An eight-port switch in the rack had started doing the thing where it works for a few hours and then quietly drops half its ports until you power-cycle it. The kind of fault that is just intermittent enough to make you doubt the cables, the patch panel, and your own memory before you accept it is the box itself. I had a spare, so the easy answer was to bin it. But it was a good switch, fanless and silent, and the failure pattern smelled like power rather than logic. So onto the bench it went.
Opening it up, the cause was almost insultingly obvious once I had the board under decent light. One electrolytic capacitor near the power input had a domed top, the classic bulge that means it has had enough. Capacitor plague has been killing cheap boards for years, and a tired cap on a power rail will sag under load, which is exactly the sort of thing that makes a device fall over only when it is warm and busy.
The repair is not complicated, but it rewards patience. The notes I keep telling myself, mostly because I keep forgetting them:
- Read the value off the dead part before you remove it. This one was 1000µF at 16V, low-ESR, and you want a like-for-like replacement, not just any cap with the right number on it.
- Match the polarity. Electrolytics care, and they express their disappointment loudly.
- Flux is not optional. A flux pen on the old joints made the difference between fighting the solder and watching it just let go.
I used hot air to lift the old cap rather than wrestling a single iron against a board with a big ground plane that wicks heat away faster than a 25W iron can supply it. A bit of preheat, gentle persuasion with tweezers, and the old part came free without lifting a pad. Cleaned the holes with desoldering braid, dropped the new cap in respecting the polarity stripe, and reflowed both legs with fresh solder.
old: 1000µF 16V, top domed, ESR high
new: 1000µF 16V low-ESR, 105°C rated
The 105°C rating matters more than people expect. A cap sitting next to a warm regulator lives a hard life, and the cheap 85°C parts that fail first are usually why the board died in the first place. Spending a few pence more on a better-rated replacement is the cheapest reliability upgrade you will ever make.
Powered up on the bench, left running with a laptop hammering traffic through it for an afternoon, and the ports stayed up. It went back in the rack the same evening and has behaved since. Total cost, somewhere under a pound for the capacitor and an hour I would otherwise have spent feeling vaguely guilty about throwing away a working chassis over one tired component.
There is a particular satisfaction in fixing a thing rather than replacing it, and it is not only sentiment. A device you have had open and understood is a device you trust differently. I know exactly what failed in that switch and exactly what I did about it, which is more than I can say for the shiny replacement I nearly bought.