The link light is on, small tests pass, and the network still falls apart every morning. A forced speed setting from years ago is one of networking's best-hidden faults — here is how it works and how to end it.
Most uplink faults are physical, but this one is self-inflicted and old. Somewhere in the switch's past, a technician locked a port to a fixed speed and duplex — often as a cure for a long-forgotten problem — and the other end was left on auto. The result is a duplex mismatch: a link that negotiates, lights up, passes casual tests, and quietly shreds real traffic under load. It is worth understanding precisely, because it masquerades as cable faults, switch faults and "the network is slow" complaints — and because the fix takes ninety seconds once both ends are actually looked at.
Auto-negotiation is a conversation. When one port is manually forced — say, locked to 1000 full duplex — it stops having that conversation and simply announces nothing. The auto side, hearing silence, falls back to a safe guess from the old rules: it picks the speed it can sense and chooses half duplex. Now one end believes both sides may talk at once, and the other believes they must take turns. Neither is "broken", the link light is on, and the stage is set for the strangest fault in networking: a link that passes ping and destroys file copies.
The forced side transmits whenever it likes. The half-duplex side, mid-transmission, sees the other's traffic and treats it as a collision — backing off, jamming, and retrying. Light traffic sneaks through in the gaps, so browsing and ping look acceptable. Sustained traffic — a backup, a big file copy, a room full of users at 9 AM — collides constantly. The counters tell the truth: late collisions on the half-duplex side and FCS/CRC errors on the full-duplex side, climbing exactly when the load does. If your uplink is fine at night and terrible in the morning, read this paragraph again.
Gigabit over copper was designed around negotiation — the two ends even use it to elect a timing master and slave. Many gigabit ports therefore cannot truly be "forced": set one to 1000/full manually and the other to auto, and you may get no link at all, a 100 Mbps link, or a mismatched one, depending on brands. There is no performance prize for forcing, either: a negotiated gigabit link and a forced one move data identically. The manual setting only ever buys you the failure modes.
Set both ends of the uplink to auto negotiation and let the conversation happen. Retest under load and confirm speed, duplex agreement, and still error counters. The standing rule for the site: ports stay on auto unless a specific old device is documented to need a fixed setting — and that exception gets a label on the cable and a line in the site notes, so it is a decision and not a fossil. If a port genuinely will not negotiate cleanly with every setting on auto, the suspicion returns to the cable's pairs, which is where this whole guide set began.
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 VideoPing packets are tiny and infrequent, so they slip through the gaps in the argument between the two ends. A file copy is a continuous stream: the half-duplex side keeps colliding with the full-duplex side's transmissions, backing off and retrying, while the full-duplex side logs the wreckage as FCS errors. Load is the trigger, which is why the fault seems to appear at busy times and vanish at night.
No. A negotiated gigabit link performs identically to a forced one — forcing buys nothing and risks a mismatch, and gigabit over copper was designed to negotiate, including electing a timing master. Leave ports on auto at both ends. The only honest exception is a specific legacy device that misbehaves on auto, and then both ends get set identically and documented.
Compare the operational speed and duplex on both ends of the uplink — a mismatch shows one side at full duplex and the other at half. Then read the error counters: late collisions climbing on the half-duplex side and FCS or CRC errors climbing on the full-duplex side during a file copy confirm it. Setting both ends back to auto and re-testing under load should freeze the counters.