Cabling and the Physical Layer

What the physical layer does

The physical layer, layer 1 of the OSI model, moves raw bits from one device to the next. It does not know about addresses or frames; its job is to turn ones and zeros into something that can travel through a medium, and back again. There are three ways to do that:

• Electrical signals on copper: changes in voltage on twisted-pair or coaxial cable.
• Light in fiber-optic cable: pulses from a laser or light-emitting diode (LED) through a thin strand of glass.
• Radio waves through the air: Wi-Fi, Bluetooth, cellular. See Wireless Networking.

The physical layer also defines the practical details: cable types, connectors, pin assignments, maximum lengths, and how bits are encoded as signals. Layer 1 problems are the most common cause of network outages, and the easiest to overlook: before you debug a routing table, check the cable.

Twisted-pair copper

Twisted pair is copper cable with eight wires twisted together in four pairs. Each pair carries a signal as the difference between its two wires. Interference from outside hits both wires of a pair almost equally, so the difference — the signal — survives. The twisting also reduces crosstalk, interference between neighboring pairs. It is the standard cable from a wall outlet to a desk, and between devices in a rack.

UTP and STP

Unshielded twisted pair (UTP) relies on the twisting alone. It is cheap, thin and flexible, and it is what most offices use. Shielded twisted pair (STP) adds foil or braid around the pairs, or around the whole cable, for environments with heavy electrical noise such as factories or cable runs near motors. The shield only works when it is grounded properly at the ends, which makes STP more work to install. Names like F/UTP and S/FTP describe exactly what is shielded: the part before the slash is the overall shield, the part after it the shield per pair.

Cable categories

Twisted-pair cable comes in categories (Cat) defined by the Telecommunications Industry Association (TIA). A higher category supports higher signal frequencies, and so higher speeds.

The 100 m limit is for the whole channel: at most 90 m of fixed cable in the wall (the permanent link) plus 10 m of patch cords at both ends together. Beyond that the signal becomes too weak and distorted, and errors appear — often not as a dead link, but as a link that works badly.

RJ45 and wiring standards

RJ45 is the name everyone uses for the eight-pin modular connector on twisted-pair Ethernet cables and ports. (Strictly speaking, the connector is an 8P8C — eight positions, eight contacts — but nobody in a network team calls it that.) Which colored wire goes to which pin is fixed by one of two TIA standards, T568A and T568B. They differ only in that the orange and green pairs trade places.

Either standard works, as long as a building uses one consistently. 10BASE-T and 100BASE-TX use only pins 1, 2, 3 and 6 (two pairs); 1000BASE-T and faster use all four pairs.

Straight-through, crossover and Auto-MDIX

A device must send on the pins where the other end listens. PCs, routers and access points send on pins 1 and 2 and receive on 3 and 6 (called MDI, medium dependent interface); switches do the opposite (MDI-X, MDI crossover). So:

• A straight-through cable has the same standard on both ends (usually T568B). It connects unlike devices: PC to switch, router to switch.
• A crossover cable has T568A on one end and T568B on the other. It connects like devices: switch to switch, PC to PC, router to router, and also router to PC.

In practice you rarely need a crossover cable anymore. Auto-MDIX lets a port detect the cable type and swap its send and receive pairs automatically. It is on by default on current switches and is defined as an optional feature of 1000BASE-T, so on current equipment a straight-through cable almost always works. The one special cable you still meet is the rollover (console) cable, with pins 1–8 reversed, for the console port of Cisco devices.

Fiber-optic cable

A fiber-optic cable carries data as pulses of light through a thin strand of glass. The strand has a core that carries the light and a cladding around it (125 µm across) that reflects the light back into the core. A link normally uses two fibers, one for each direction: the transmit side of one device connects to the receive side of the other. Fiber has three big advantages over copper:

• Distance: from hundreds of meters to tens of kilometers, instead of 100 m.
• Bandwidth: 10, 100 and 400 Gbit/s links are routine.
• Immunity to interference: light is not affected by electromagnetic interference (EMI), and fiber carries no electricity between buildings, so it is also safe against ground-potential differences and lightning.

The downsides are cost (especially of the optics at both ends) and care: fiber must not be bent too sharply, and a speck of dust on a connector can block the light.

Single-mode and multimode

Multimode fiber (MMF) has a relatively wide core of 50 or 62.5 µm. Light enters at many angles and travels along many paths (modes) of slightly different length, so a sharp pulse smears out over distance. That limits multimode to a few hundred meters at 10 Gbit/s, but the wide core accepts cheap light sources (typically at 850 nm), which makes it the usual choice inside a building or data center. Grades are OM1 to OM5; OM3 and OM4 (aqua jacket; OM4 sometimes violet) are the common ones today.

Single-mode fiber (SMF) has a core of about 9 µm, so narrow that light follows essentially one path. Pulses stay sharp over long distances: 10 km, 40 km or more, using lasers at 1310 or 1550 nm. The optics are more expensive, but the cable itself is not. Single-mode (grades OS1 and OS2, usually a yellow jacket) is used between buildings, on campuses and by providers.

Fiber connectors

• LC (Lucent Connector): small, with a latch like RJ45, usually in duplex pairs. The standard connector on SFP transceivers.
• SC (Subscriber Connector): larger and square, push-pull. Common on older equipment and in provider networks.
• ST (Straight Tip): round with a bayonet twist lock. Older installations.
• MPO/MTP: one connector with several fibers, usually 8, 12, 16 or 24, for 40 and 100 Gbit/s links that use several fibers in parallel.

Connector end faces are either UPC (flat polish, blue) or APC (angled polish, green). They must not be mixed: an APC connector in a UPC port leaves an air gap and loses most of the light.

Transceivers

A transceiver (transmitter plus receiver) is the component that converts between the electrical signals inside a device and the signals on the medium. On switches and routers, fiber ports are empty cages into which you plug a small, hot-swappable transceiver module (an "optic"). That way one switch model can serve short multimode links and long single-mode links: you only change the module.

Both ends of a fiber link need matching optics: the same fiber type (multimode or single-mode) and the same wavelength. For short links inside a rack, a direct attach copper (DAC) cable — a twinax cable with a transceiver molded on each end — is cheaper than two optics and a fiber. Some vendors only accept their own coded optics, so check compatibility before ordering.

Coaxial cable

Coaxial (coax) cable has one central copper conductor, an insulating layer, and a braided shield around it. Early Ethernet (10BASE5 and 10BASE2) used coax in a bus topology, but that is history. Today you meet coax at the edge of the network: cable internet uses RG-6 coax with F-type screw connectors, following the Data Over Cable Service Interface Specification (DOCSIS) standards, between the street and the cable modem.

Ethernet physical standards

IEEE 802.3 names each Ethernet physical standard with a code: the speed in Mbit/s (or with a G for Gbit/s), BASE for baseband signaling, and a suffix for the medium. T means twisted pair; for fiber, S is short wavelength (850 nm, multimode), L is long wavelength (1310 nm) and E is extended (1550 nm).

Power over Ethernet

Power over Ethernet (PoE) sends electrical power over the same twisted-pair cable as the data. A wireless access point, IP phone or security camera then needs only one cable, and no power outlet near the ceiling. The device that supplies power (usually a PoE switch, sometimes an injector) is the power sourcing equipment (PSE); the device that uses it is the powered device (PD). The PSE first detects a PD by its signature resistance, so a normal laptop plugged into a PoE port receives no power, and then learns which power class it needs.

The difference between the two columns is lost as heat in the cable. A PoE switch also has a total power budget for all its ports together, which is often less than the maximum per port times the number of ports. Planning a new floor of access points is partly a sum:

CLASS_WATTS = {2: 7.0, 3: 15.4, 4: 30.0}   # power reserved per port, by PoE class

devices = [("IP phone", 2, 24), ("Access point", 4, 6), ("Camera", 3, 4)]
budget = 370.0                             # the switch's total PoE budget in watts

total = 0.0
for name, poe_class, count in devices:
    watts = CLASS_WATTS[poe_class] * count
    total += watts
    print(f"{count:>2} x {name:<13} class {poe_class}  {watts:6.1f} W")

print(f"Needed: {total:.1f} W of {budget:.1f} W")
if total > budget:
    print(f"Short by {total - budget:.1f} W: some ports will get no power")

24 x IP phone      class 2   168.0 W
 6 x Access point  class 4   180.0 W
 4 x Camera        class 3    61.6 W
Needed: 409.6 W of 370.0 W
Short by 39.6 W: some ports will get no power

Real switches can be smarter — many allocate the power a device actually requests through the Link Layer Discovery Protocol (LLDP) or Cisco Discovery Protocol (CDP) instead of the class maximum — but a class-based sum is the safe estimate. On a Cisco switch, show power inline shows the budget and what each port draws.

Common physical problems

Many tickets that look like "the network is slow" or "the connection keeps dropping" end at layer 1. The usual suspects:

• A bad patch cable: a crushed cable under a desk, a broken latch, a self-made cable with the wrong pin order. Swap it first; it is the cheapest test there is.
• Too long or the wrong category: a link that negotiates to 100 Mbit/s instead of 1 Gbit/s often has a pair that does not work, or a run over 100 m.
• Interference: UTP laid alongside power cables, motors or fluorescent lights.
• Dirty or damaged fiber connectors: dust on an end face causes high loss. Inspect and clean every connector before plugging it in.
• Swapped fibers (transmit to transmit) or mismatched optics (a single-mode optic on multimode fiber, or different wavelengths at each end): the link stays down.
• Duplex mismatch: one side full duplex, the other half duplex. Strictly a layer 2 setting, but it shows up as the same kind of errors; see Ethernet and MAC Addresses.

The interface counters on the switch show whether frames arrive damaged:

Switch1# show interfaces GigabitEthernet1/0/12
GigabitEthernet1/0/12 is up, line protocol is up (connected)
  Hardware is Gigabit Ethernet, address is 0000.5e00.530c (bia 0000.5e00.530c)
  MTU 1500 bytes, BW 1000000 Kbit/sec, DLY 10 usec,
…
  Full-duplex, 1000Mb/s, media type is 10/100/1000BaseTX
…
     184223 packets input, 51872311 bytes, 0 no buffer
     Received 1421 broadcasts (1305 multicasts)
     0 runts, 0 giants, 0 throttles
     2319 input errors, 2311 CRC, 8 frame, 0 overrun, 0 ignored
…
     0 output errors, 0 collisions, 0 interface resets

• is up, line protocol is up: layers 1 and 2 are working, so the link itself is established.
• Full-duplex, 1000Mb/s: the negotiated duplex and speed.
• 2311 CRC: frames that failed the cyclic redundancy check (CRC), because their Frame Check Sequence did not match — the bits were damaged on the way. A CRC count that keeps growing points to a cable, connector, interference or duplex problem. Run clear counters GigabitEthernet1/0/12 and check again after a few minutes to see whether errors are still increasing.