brand failure. It's a physics failure.
Getting Practical with a Fluke 117 Multimeter
When I'm triaging a rush order or a field failure, I start with the cheapest suspicion. I don't immediately blame the CPU, the memory, or the firmware. I look for the part that costs less than a pizza.
I opened the switch's power supply board and found a row of relays. The one I needed to check was labelled with a Kemet part number ending in 3210. I want to say the full part number was something like a Kemet 3210, but I'd have to dig up the photo to give you the exact suffix. Don't quote me on that one.
I set my Fluke 117 multimeter to resistance mode and measured the coil pins. The datasheet ballpark for that relay was around 400 ohms, give or take. I want to say 400, but it might have been 350. The Fluke 117 read OL. Open loop. What I mean is the coil had no continuity. No energizing possible. The relay was dead.
If you've ever used a Fluke 117, you know why it's the meter I grab first. It's not the fanciest, but it's reliable, it's easy to read under bad lighting, and it's been with me through a hundred field diagnostics. Trust me on this one: a good multimeter pays for itself the first time it saves you from replacing a $5,000 switch for a $5 part.
The Replacement Problem
The dead relay was easy to find. Getting a replacement was not.
The client had a “compatible” relay in their drawer. It had the same coil voltage and same pinout. It was about $3 cheaper than the genuine Kemet part. It was also the kind of part that makes me nervous.
Here's the thing: relays have contact ratings, insulation ratings, response times, and surge behaviors. Two relays can look identical and behave differently when the circuit is under load. The “just use an equivalent” advice ignores the nuance. It's tempting to think you can swap any relay with the same footprint and voltage—but you can't. Not always.
Per FTC guidelines (ftc.gov), advertising claims need to be truthful and substantiated. When someone says “equivalent to Kemet,” I want to see a datasheet. If the vendor can't show me one, that's a red flag. In an emergency, I don't need a “probably.” I need a “guaranteed.”
Paying for Certainty
We didn't have a Kemet 3210 relay in the van. The nearest authorized distributor had one in stock, but it would require an overnight freight charge of $96. The client looked at me like I had just quoted them a small ransom.
Let's put that $96 in context. The product launch was worth tens of thousands in booked orders. Missing the Thursday deadline would have triggered a penalty clause. The client's alternative was to install the non-genuine relay and hope. Hope is not a logistic strategy.
We paid the $96. The distributor said the relay would arrive by 10:30 a.m. It showed up at 10:14. I installed it, cycled the switch, and the network stopped dropping. The launch went off without a single network complaint. That's the last time I heard about that client's old switch, which is exactly the way I like it.
What I'd Do Differently
Not perfect. I wish I had double-checked the relay's part number when I first arrived instead of writing it down as “3210” and then second-guessing myself. I also wish the client hadn't waited until 6:47 p.m. to call. But emergency work is like that. You work with what you have.
Since that night, I've added a photo of the Kemet 3210 relay to my phone's field notes. It sounds small, but when you're in a hurry, a photo is worth a thousand part-number searches.
Bottom line: in a crisis, the premium you pay for guaranteed delivery isn't for speed. It's for certainty. A cheap relay that costs $3 less can cost you a $30,000 event. The “probably on time” promise is the biggest risk in any urgent order. I'd rather pay for a “definitely” every time.
And for the record, the “switches vs Cisco” argument is fine for planning meetings. But out in the field, it doesn't matter what name is on the front of the box. What matters is what's on the board inside—and whether your multimeter is in your bag.
