Home › Guides › Modules et connecteurs

SFP, SFP+, XFP and connectors: what an engineer needs to know

An optical link that will not come up is almost always fixed by three checks: the right module, the right fibre, the right connector. The trouble is that the vocabulary mixes three distinct things, the mechanical form factor, the electrical standard and the connector, and catalogues print them on one line. This guide separates the three, gives the typical field values worth knowing, and names the mistakes that cost a night on call.

Estimate my cost →

The form factor is not the standard, and that is the founding confusion #

An SFP is a mechanical and electrical FORM FACTOR, defined by an agreement between manufacturers, not by a standards body. Optical performance is defined by the standard, for example 10GBASE-LR published by IEEE 802.3. The practical consequence: two modules of the same form factor can be entirely incompatible, and two modules of different form factors can talk to each other perfectly if they implement the same standard. An SFP 10GBASE-LR and an XFP 10GBASE-LR interoperate without difficulty. When comparing two part numbers, compare the standard first.

The ladder of form factors, and what each carries #

SFP: one lane, 100 Mbit/s to 1 Gbit/s, LC duplex connector. The reference form factor, still deployed everywhere.
SFP+: one lane at 10 Gbit/s, same dimensions as SFP. Much of the signal conditioning moved to the host equipment, which is how the size was kept.
XFP: also one lane at 10 Gbit/s, but bulkier, and it performs clock recovery itself. It predates SFP+ and survives on transport equipment.
SFP28: one lane at 25 Gbit/s, same footprint. It also serves to break a 100 Gbit/s port into four links.
QSFP+: four lanes of 10 Gbit/s, so 40 Gbit/s. QSFP28: four lanes of 25, so 100 Gbit/s.
QSFP-DD and OSFP: eight lanes, for 400 Gbit/s and beyond. Noticeably larger and hungrier, worth checking before densifying a cabinet.

What the module name actually tells you #

The suffixes are not decorative. SR works at 850 nanometres over multimode fibre and covers a few hundred metres depending on fibre grade. LR works at 1310 nanometres over single-mode fibre for about ten kilometres. ER aims at around forty kilometres and ZR at roughly eighty, the latter being market usage rather than a published standard. The classic trap is not the bit rate, it is the fibre: a single-mode module on a multimode patch cord gives a link that sometimes comes up over a few metres and collapses as soon as the run gets longer. Check the fibre before suspecting the module.

The receive window has two edges, not one #

Everyone knows the sensitivity threshold, below which the receiver can no longer tell anything apart. The second edge is far less present in people's minds: above a certain power the receiver saturates and the link turns erratic while everything looks generously sized. That is exactly what happens when eighty-kilometre optics are used to join two rooms two kilometres apart. The cure is not another module, it is a calibrated attenuator. Both thresholds are printed on the datasheet.

Between the sensitivity threshold and the saturation threshold. Both edges matter.

The budget calculation, in one line #

Minimum launch power minus sensitivity threshold equals available budget. Against it, the sum of losses: fibre attenuation, typically around 0.35 dB per kilometre near 1310 nanometres and 0.20 to 0.25 dB per kilometre near 1550; each mated connector pair, often 0.2 to 0.5 dB in practice; each splice, around 0.05 to 0.1 dB; plus an ageing margin you always keep, two to three decibels. These are usual orders of magnitude, not a guarantee: the datasheets for your fibre and your modules are what count. If total loss exceeds the budget, the link will not come up, and no configuration recovers physics.

Digital monitoring, the tool that settles most arguments #

Modern modules publish their internal readings: temperature, supply voltage, laser bias current, transmitted power and above all received power. That last figure, compared with the datasheet sensitivity threshold, settles in thirty seconds a discussion that would otherwise take a day. Three cases orient the diagnosis. Received power is absent or collapsed: the fibre is cut, badly mated or dirty. It is low but present: the budget is too tight, look for a poor splice or a contaminated connector. It is healthy but the port counts errors: look at saturation or a standard mismatch.

Connectors, and the one detail that matters when ordering #

LC: small ferrule, the standard for pluggable modules, almost always duplex. This is what you meet in ninety per cent of cases.
SC: a ferrule twice the size, push-pull latching. Found on distribution frames, on carrier handover sockets and on older equipment.
MPO, sometimes called MTP: twelve or sixteen fibres in one connector, for parallel optics where lanes travel on separate fibres. Two specific traps, cable polarity, which exists in several incompatible conventions, and the presence or absence of alignment pins, two pinned connectors never mating together.
FC and ST: the older ones, screwed for the first, bayonet for the second. Still around in labs and on legacy installations.
The detail that matters: a four-lane optic does not necessarily mean MPO. Variants that multiplex their four colours onto a single fibre pair use ordinary LC.

PC, UPC, APC: why some connectors are cut at an angle #

A fibre end reflects part of the light back towards the transmitter. Polishing limits that return. A conventional domed face already returns far less than a flat one, an ultra polished face less again, and a face angled at eight degrees deflects the echo out of the core, which is by far the best result. That is why fibre access networks, more sensitive to those reflections, use the angle. The absolute rule fits in one sentence: an angled face never mates with a straight one. Contact happens on an edge, loss is catastrophic and both ferrules are ruined. The colour code is your safeguard, green marks the angle, blue marks straight polishing on single-mode.

The light returned depending on the end face. To the right of the line, the fibre opposite.

The mistakes that cost a night on call #

Mating a green face onto a blue one. Major loss, and both connectors are fit for the bin.
Confusing the green of the angle with the light green of a recent multimode patch cord. Read the cable reference, not just its colour.
Fitting long-reach optics on a short link with no attenuator, then hunting the fault on the switch side.
Plugging in without cleaning. One speck of dust on a single-mode core is enough to ruin the budget, and it transfers to the opposite face on first contact.
Forgetting that some equipment refuses modules whose internal identifier does not match the equipment vendor. It does not show on the optical datasheet, it shows in the switch log, and it is checked BEFORE ordering.
Mixing two polarity conventions on MPO links. Links come up in improbable pairs, and diagnosis takes hours if you do not think of it.

What to remember #

The form factor tells you what fits the chassis, the standard tells you whether the link can work, the budget tells you how far, and digital monitoring tells you what is actually happening. An engineer who checks those four points in that order eliminates the great majority of optical faults without getting in the car. The rest is a matter of cleanliness, and that is not a joke: dust is the leading cause of failure on a properly designed optical plant.

FAQ #

Will a module from one vendor work in another vendor's equipment?

Optically yes, provided both ends implement the same standard. Any blocking happens elsewhere: some equipment checks an identifier written in the module memory and refuses those it does not recognise. Check the equipment documentation before ordering, not after.

Can I put an SFP in an SFP+ port?

Mechanically yes, and many devices accept the module at the lower rate. It is not universal, however, as some ports do not negotiate down. The chassis documentation is the only reliable answer.

How do I know whether my fibre is single-mode or multimode?

The cable jacket carries a standard reference, and the colour code gives an indication. If in doubt, the core diameters differ substantially and a meter tells them apart immediately. Do not guess: a single-mode module on multimode fibre gives an unstable link that is hard to diagnose.

What is the difference between UPC and APC in one sentence?

Both limit the light returned to the transmitter, but the face angled at eight degrees does it far better by throwing the echo out of the core. That is why it is used on access networks. It never mates with a straight face.

Do I need an attenuator on a short link?

If you are using optics designed for tens of kilometres over a few hundred metres, very probably. Compare the received power reported by the module with the saturation threshold on its datasheet: that is the check that settles it.

Written on 6 September 2026.