It makes a failure on one side of the converter visible on the other - without it, a broken fibre can leave both switches believing the link is up.

A basic converter is two independent interfaces glued together. The copper side sees the switch, the fibre side sees the far converter, and neither knows about the other.

If the fibre breaks, the local switch still sees a copper link to the converter and keeps sending traffic into a path that goes nowhere. Nothing indicates a fault except that traffic disappears - and where the link is one of two redundant paths, spanning tree does not fail over because it never sees the link go down. That is far worse than a clean failure.

Link fault pass-through propagates the failure: when the fibre drops, the converter drops the copper link too, so the switch sees the port go down and reacts normally.

It should be regarded as a required feature rather than an option on any link that carries production traffic or forms part of a redundant topology.

Multimode for links inside a building or a small campus; single-mode for anything longer, and increasingly as the default for new installations.

Multimode has a larger core, uses cheaper optics, and suits distances up to a few hundred metres depending on grade and speed. It is the traditional choice for building backbones.

Single-mode has a much smaller core and supports far greater distances - kilometres routinely, and much more with appropriate optics. Optics were historically more expensive; the difference has narrowed considerably.

Because the fibre itself is a small part of an installation's cost while the labour of installing it is large, single-mode is increasingly specified for new work even at short distances - it does not need replacing when speeds increase or distances extend.

What matters practically is that the converters at both ends match each other and the fibre. Mixing modes does not work, and a multimode optic on single-mode fibre is a recurring cause of a link that will not establish.

Managed on anything that matters - the ability to see optical levels and to be alerted on failure repays the difference quickly.

An unmanaged converter reports nothing. When it fails, the symptom is a dead link and the diagnosis depends on somebody remembering the converter exists and being able to reach it.

A managed converter has an address and reports link state, optical transmit and receive levels, and errors. It appears in the monitoring system, and its optical level reading is genuinely valuable: a gradual decline indicates a connector, splice or bend degrading, which can be fixed before the link fails.

Managed units also usually implement link fault pass-through properly and allow loopback testing, which speeds fault isolation on a long run.

For a temporary link or a low-value connection, unmanaged is reasonable.

Where several conversions occur in one place, a managed chassis with hot-swappable modules and redundant power is the better arrangement than several standalone units.

Not on its own - it converts media rather than extending copper. The extension comes from the fibre section.

The 100 metre limit applies to each copper segment. A converter at each end with fibre between them gives 100 metres of copper, a long fibre run, and another 100 metres of copper - which is how a device 800 metres away is reached.

What does not work is putting a converter in the middle of a copper run to double it, because there is no fibre section.

For genuine copper extension over existing cable, the correct product is an Ethernet extender, which uses different signalling to trade speed for distance.

The fibre approach is preferable wherever fibre can be installed: full speed, far greater distance, and no electrical path. The extender is for cases where existing copper is all there is and installing fibre is not possible.

Because it carries no electrical path - which removes earth potential differences and surge, both of which damage equipment.

Two buildings have separate earthing systems, and a genuine voltage difference can exist between them. A copper cable joining them creates a path for current to flow, which can damage equipment at both ends and is a safety issue.

The more dramatic case is lightning. A copper run between structures is an excellent collector of induced energy from a nearby strike, and the damage typically appears at the switch ports rather than at the cable.

Fibre carries light. There is no conductor and no path, so both problems disappear entirely - provided the fibre is genuinely all-dielectric, since armoured fibre with a metallic strength member reintroduces one unless it is bonded correctly.

This is why cabling standards and practice call for fibre between buildings regardless of distance, and it is the strongest argument for a converter pair even on a link of thirty metres.

Dirty or damaged connectors, marginal optical levels, mismatched settings and failing power supplies - in roughly that order.

Contamination on a fibre connector is the leading cause of optical problems by a wide margin. A speck of dust on a fibre end face attenuates significantly and can cause a link that works, then does not, then does again. Inspect and clean rather than assuming.

Marginal optical levels leave no margin for temperature change or a slightly loose connector. A managed converter's level reading shows this; without one it is invisible.

Setting mismatches - speed, duplex, or fibre mode - produce links that establish and pass traffic poorly.

Power supplies fail, and a marginal one produces intermittent behaviour rather than a clean failure, which is much harder to diagnose. Substituting a known-good supply is a quick test worth doing early.

Also check the fibre patch leads. They are handled, bent and trapped in cabinet doors more often than anything else in the path.

Ideally yes. Different models generally interoperate but matched pairs avoid the settings and behaviour differences that cause obscure faults.

The optical interface is standardised, so converters from different manufacturers usually establish a link provided speed, mode and wavelength match.

What differs is behaviour. Link fault pass-through implementations vary and do not always interwork. Auto-negotiation handling on the copper side differs. Some units require a switch setting for pass-through that the other end does not.

The result is a link that works but fails to signal faults correctly, which is precisely the failure mode that pass-through exists to prevent.

Matched pairs also simplify sparing: one spare model covers both ends.

Where single-fibre bidirectional converters are used, the pair is mandatory and complementary - the two ends transmit on different wavelengths, so they are explicitly sold as A and B units and are not interchangeable.