A cable is the one part of a network that cannot be reconfigured. Everything above it is software and can be changed in an afternoon; the run through the wall decides what that software is allowed to attempt for the next fifteen years.
This is the fourth post in a series working through networking from the bottom, and it covers layer 1: what the cable is made of, what the connector on the end of it is, how fast the link can run, and whether it can also carry power. The previous posts covered what a network is, network topologies and the Open Systems Interconnection (OSI) and TCP/IP models.
What the cable decides
Three things, and only three:
- How far the signal can travel before it degrades past recovery.
- How fast the link can run over that distance.
- How well it resists interference from everything else in the building.
Every cable choice is a trade between those and cost. The reason there are several standards rather than one is that a run between two racks and a run between two cities are not the same problem.
| Type | What it carries | Signal | Typical use |
|---|---|---|---|
| Twisted pair | Electrical current on copper pairs | Electrical | Ethernet to a desk or an access point |
| Coaxial | Electrical current on a shielded core | Electrical, radio frequency | Cable internet, television, antennas |
| Fibre-optic | Pulses of light in a glass core | Optical | Building-to-building, data centre, long distance |
| Auxiliary | Analogue sound | Analogue audio | Not a networking cable at all |
Copper cabling: twisted pair and coaxial
Copper cabling carries data as electrical signals along copper conductors. It is cheap, easy to terminate, and it is what nearly every desk in the world is connected with.
Its limits are all consequences of being an electrical conductor. The signal weakens over distance, and it picks up electromagnetic interference (EMI) from motors, fluorescent lighting, power cabling and the other pairs inside the same jacket.
Twisted-pair cable is the answer to that second problem. Pairs of insulated copper wires are twisted around each other along the whole run, so interference hits both wires of a pair almost equally and cancels out when the receiver reads the difference between them. Twisting also reduces crosstalk, which is one pair inducing a signal in its neighbour.
Two variants:
- Unshielded twisted pair (UTP) relies on the twisting alone. It is thinner, cheaper and easier to pull, and it is what most buildings use.
- Shielded twisted pair (STP) adds a conductive foil or braid around the pairs or the whole bundle. It costs more, is stiffer, needs correct grounding to work, and it is worth it in an industrial environment or alongside heavy power runs.
Coaxial cable puts a single central conductor inside insulation, wraps that in a conductive shield, and jackets the result. The shield is the point: it surrounds the signal conductor completely, which makes coaxial resistant to interference and suitable for radio-frequency work. It carries cable internet and television into homes, and it connects antennas. Its common connectors are the Bayonet Neill-Concelman (BNC) type, the F-type used for television, and SubMiniature version A (SMA) on radio equipment.
Fibre-optic cabling: single-mode and multi-mode
Fibre-optic cable carries data as pulses of light through a glass or plastic core rather than as current through metal.
That single difference removes most of copper's limits at once. Light is not affected by electromagnetic interference, so a fibre run can pass a transformer room without caring. Loss over distance is far lower, so runs are measured in kilometres rather than the hundred metres copper manages. There is no electrical path between the two ends, so a surge at one end cannot travel to the other.
Two kinds, and the difference is the width of the core:
- Single-mode fibre (SMF) has a very narrow core, around 9 micrometres, so light travels one path down it. Used for long distances, from campus links to intercontinental cable.
- Multi-mode fibre (MMF) has a wider core, typically 50 or 62.5 micrometres, so light bounces along several paths. Cheaper optics, shorter reach, and normal inside a building or a data centre.
The costs are real. Fibre is more fragile than copper, a bend tighter than its minimum radius attenuates the signal, terminating it needs a fusion splicer rather than a crimp tool, and the transceivers at each end cost more than a copper port.
The Cat standards compared
Twisted-pair cable is sold against category (Cat) standards, which define the electrical characteristics of the cable and therefore the speeds it can support over a given distance.
| Standard | Bandwidth | Speed at 100 m | 10 Gbps | Status |
|---|---|---|---|---|
| Cat5 | 100 MHz | 100 Mbps | No | Obsolete for new installations |
| Cat5e | 100 MHz | 1 Gbps | No | Still very widely installed |
| Cat6 | 250 MHz | 1 Gbps | Up to about 55 m | Common |
| Cat6A | 500 MHz | 10 Gbps | Yes, to 100 m | Current choice for new structured cabling |
A few things that table does not say outright.
The 100-metre figure is the whole Ethernet channel, not just the cable on the drum: it is normally taken as 90 metres of fixed horizontal cabling plus up to 10 metres of patch leads at each end.
The "e" in Cat5e is enhanced, and the enhancement is tighter crosstalk limits rather than more bandwidth. Cat5e and Cat5 both run at 100 MHz; only Cat5e is specified for gigabit.
Cat6 supports 10 Gbps only over a shortened run, commonly quoted as 55 metres and less in a bundle where cables interfere with each other. If a run needs 10 Gbps at full length, that is what Cat6A is for.
Cat6e and Cat6A
There is no Cat6e.
Cat6 and Cat6A (category 6 augmented) are the standards; "Cat6e" is a marketing label a manufacturer can print on a cable that is somewhere between them, and it means whatever that manufacturer decided. There is no specification behind it and nothing you can hold anyone to.
If someone offers Cat6e, the question to ask is which real category it is certified to. The answer is Cat6, Cat6A, or nothing.
RJ45 connectors
The connector on the end of a twisted-pair Ethernet cable is a registered jack 45 (RJ45). Its precise name is 8P8C, meaning eight positions and eight contacts, and that is what you will find in a specification.
It is worth being pedantic about one thing here, because it causes real confusion: RJ45 is the connector, not the cable. "An RJ45 cable" is not a thing. A cable is Cat5e or Cat6A; the RJ45 is the plastic plug crimped onto it.
You will find it on computers, switches, routers, access points and wall ports, and it is a passive connector: it holds eight wires against eight contacts and does nothing else.
Small form-factor pluggable modules
A small form-factor pluggable (SFP) module is not a connector. It is a hot-swappable transceiver that slides into an SFP cage on a switch or a router and contains the electronics and optics needed to turn the port's electrical signals into whatever the medium requires.
That is the whole reason SFPs exist: the port is generic and the module decides the medium. The same switch port takes a multi-mode fibre module today and a single-mode module for a longer run next month, and a copper SFP presents an RJ45 socket if that is what is needed.
| RJ45 | SFP | |
|---|---|---|
| What it is | A passive connector | An active transceiver module |
| Medium | Copper twisted pair | Fibre or copper, depending on the module |
| Changing the medium | Requires a different port | Swap the module |
| Typical use | A device to a switch, up to 100 m | Switch to switch, switch to router, long links |
Link speed
Link speed is the rate at which a link can transmit, negotiated between the two interfaces at each end and measured in bits per second.
The common Ethernet steps are 10 Mbps, 100 Mbps (Fast Ethernet), 1 Gbps (Gigabit Ethernet), 2.5 Gbps, 5 Gbps and 10 Gbps, with 25, 40, 100 Gbps and beyond in data centres.
Two things are worth holding on to. A link negotiates to the highest speed both ends and the cable support, so a 10 Gbps switch port with a Cat5e run and a gigabit network card gives you a gigabit link and no error anywhere. And link speed is bandwidth in the sense of the first post in this series: it is a ceiling, and throughput is what you actually get under it.
You can read the negotiated speed of an interface directly:
ethtool <interface>
It reports the speed and duplex the link actually settled on, which is the number to check when a gigabit link is somehow moving data at a hundred megabits.
Power over Ethernet
Power over Ethernet (PoE) carries electrical power along the same twisted-pair cable as the data, so a device can be installed where there is a network drop and no power socket.
This is why ceiling access points, security cameras, desk phones and door controllers are practical to deploy at all. One cable, one run, one thing to install.
| Standard | Common name | Power at the source | Typical use |
|---|---|---|---|
| IEEE 802.3af | PoE | Up to 15.4 W | Phones, basic access points |
| IEEE 802.3at | PoE+ | Up to 30 W | Pan-tilt-zoom cameras, larger access points |
| IEEE 802.3bt | PoE++ | Up to 60 W or 90 W depending on type | Wi-Fi 6 and 7 access points, small displays |
Those are the figures the power sourcing equipment supplies. The device at the far end receives less, because the cable itself dissipates some of it as heat, and a longer run loses more. Budget for the number at the device, not the number on the switch.
Institute of Electrical and Electronics Engineers (IEEE) standards are what these are: 802.3 is Ethernet, and af, at and bt are amendments to it.
Link speed compared with Power over Ethernet
These get conflated constantly, and they have nothing to do with each other.
| Link speed | Power over Ethernet | |
|---|---|---|
| Concerns | Moving data | Delivering electrical power |
| Measured in | Mbps or Gbps | Watts |
| Set by | Negotiation between both interfaces and the cable | The switch's PoE standard and the device's class |
| Example | 1 Gbps | 30 W of PoE+ |
A port can be 1 Gbps with PoE, 10 Gbps with PoE, or 10 Gbps with none. PoE does not make a link faster and a fast link does not imply power.
What comes next in this series
The next post covers layer 2: what an Ethernet frame contains field by field, how a switch learns which device is on which port, what it does when it does not know, and the difference between a collision domain and a broadcast domain.
These posts follow my own notes as I work through the material, so they go in the order the topics are learned rather than the order they would be taught in a reference.
