Ethernet

On this page 7

Ethernet is the family of wired local area network technologies standardized as IEEE 802.3, which defines both the physical layer (cables, connectors and signaling) and the data link framing that devices use to exchange data. In networking, it is the default way to connect computers, servers, switches, routers, access points and storage inside a building or a data center.

Ethernet began at Xerox PARC in the 1970s as a shared coaxial bus and became an IEEE standard in 1983. Its frame format has stayed remarkably stable while link speeds grew from 10 Mbit/s to 800 Gbit/s, which is the main reason it displaced competing LAN technologies such as Token Ring and FDDI.

How Ethernet moves data in frames

Ethernet moves data in self-contained units called frames, each addressed with a 48-bit MAC address of the sender and the receiver. A network interface reads the destination address of every frame it receives and accepts the frame only if the address matches its own, a broadcast address or a multicast group it has joined.

Every frame ends with a frame check sequence (FCS), a 32-bit cyclic redundancy check computed over the addresses, type field and payload. A receiver that calculates a different value silently discards the frame; Ethernet itself never retransmits, so recovery is left to higher layers such as TCP. The table below lists the fields of a standard frame on the wire.

FieldSizePurpose
Preamble and SFD8 bytesClock synchronization and start-of-frame marker
Destination MAC6 bytesReceiving interface, multicast group or broadcast
Source MAC6 bytesSending interface
802.1Q tag4 bytes (optional)VLAN ID and priority bits
EtherType2 bytesPayload protocol, such as 0x0800 for IPv4 or 0x86DD for IPv6
Payload46 to 1,500 bytesPacket of the upper-layer protocol, padded if shorter
FCS4 bytesCRC-32 error detection
Fields of an Ethernet II frame

Counting from the destination address to the FCS, a frame is between 64 and 1,518 bytes, or 1,522 bytes with a VLAN tag. The 1,500-byte payload limit is the familiar Ethernet MTU. Between frames, a sender must leave an interframe gap of 96 bit times, so a full-size frame occupies 1,538 byte times on the wire and the best-case payload efficiency is 1,500 / 1,538, or about 97.5%.

From shared bus to switched full duplex

Classic Ethernet used CSMA/CD (carrier sense multiple access with collision detection): stations listened before sending, detected collisions on the shared medium and retried after a random backoff. The 64-byte minimum frame size comes from this era, because a sender had to still be transmitting when a collision signal returned from the far end of the segment.

Modern Ethernet is switched and full duplex. Every device has a dedicated link to a switch port, transmits and receives at the same time on separate signal paths, and never experiences collisions, so CSMA/CD is effectively unused. Forwarding decisions are made by the network switch, which learns which MAC address sits behind which port; that behavior belongs to the neighboring topics of Layer 2 switching and link aggregation rather than to Ethernet itself.

Links agree on speed and duplex through auto-negotiation, defined in the 802.3 standard for twisted-pair media. Forcing settings on one side while the other side negotiates is a classic cause of a duplex mismatch, in which the link works but performs poorly under load.

Ethernet speeds, media and reach

Each Ethernet variant is named by speed, signaling and medium: in 1000BASE-T, 1000 is the rate in Mbit/s, BASE means baseband signaling and T means twisted pair. Copper links over twisted pair are limited to 100 meters per channel, usually 90 meters of fixed cabling plus 10 meters of patch cords, while fiber reaches hundreds of meters to tens of kilometers. The table compares common variants.

VariantStandardMediumMaximum reach
100BASE-TX802.3u (1995)Cat 5 twisted pair, 2 pairs100 m
1000BASE-T802.3ab (1999)Cat 5e twisted pair, 4 pairs100 m
2.5GBASE-T / 5GBASE-T802.3bz (2016)Cat 5e or Cat 6 twisted pair100 m
10GBASE-T802.3an (2006)Cat 6a twisted pair100 m (about 55 m on Cat 6)
10GBASE-SR802.3ae (2002)Multimode fiber, 850 nm300 m on OM3, 400 m on OM4
10GBASE-LR802.3ae (2002)Single-mode fiber, 1310 nm10 km
100GBASE-SR4802.3bm (2015)Multimode fiber, 4 lanes100 m on OM4
Common Ethernet physical-layer variants

In data centers, 25, 100 and 400 Gbit/s optical and direct-attach copper links dominate, plugged into SFP28, QSFP28 or QSFP-DD module cages. In offices, 1 Gbit/s to the desk remains standard, while 2.5 and 5 Gbit/s variants exist mainly to feed fast Wi-Fi access points over existing Cat 5e and Cat 6 cabling.

Power over Ethernet

Power over Ethernet (PoE) delivers DC power over the same twisted-pair cable that carries data, so phones, cameras and access points need no separate power supply. The switch or injector acts as power sourcing equipment and powers a device only after detecting a valid signature resistance, which protects ordinary network cards. Power arriving at the device is lower than the power sent, because some is lost as heat in the cable.

StandardCommon namePower at the switch portPower available at the device
802.3afPoE (Type 1)15.4 W12.95 W
802.3atPoE+ (Type 2)30 W25.5 W
802.3bt Type 3PoE++ (4-pair)60 W51 W
802.3bt Type 4PoE++ (4-pair)90 W71.3 W
IEEE PoE classes and power budgets

When planning PoE, the decisive figure is the switch’s total power budget, not the per-port maximum. A 24-port switch with a 370 W budget cannot supply 30 W to every port at once, so heavy loads such as pan-tilt-zoom cameras must be counted explicitly.

Common Ethernet problems and their causes

Most Ethernet faults show up in interface error counters long before users report them, so reading those counters is the first diagnostic step. The table maps typical symptoms to their usual causes and fixes.

SymptomLikely causeFix
Rising CRC or FCS errorsDamaged cable, poor termination, interferenceTest and replace the cable or patch cord
Late collisions on a full-duplex networkDuplex mismatchSet both ends to auto-negotiation
Link at 100 Mbit/s instead of 1 Gbit/sBroken pair in a 4-pair cableRe-terminate or replace the cable
Link flappingFaulty optic, dirty fiber connector, PoE overloadClean or swap the optic, check the power budget
Large transfers stall, small ones workMTU mismatch with jumbo framesUse the same MTU on every hop of the path
Ethernet symptoms, causes and fixes

A frequent misconception is that a higher cable category makes a link faster. The speed is set by the interfaces at both ends; a better cable only provides the margin a given speed requires. Likewise, jumbo frames of about 9,000 bytes are not part of the base standard and help only when every device on the path supports the same size.

In simple terms

Ethernet is the common language of wired network devices. It decides what the cable and plug look like, how signals travel through the wire and how data is packed into small labeled envelopes called frames.

Each envelope carries the address of the sender and the receiver, plus a check value that reveals damage on the way. A switch reads the addresses and delivers each envelope to the right port, much like a mail room sorting letters. The key point is that Ethernet is the dependable wired foundation under almost every office and data center network.

A practical example

A video production studio connects twelve editing workstations and a network storage server to one managed switch. The workstations use 2.5GBASE-T over existing Cat 6 cabling, and the storage server uses a 10GBASE-T port so that several editors can read footage at the same time.

Copying a 50 GB project at an actual throughput of about 2.35 Gbit/s takes roughly 170 seconds, compared with about 7 minutes on a 1 Gbit/s link delivering 940 Mbit/s. After one workstation shows rising CRC errors and slow transfers, the administrator replaces a crushed patch cord and the counters stop increasing. The upgrade cut large file copies by more than half without pulling a single new cable.