The switches are fine, the plugs are fine — and the uplink still drops. On long runs the fault lives in the middle: metres nobody measured, noise nobody routed around, and damage nobody can see from either end.
Desk cables forgive almost everything; building runs forgive nothing. A switch-to-switch link that climbs floors collects every weakness at once — length against the standard's limit, electrical noise from the building's machinery, and physical damage hidden in risers and ceiling voids. It is no accident that the video's story is set in a 12-storey building, with an uplink per floor and a mouse as the surprise culprit. This guide gives the long run its own diagnosis: the distance budget, interference discipline, mid-run damage testing, and the point where copper should honestly retire in favour of fibre.
Ethernet over twisted pair is specified for a 100 metre channel: up to 90 metres of fixed horizontal cable plus patch cords at both ends. In buildings, runs get estimated by eye — "ground to fifth floor, it must be sixty metres" — while the actual route zigzags through risers, across ceilings and around lift shafts, quietly doubling the guess. A run over budget often still links — at 100 Mbps, or at gigabit with errors — which is worse than failing outright, because it invites weeks of blaming the switches. If a long uplink misbehaves, measure or re-route before anything else; a cable certifier's length reading settles it, and so does testing the same switches together on a short cord in one room.
A long cable collects the building's electrical weather. The honest rules: do not run network cable lashed to power cables, keep separation at crossings (cross power at right angles where you must), and be suspicious of routes past lift motors, big AC units and old fluorescent ballasts. Quality cable and correct pairing are the first defence — which is why a split-pair fault that a desk cable survives can be fatal on a riser run: the untwisted signal has no noise cancellation left exactly where the noise is worst. If link errors rise when the lift runs or the AC kicks in, you have found your correlation.
The video's example deserves its fame: in a 12-storey building, a mouse cut the uplink cable mid-route, and one damaged pair turned a building's network into a detective story. Mid-run damage has a signature — the fault appeared suddenly, nothing at either end changed, and it often takes out one or two pairs rather than all eight (a bite, a staple through the jacket, a cable crushed under a ceiling grid or a door). The test that isolates it: prove both switches and both patch cords in one room, then reconnect sections of the fixed run one at a time, or substitute a spare run if the building has one. Buildings wired with a spare cable in the riser — pulled during installation, when it was cheap — forgive their owners for decades.
Diagnosis discipline for long runs mirrors the fault's geography. Confirm the switches talk on a short factory cord. Then test the fixed run alone with a wire-map tester at both ends — checking continuity, pairing, and length if the tester offers it. Then reintroduce the patch cords. Each stage that passes removes a section from suspicion. What does not work: replacing both ends' plugs repeatedly while the damage sits untouched in a ceiling void on the seventh floor.
Some links should stop being copper. Between separate buildings, copper joins two different earthing systems, and voltage differences plus lightning strikes ride the cable into your switch ports — a run that kills a port every monsoon is making this argument for you. For long or noisy routes inside a building, fibre also ends the 100 metre negotiation entirely. A fibre uplink with two media converters or SFP switches costs more than a cable and less than repeated storm damage and downtime. It needs a proper installation — bend radius respected, connectors kept clean — but once in, inter-building links simply stop being a recurring character in your fault log.
See the live troubleshooting in the video — the same cable tested with a laptop, then between two switches, and the wiring fault found on camera.
▶️ Watch Full VideoThe Ethernet channel limit is 100 metres end to end — up to 90 metres of fixed cable plus patch cords at both ends. Beyond that, links may still appear but with falling speed and rising errors. In buildings the real routed length is usually far more than the straight-line guess, so measure the run or test the switches together on a short cord before blaming hardware.
Yes, on long runs especially. A twisted pair rejects interference only while its pairs are correctly twisted — and long runs pass lift shafts, AC plant and power cabling that radiate noise. The symptoms are error counters climbing in step with building machinery. Keep network runs separated from power, cross at right angles where unavoidable, and remember a split-pair cable on a noisy riser has no defence left.
Three situations argue for it: links between separate buildings, where copper also imports earth differences and lightning surges that kill switch ports; routes beyond the 100 metre copper limit; and runs through heavy electrical noise that keep failing copper tests. Fibre costs more to install and then removes the whole class of faults — no surge path, no interference, no distance argument.