Ethernet

In short: The dominant technology for wired local area networks (LANs) — defines how devices exchange data over cable in the form of frames.

In more detail: Ethernet covers both the physical cabling (see Cat categories and specifications) and the frame format at layer 2 of the OSI model, including MAC addresses for addressing devices in the local network. Speeds range from 10 Mbit/s (historically) up to several hundred Gbit/s.

In Depth

Frame structure

An Ethernet frame has a standardised structure that is sent along with every transmission:

| Destination MAC (6 bytes) | Source MAC (6 bytes) | EtherType (2 bytes) | Payload (46-1500 bytes) | CRC checksum (4 bytes) |

The EtherType field states which protocol the payload contains (e.g. IPv4, IPv6 or ARP), and the CRC checksum (cyclic redundancy check) at the end allows the recipient to detect transmission errors at the bit level and simply discard damaged frames instead of passing on faulty data. The maximum payload of 1500 bytes (maximum transmission unit, MTU) is the historical default and the reason why larger amounts of data have to be split into several packets at higher protocol levels — some modern networks optionally support larger “jumbo frames” (up to 9000 bytes), which reduces the relative header overhead per transferred byte at very high data throughput (e.g. in data centres).

From shared medium to full duplex

Historically, Ethernet was a shared-medium method with CSMA/CD (carrier sense multiple access/collision detection): all devices on a shared cable segment listened before sending to see whether the line was free (“carrier sense”), and if two devices sent at the same time (“collision”) both had to abort immediately and try again after a randomly chosen waiting time — a method that worked well with a few devices but became increasingly inefficient as the number of participants rose, because collisions occurred more often. Modern networks with switches have eliminated this problem completely: every port has its own collision-free full-duplex connection to the switch (each device effectively hangs on its own dedicated “cable segment”), which lets all connected devices send AND receive at the same time without disturbing each other. CSMA/CD is therefore practically obsolete in modern Ethernet installations, but technically remains part of the standard for the rare case of real half-duplex connections.

Speed designations

The speed designations follow a fixed, easily readable convention: the number states the speed in Mbit/s (or with “G” in Gbit/s), “BASE” stands for baseband transmission (the entire cable is used for one signal, no multiplexing of several channels), and the suffix describes the transmission medium — e.g. “1000BASE-T” (1000 Mbit/s over twisted-pair copper cable), “10GBASE-SR” (10 Gbit/s over fibre, “short range”) or “100BASE-FX” (100 Mbit/s over fibre). The matching cable category limits the maximum achievable speed over copper in each case — a Cat5e cable, for example, supports at most 1000BASE-T; 10GBASE-T needs at least Cat6a.

Auto-negotiation

When a connection is established, two devices connected via Ethernet automatically negotiate the highest speed and duplex mode they both support (“auto-negotiation”) — if this process goes wrong (e.g. due to outdated drivers or a misconfiguration on one side), the result can be a connection that seems to work but is inexplicably slow with many transmission errors, because the two sides have “agreed” on different duplex settings.

See also: MAC addresses, Cat categories and specifications, Frames, Switch