Before you read. Two fibre patch leads, both yellow, both with the same connectors, both apparently identical. Between two switches in the same room either of them works.
Move one of those switches to a building 300 metres away and one lead would carry the traffic and the other would not, and no amount of testing in the room would have told you which.
What is different inside the glass, and what would have told you?
Copper had one variable that mattered and a category number that summarised it. Fibre has four things that have to agree, they are chosen independently, and getting one of them wrong produces a link that comes up. That last part is what makes this topic worth more than its share of the exam.
Some words you will need
- core
- The centre of the fibre, where the light travels. Its diameter is the whole distinction below.
- cladding
- The glass around the core, with a lower refractive index, which keeps light in the core.
- mode
- One possible path light can take down the fibre. A wider core allows more of them.
- dispersion
- A pulse of light spreading out as it travels, until it overlaps the pulse behind it.
- wavelength
- The colour of the light, measured in nanometres. The transceiver picks it and the fibre is designed for it.
- transceiver
- The removable module that converts between electrical signals and light. Transmitter and receiver in one.
What breaks without this
A link comes up and the circuit is still wrong. Copper mostly fails honestly. A fibre mismatch frequently gives you a link light and an interface that says up, with errors accumulating underneath, which is the hardest kind of fault to be handed.
You order the wrong part and lose a week. A transceiver is specific to a fibre type, a wavelength, a distance and a speed. Ordering by the shape of the socket gets you something that fits perfectly and does not work.
Somebody cleans a connector with their shirt. Contamination is the most common fibre fault in the field by a wide margin, and the instinct to wipe something on your sleeve makes it permanently worse.
Two kinds of glass
The difference between the two fibre types is the diameter of the core, and everything else follows from it.
Multimode has a wide core, either 50 or 62.5 micrometres. Wide enough that light entering at slightly different angles takes visibly different paths down the fibre, bouncing off the cladding boundary at different rates. Each of those paths is a mode, hence the name.
Single mode has a core of roughly 9 micrometres. That is narrow enough that only one path is possible, so the light travels straight down the middle.
The consequence is what matters. In multimode, light that took a longer bouncing path arrives later than light that went more directly, so a pulse that left as a sharp spike arrives smeared out in time. That is modal dispersion, and it puts a ceiling on distance: send pulses faster and they start to overlap the smeared tail of the one before, until the receiver cannot tell them apart. Higher speed therefore means shorter distance, on the same fibre.
Single mode has no modal dispersion because there is only one mode. It goes tens of kilometres, and in long haul use considerably further.
So why does multimode exist at all? Because a wide core is far more forgiving. Alignment tolerances are looser, connectors are cheaper, and the light sources that work with it are cheaper than the ones single mode needs. Inside a building, where nothing is more than a few hundred metres from anything else, multimode is the cheaper answer to a problem single mode would over-solve.
If you already work on networks: the wavelengths, and the second kind of dispersion
Modal dispersion is the one that explains the core diameters. There is another, and it explains the wavelengths.
Multimode systems run at 850 nanometres, and sometimes at 1300. Single mode runs at 1310 and 1550. Those are not arbitrary and they are not interchangeable, and the reasoning is different in each case.
The multimode choice is economic. At 850 nm you can use a vertical cavity surface emitting laser, which is cheap to make and cheap to drive, and the fibre is designed to perform well there. Most of the cost advantage of multimode is actually the cost advantage of that light source.
The single mode choice comes from the glass. ITU-T G.652, which is the recommendation that defines ordinary single mode fibre, describes fibre designed around a zero-dispersion wavelength near 1310 nm and usable in both the 1310 and 1550 regions. Zero dispersion here means chromatic dispersion, which is the second kind: a light source does not emit one perfect colour but a narrow band of them, and different colours travel at slightly different speeds through glass, so the pulse spreads. Near 1310 nm those effects largely cancel in standard fibre.
At 1550 nm chromatic dispersion is worse, and 1550 is used anyway, because attenuation is lower there than anywhere else. That is the trade for very long distances: accept dispersion you can compensate for, and lose less signal per kilometre.
Two practical consequences. The wavelength is a property of the transceiver, not of the fibre, so a fibre run does not have a wavelength until you plug something into it, and two transceivers at opposite ends have to have chosen the same one. And this is why single mode transceivers come in a confusing spread of variants with different reaches: they are different lasers at different wavelengths with different power budgets, all fitting the same socket.
ITU-T also publishes G.657 for bending-loss insensitive single mode, which is the same fibre made tolerant of tight bends for use inside buildings, where somebody is eventually going to route it round a corner it was not designed for.
Naming the glass
The category system for fibre is two letters and a number, and it splits the two types cleanly.
OM numbers are multimode. They run OM1 through OM5 and each one is a bandwidth grade rather than a distance, in the same way a copper category was a bandwidth rather than a speed. Higher numbers carry a given speed further.
OS numbers are single mode, OS1 and OS2, and they differ by attenuation and by how the cable is constructed rather than by anything about the core.
| Grade | Type | Core | Typically |
|---|---|---|---|
| OM1 | Multimode | 62.5 µm | Legacy. Orange jacket |
| OM2 | Multimode | 50 µm | Legacy. Orange jacket |
| OM3 | Multimode | 50 µm | 10G to 300 m. Aqua jacket |
| OM4 | Multimode | 50 µm | 10G further than OM3. Aqua or violet |
| OM5 | Multimode | 50 µm | Adds short wavelength multiplexing. Lime green |
| OS1 | Single mode | 9 µm | Indoor, tight buffered. Yellow |
| OS2 | Single mode | 9 µm | Outdoor and long haul, loose tube. Yellow |
Do not mix OM1 with anything. The 62.5 µm core against 50 µm is a physical mismatch, and joining them loses a substantial fraction of the light in one direction. Buildings cabled in the 1990s are full of OM1, and connecting a new OM3 run to an existing OM1 backbone is a fault that measures as unexplained loss.
If you already work on networks: the jacket colours are a convention, and they have drifted
The colour table above is how it is taught and how it usually is, and it is worth knowing exactly how much weight it will bear.
Aqua for OM3 and OM4 is close to universal, and it is genuinely useful: an aqua jacket in a rack is almost certainly multimode rated for 10G. Yellow for single mode is similarly reliable. Those two are safe reflexes.
The rest has drifted. OM4 appears in both aqua and violet depending on the manufacturer and the year, and violet was introduced precisely because aqua could not distinguish OM3 from OM4. Lime green for OM5 is the intended convention and OM5 is uncommon enough that you may never see one. Orange covers both OM1 and OM2, which is exactly the pair you most need to tell apart, because one is 62.5 µm and the other is 50.
And none of it is guaranteed. Jacket colour is a manufacturing convention rather than a requirement, and a cable made to a customer specification can be any colour at all. Data centres with a colour scheme for purpose rather than for fibre type are common, and in one of those the colour tells you which system a link belongs to and nothing about the glass.
So the reliable answer is printed on the jacket alongside everything else, in the same small type as the copper markings from the previous topic. It gives the type, the core and cladding diameters as a pair such as 50/125, and the OM or OS grade. Reading it takes ten seconds and it is the difference between knowing and assuming, which on an OM1 backbone is a whole afternoon.
The connectors
Four connector types come up, and they are best remembered by how they attach rather than by their initials.
| Connector | How it attaches | Where you meet it |
|---|---|---|
| LC | Small latch, like a phone plug. Usually a duplex pair | Almost everything current. The default |
| SC | Square, push and pull | Older equipment, some providers, patch panels |
| ST | Bayonet, twist to lock, round | Legacy multimode installations |
| MPO | Multi-fibre ribbon, one connector carrying 12 or more | 40G and 100G, and structured trunks |
The initials do expand, and the exam’s list of connector types prints them in full: SC is subscriber connector, LC is local connector, ST is straight tip, MPO is multi-fibre push on. Only one of the four describes the object. A straight tip really is a straight ferrule with a bayonet collar round it, and the other three names tell you nothing you can use while standing in a comms room holding one, which is why the table is about how they attach.
LC is what you will handle most. It is small, which is the point: two of them fit where one SC would, so a switch port takes a duplex LC and gets transmit and receive in the space a single older connector needed.
MPO is the odd one because it is not a single link. It carries a ribbon of fibres in one body, and speeds such as 40G were originally built by running four lanes in parallel over four pairs of fibres inside one MPO. That makes polarity a real concern: the fibres have to arrive in an order the far end expects, and MPO trunks come in several polarity types that are not interchangeable.
Two more things on the end face. Connectors are polished either flat, called PC or UPC, or at a slight angle, called APC. Angled connectors are green and flat ones are blue, and the angle exists to send reflected light into the cladding rather than straight back down the fibre. Never mate an angled connector to a flat one. They do not sit together correctly, the loss is high, and on some combinations you can damage the end faces.
If you already work on networks: contamination, which is the fault you will actually meet
Ask anyone who works with fibre what breaks and they will say dirt, and they will say it before you have finished the question.
The scale explains it. A single mode core is 9 micrometres across. A typical dust particle is comparable in size, and a fingerprint is enormous. One particle sitting on the end face can block a meaningful fraction of the light, and because the two end faces are pressed together under spring pressure, it does not simply sit there: it gets ground into the glass, which turns a cleanable problem into a scratched ferrule and a connector that has to be replaced.
Three habits follow from that, and they are the whole practice.
Caps stay on until the moment of connection, on both the connector and the port. The dust cap is not packaging.
Clean before every connection, with a proper cleaner: a cassette tool, or lint-free wipes with the right solvent. Not a shirt, not a tissue, not compressed air from a can, all three of which either deposit more contamination or drive it further in.
And inspect if you can. A fibre inspection scope shows the end face magnified, and once you have seen a contaminated one next to a clean one you stop treating this as fussiness. It is the reason a link that measured fine last month is now losing 3 dB.
The diagnostic pattern worth carrying: a fibre link that has degraded over time, with no change to the equipment and no physical damage to the run, is contamination until proven otherwise, and it is usually the last connector somebody touched.
The transceiver, and what a form factor is
The module that plugs into the switch is where most of the confusion lives, and one distinction clears most of it up.
A form factor is a shape and an electrical interface. It says what fits in the cage and how fast the lane runs. It says nothing about the light.
The optic is what is inside. Which wavelength, which fibre type, how much power out, how sensitive the receiver, and therefore how far it reaches.
| Form factor | Typical speed |
|---|---|
| SFP | 1 Gbps |
| SFP+ | 10 Gbps |
| SFP28 | 25 Gbps |
| QSFP+ | 40 Gbps, four lanes of 10 |
| QSFP28 | 100 Gbps, four lanes of 25 |
The Q is for quad, and it is the reason the numbers work: a QSFP is four lanes in one module, which is why 40G was four times 10 and 100G was four times 25. That also explains why a QSFP can often be broken out into four separate links with the right cable, which is a genuinely useful thing to know exists.
Two modules of the same form factor can be completely different optics. An SFP+ might be 10GBASE-SR at 850 nm for multimode over a few hundred metres, or 10GBASE-LR at 1310 nm for single mode over ten kilometres. Same shape, same socket, same switch, and they will not talk to each other.
The same cages carry protocols other than Ethernet. Fibre Channel is a separate storage networking protocol with its own speeds and its own transceivers in the same form factors, which is why a box of SFPs pulled from a storage environment may look right and be built for a different protocol entirely.
If you already work on networks: why a mismatch gives you a link that comes up, and then does not work
The unhelpful thing about optical mismatches is that link status is a poor test of them. The interface goes up, the light is present, and the problem shows up as errors or as a link that fails at a distance nobody has tested at.
A single mode transceiver on multimode fibre is the classic. The narrow beam enters a wide core, launches into multiple modes, and over a short distance enough of it arrives to establish a link. Over a real run the modal dispersion it just created smears the signal and the link degrades or drops. So it works on the bench and fails in the building, which is the worst possible failure schedule.
Too much light is the mistake nobody expects, because more signal sounds safe. A long-reach single mode transceiver is built to put out enough power to survive forty kilometres. Plug two of those into a three metre patch lead and the receiver is saturated: it is being shouted at, and it cannot resolve the pulses. The link comes up and errors constantly. The fix is an inline attenuator, which is a component whose entire purpose is throwing signal away, and which sounds absurd until you have met this fault.
Wavelength mismatch between the two ends produces either nothing or a poor link, depending on how far apart they are. Two ends both need the same colour.
The diagnostic that settles all of these is the transceiver’s own reporting. Modules support digital diagnostics, and a switch will show transmit power, receive power, temperature and bias current per port. Receive power against the optic’s specified range answers the question directly: too low means loss, contamination or the wrong optic, and too high means saturation. That single number resolves more fibre arguments than any amount of swapping parts, and it is the first thing to look at rather than the last.
Making the two ends agree
Four things have to match, and they are ordered here by how often each one is the problem.
The fibre type. Single mode transceivers on single mode fibre, multimode on multimode. This is the one at the top of the page, and it is why the two yellow patch leads were not interchangeable.
The wavelength. Both transceivers, the same colour of light.
The connector and polish. Physically compatible, and angled to angled or flat to flat, never mixed.
The protocol and speed. Ethernet to Ethernet, and the same speed at both ends, since unlike copper there is generally no auto-negotiation to rescue a mismatch.
The reason this is a list rather than a single decision is that the four are bought separately, often by different people at different times. The fibre was installed by a contractor five years ago, the transceivers came with a purchase order last month, and the patch leads came out of a drawer.
Which suggests the habit worth building: write down what a link is made of. Fibre type and grade, transceiver part number at each end, wavelength, connector type. It takes a line in the documentation and it turns a future fault from an investigation into a comparison.
If you already work on networks: where the transceiver specifications actually live, and why vendor locking exists
The form factors are not defined by any of the standards this page cites. They come from the SFF committee, whose specifications are published through SNIA and are free to read, and which define the mechanical shape, the electrical interface, and the management interface a module presents.
That management interface is why a switch can tell you a module’s manufacturer, part number, serial number, and its live optical power readings. The module carries a small amount of memory describing itself, and the switch reads it on insertion.
Which is also where vendor locking comes from. Some switch vendors check the manufacturer field and refuse to enable a port with a module that is not theirs, sold at several times the price of an identical optic from a third party. The practice is contested, the third party market exists and codes its modules to match, and most vendors provide a command to permit unsupported optics while declining to support the result.
Two things worth carrying into a purchasing conversation. The optic itself is frequently manufactured by the same handful of companies whatever name is on it, so the technical argument for the premium is weaker than it is presented as. And the support argument is not nothing: if a link misbehaves and the optics are third party, the first response you get will be to replace them with the vendor’s own, and you will have to do it before anyone looks further.
Prove it
No commands here either, so the evidence is again a document and a question only that document answers.
ITU-T G.652. ITU-T recommendations are free to download, which makes this the one primary source in these two topics you can actually read. Open it and find what it says about the wavelength regions the fibre is intended for, then answer: does the recommendation define a single wavelength for single mode fibre, or a fibre that works across more than one, and which one is it designed around?
The SFF specifications, published through SNIA. Find the specification for a form factor named on this page. Answer a narrower question: does it specify the optical characteristics of the module, or the mechanical and electrical interface it presents? The answer tells you why two modules of the same form factor can be entirely different optics, which is the confusion this topic exists to remove.
Then, if you have any access to network equipment with fibre in it, read the transceiver diagnostics on a working port. Every managed switch exposes transmit and receive power per module. Comparing a working link’s receive power against the optic’s specified range is the single most useful fibre skill there is, and it is easiest to learn on a link that is behaving.
What trips people up
1. Assuming a link light means the optics match
A single mode transceiver on multimode fibre will frequently bring a link up over a short patch lead and fail over a real distance. Fibre mismatches are not excluded by link status, which is why the transceiver’s power readings matter more than the interface state.
2. Thinking more light is safer
A long-reach transceiver on a short link saturates the far receiver and produces constant errors. The fix is an attenuator, deliberately throwing signal away. More power is not a margin, it is a specification with a lower bound and an upper one.
3. Reading a form factor as a specification
SFP+ is a shape and a speed. It says nothing about wavelength, fibre type or distance. Two SFP+ modules can be incompatible with each other and both perfectly correct.
4. Mixing 62.5 and 50 micrometre multimode
OM1 has a 62.5 µm core and everything since has 50. Joining them loses a substantial amount of light in one direction, and it presents as unexplained loss on a run where every component is individually fine.
5. Mating an angled connector to a flat one
Green is angled, blue is flat. They do not seat correctly together, the loss is severe, and the end faces can be damaged. This is a five second visual check that people skip because both ends fit.
6. Cleaning a connector with whatever is to hand
A core is 9 micrometres across and a fingerprint is enormous by comparison. Wiping an end face on clothing or with a tissue drives contamination in and scratches the ferrule when it is mated under spring pressure. Cleaning tools are cheap and the alternative is replacing connectors.
Work it through
A campus has two buildings 400 metres apart. The existing link between them is 1 Gbps over multimode fibre installed in 2003, and it needs to become 10 Gbps. Somebody has priced 10G transceivers and the plan is to swap them at both ends.
Take the distance and the fibre together, because they decide everything. Four hundred metres is comfortably beyond what multimode carries at 10G. OM3 manages 10G to around 300 metres and OM4 further, but the fibre was installed in 2003, which almost certainly makes it OM1 or OM2, and neither carries 10G anywhere near 400 metres. So swapping the transceivers gives you a link that comes up on the bench and does not work between the buildings.
Establish what is actually in the ground before pricing anything else. The jacket markings at either end give the type and the core diameter, and the installation records may exist. That one check decides between three very different projects.
If it is OM1 or OM2, which is likely, there are two honest options. Pull new fibre, and if you are pulling anyway then pull single mode, because the cost difference is in the labour rather than the glass and single mode removes the distance question permanently. Or, if there are spare strands and the run is short enough for a different technology, look at what else the existing fibre supports at lower speed.
The option to be suspicious of is anything promising 10G over old multimode at 400 metres. It exists, at a price, using specialised optics, and it is a way of paying transceiver money to avoid a cable project while inheriting the constraint permanently.
The thing to write down at the end, whichever way it goes: the fibre type, the grade, the transceiver part numbers, and the measured receive power at both ends on the day it was commissioned. That last number is what a future fault gets compared against, and nobody ever regrets having recorded it.
Try it
Read a fibre jacket. If you can find a fibre patch lead anywhere, read the printing. It gives the type, the core and cladding as a pair such as 50/125, and the OM or OS grade. Compare what it says to the jacket colour and see whether the convention held.
Open G.652. It is free, it is the only primary source in these two topics you can read without paying, and the exercise in Prove it is a ten minute read. Knowing what a fibre recommendation actually contains is worth more than the specific answer.
Look at transceiver diagnostics. On any managed switch with fibre, find the command that shows optical power per module. Compare the receive power to the optic’s datasheet range. If you have no equipment, read a transceiver datasheet instead and find the receiver sensitivity and the maximum input power, which are the two numbers the panels above are about.
Check yourself
What physically differs between single mode and multimode fibre, and what follows from it?
The core diameter. Single mode is about 9 micrometres, multimode is 50 or 62.5.
A wide core allows light to travel by several different paths, each of which is a mode. Light taking a longer bouncing path arrives later, so a pulse spreads out as it travels. That is modal dispersion and it limits distance, more severely as speed goes up. A 9 micrometre core allows only one path, so there is no modal dispersion and the fibre carries signals tens of kilometres.
Multimode exists because the wide core is more forgiving. Alignment is looser, connectors are cheaper, and the light sources are cheaper, which makes it the economical choice inside a building.
A 10G link between two switches works on a 2 metre patch lead in the lab and fails when installed over 250 metres of existing fibre. What is likely?
Most likely a fibre type mismatch, and specifically a single mode transceiver on multimode fibre, or 10G optics on multimode that is too old a grade for the distance.
A narrow beam entering a wide core launches multiple modes, and over two metres enough of the signal arrives that the link comes up. Over 250 metres the dispersion that creates is enough to break it.
The check is the jacket marking on the installed fibre and the transceiver part numbers, and then the receive power reading at each end. Link status will not tell you, because it already said the link was up.
Why would anyone deliberately install an attenuator on a fibre link?
To stop the receiver being saturated.
A long-reach transceiver is built to put out enough power to survive tens of kilometres. Over a short link that power arrives almost undiminished and the far receiver cannot resolve the pulses, so the link comes up and errors continuously.
An attenuator throws away a specified amount of signal to bring the received power back inside the optic’s working range. Optical receivers have a maximum input power as well as a minimum sensitivity, and only the minimum is intuitive.
Two SFP+ modules both fit the same switch port. Does that mean they will work together?
No. A form factor is a shape and an electrical interface, and it says nothing about the light.
One SFP+ could be an 850 nm multimode optic for a few hundred metres and the other a 1310 nm single mode optic for ten kilometres. Both fit, both are correct, and they cannot talk to each other.
The same cages also carry protocols other than Ethernet, so a module pulled from a Fibre Channel environment will fit and is built for something else entirely.
A fibre link that has worked for two years is now losing signal. Nothing has been reconfigured and the run has not been damaged. What is the first thing to check?
Contamination on the connector end faces, and specifically the last one somebody touched.
A single mode core is 9 micrometres across, so a dust particle or a fingerprint blocks a meaningful fraction of the light. Because the faces are mated under spring pressure, contamination gets ground in, which can turn a cleanable problem into a scratched ferrule.
Check it by reading receive power at both ends and comparing against the optic’s range, then clean the connectors with a proper cleaning tool rather than anything to hand. Gradual degradation with no configuration change is contamination until proven otherwise.
What is the difference between an OM number and an OS number, and why must OM1 not be mixed with the rest?
OM grades are multimode and OS grades are single mode. The OM number is a bandwidth grade, so a higher number carries a given speed further, in the same way a copper category rates bandwidth rather than speed.
OM1 has a 62.5 micrometre core and OM2 onward have 50. Joining them means light crossing from a wide core into a narrow one, which loses a substantial fraction of it in that direction.
The practical version: buildings cabled in the 1990s are full of OM1, and connecting new OM3 to an existing OM1 backbone produces unexplained loss on a run where every individual component tests fine.
References
- Recommendation ITU-T G.652 - ITU-T, characteristics of a single-mode optical fibre and cable, including the wavelength regions it is designed around. Free to download. Accessed 2026-08-10.
- Recommendation ITU-T G.651.1 - ITU-T, multimode fibre for the optical access network. Accessed 2026-08-10.
- Recommendation ITU-T G.657 - ITU-T, bending-loss insensitive single mode fibre. Accessed 2026-08-10.
- SFF specifications - SNIA, the transceiver form factor specifications. Free to read. Accessed 2026-08-10.
- IEEE 802.3 Standard for Ethernet - IEEE Standards Association, which defines the Ethernet variants rather than the fibre. Accessed 2026-08-10.
Pictures. Freely licensed files from Wikimedia Commons, downloaded and served from this site rather than linked across to somebody else’s server. Each is resized and otherwise unaltered.
- LC optical fiber connector by Adamantios, CC BY-SA 3.0.
- SC optical fiber connector by Adamantios, CC BY-SA 3.0.
- LR-Link 10GBASE-SR SFP+ transceiver by Dmitry Nosachev, CC BY-SA 4.0.
Where the numbers came from. Nothing on this page is captured, for the same reason as the previous topic. The wavelength and fibre characteristics are from the ITU-T recommendations, which unlike the copper standards are free to download, so this topic rests on better evidence than the one before it. The distances given for the OM grades are industry conventions drawn from the IEEE Ethernet variants that specify them, and the jacket colours are a manufacturing convention rather than a requirement, which the panel about them says explicitly.
If you also work on Linux. Nothing here has a Linux counterpart. An operating system sees a link that is up or down and has no visibility of the glass, and transceiver diagnostics are read from the switch rather than from the host.