What Is Ethernet? How a Wired LAN Carries Your Data
Ethernet is the worldwide standard (IEEE 802.3) for wired LANs, which link devices with cables. It puts data into addressed packages called "frames." A device called a switch delivers each frame only to the device it is addressed to. Wi-Fi shares the airwaves, but each Ethernet cable serves one device alone, so speed stays steady.
What you'll learn
- What part of the connection Ethernet covers, and how it relates to the internet
- What is inside a frame, and how a switch tells destinations apart
- Why the wires in a network cable are twisted, and why cables stop at 100 m
- How speeds have grown, PoE (power over the same cable), and how Ethernet differs from Wi-Fi
Prerequisites: none. Reading "How Is Data Sent Over the Internet?" first shows how the pieces fit together.
In everyday life
On the back of a home router is a row of square sockets labeled "LAN." Plug a cable in here, and a computer or game console joins the network by wire. These cables are often called Ethernet cables or LAN cables. Ethernet is the set of rules for exchanging data over them.
In offices, cables run to sockets under desks or in the floor. Some Wi-Fi access points on the ceiling, and some security cameras, are connected by just one network cable. They have no power cord, because the same cable also carries electricity. This is called PoE, and we explain it later.
Ethernet is the standard for wired LANs
A network that links devices within a small area, such as a home or office, is called a LAN (local area network). A LAN that uses radio is Wi-Fi, or wireless LAN. The main kind of LAN that uses cables is Ethernet.
The rules are set by IEEE, an international organization of engineers, as a standard numbered "802.3." The Wi-Fi standard is "802.11," so both belong to the same "802" family. The rules cover everything from the electrical signals on the cable to how data is packaged.
Ethernet carries data only as far as the next device on the cable. Delivering it further, across networks to a distant destination, is the job of a separate set of rules called IP. This division of labor is explained in the part about "layers" in "How Is Data Sent Over the Internet?"
Born inside a research lab in 1973
Ethernet was born in 1973 at Xerox's Palo Alto Research Center (PARC) in the US. Bob Metcalfe set out the idea in a memo dated May 22 that year, and built it with David Boggs. It was made to link the lab's "Alto" computers. Its speed was about 2.94 million bits per second (2.94 Mb/s). A bit is a single digit that is either 0 or 1.
In 1980, three companies, Xerox, DEC and Intel, published a specification for 10 million bits per second (10 Mb/s). It is called "DIX," from the companies' initials. Based on it, IEEE approved the 802.3 standard in June 1983 (it was published in 1985). Because the specification was public, any company could build equipment to the same standard.
Data travels in packages called frames
Ethernet sends data in packages with a fixed layout. Each package is called a "frame." Roughly, a frame holds four things. (To keep things simple, we skip parts such as the signal that marks its start.)
- Destination and sender numbers: which device it is for, and which device sent it
- Type: a marker that says which set of rules (for example, IP) the contents follow
- Contents: the data being carried, up to 1,500 bytes per frame (1 byte is 8 bits)
- Check value: 4 bytes used to check that the frame was not damaged on the way
The device number is a 48-bit number given to each network port at the factory. It is called a MAC address. Its first part is a number that IEEE assigns to each maker. An IP address is a number for finding a destination across networks. A MAC address is a number for telling neighboring devices apart within the same LAN (see "What Is an IP Address?").
The check value at the end is called the FCS. The sender works it out from the part of the frame that runs from the destination to the contents, using a calculation called CRC. The receiver does the same calculation. If the values do not match, it throws the frame away. How the calculation works, with a diagram of the frame, is in "How CRC Works."
Ethernet itself does not resend frames that were thrown away. Recovering lost data is the job of TCP at the sender (see "TCP vs UDP: What's the Difference?").
The past: everyone shared one cable
Early Ethernet connected many devices to one thick cable. A signal sent by anyone reached everyone on the cable. Boxes called "hubs," which spread later, did much the same. A hub passed every incoming signal straight out to all of its sockets.
In this setup, if two devices sent at once, their signals overlapped and were ruined. This is called a collision. So a device first checked that the cable was free before sending. If it noticed a collision, it waited a random time and sent again. This procedure is called CSMA/CD.
Wi-Fi uses a similar procedure that checks the air is free before sending (CSMA/CA). In today's wired LANs, however, CSMA/CD is hardly used. Standards for 10 Gb/s and faster do not include it at all. That is because switches, which we look at next, got rid of collisions altogether.
Today: switches deliver only to the destination
At the center of a wired LAN today is a device called a switch. For each of its sockets (ports), a switch learns the MAC addresses of the devices beyond it. It learns in a simple way. For each frame that arrives, it writes down the "sender" number and the port it came in on, in a table.
When a frame arrives, the switch looks up the destination number in its table and sends the frame out of that port only. If the destination is not in the table yet, it sends the frame out of all the other ports. Once the destination replies, it goes into the table. Numbers that have been silent for a while are removed from the table.
Today, a device can also send and receive at the same time on its link to the switch. This is called full duplex. With network cables, speeds up to 100 Mb/s use separate wires for sending and receiving. At 1 Gb/s and above, a technique sends both ways over the same wires. Each cable serves one device alone and carries traffic both ways at once, so collisions do not happen.
A switch delivers within a LAN by looking at MAC addresses. Relaying between networks by looking at IP addresses is the job of a router. Most home routers have a small switch built in. That is why there are about four LAN ports on the back. The differences between the three devices are laid out in a table in Hub vs Switch vs Router.
Speed over the years: from 10 Mb/s to 800 Gb/s
Ethernet has raised its speed many times while keeping its basic rules for connecting. The main standards are below. (Years are when IEEE approved each one. 1 Gb/s is 1 billion bits per second.)
| Speed | Main standard | Typical use |
|---|---|---|
| 10 Mb/s | 802.3 (1983); network-cable version in 1990 | Early office LANs |
| 100 Mb/s | 802.3u (1995) | PCs of the late 1990s |
| 1 Gb/s | 802.3ab (1999, network-cable version) | Today's home PCs and routers |
| 10 Gb/s | 802.3ae (2002) | Office backbones; homes with 10 Gb/s fiber |
| 100 Gb/s | 802.3ba (2010) | Data centers |
| 400 Gb/s | 802.3bs (2017) | Data centers |
| 800 Gb/s | 802.3df (2024) | Data centers for AI |
A standard for 1.6 Tb/s (1.6 trillion bits per second), 802.3dj, has also been in the works, with the aim of finishing it in 2026. Most fast standards run over optical fiber rather than copper wire.
Why are the wires in a network cable twisted?
Cut open a network cable, and you find eight thin wires inside, twisted together in pairs. A cable made of twisted pairs of wires is called twisted-pair cable. The plug at the end has eight contacts and a shape called RJ45.
The wires come in pairs so that the signal can be sent in a way that resists noise. The sender puts opposite signals on the two wires. The receiver looks only at the voltage difference between them to decide 0 or 1. This way of sending is called differential signaling.
Noise from motors, fluorescent lights and the like lands almost equally on two wires that sit side by side. Taking the difference cancels out noise that landed equally. Twisting the pair makes each wire move closer to the noise source and away from it, over and over. So the amount of noise on the two wires tends to even out.
USB uses the same method of sending on a pair of wires (see "What Is USB?"). A table comparing it with sending on a single wire is in Single-Ended vs Differential Signaling.
Cable "categories," and optical fiber
Network cables come in grades such as "Category 5e" and "Category 6A." A higher number means the cable is built to carry signals that change faster (higher-frequency signals). Here is a rough guide.
- Category 5e: 1 Gb/s up to 100 m. Often enough for a home
- Category 6: can also carry 10 Gb/s, but only for a few tens of meters
- Category 6A: 10 Gb/s up to 100 m
- Category 8: for 25 and 40 Gb/s, but only up to 30 m. Mainly for data centers
The standard limits a network cable to 100 m in total. Fixed wiring, such as inside walls, can be up to 90 m. The cords at both ends can add up to 10 m. The longer the cable, the weaker the signal gets, and the harder it is to tell apart from noise.
In data centers and between buildings, optical fiber cable is also common. Optical fiber sends signals as light through a thin strand of glass. It is not affected by electrical noise, and it reaches far. The chips inside the parts that convert between light and electricity are introduced in "What Is Silicon Photonics?"
PoE: sending power over a network cable
There is a way to send electricity along with data over the wires of a network cable. It is called PoE (Power over Ethernet). It is handy for Wi-Fi access points on ceilings, security cameras and office phones. They can go where there is no power outlet, with just one cable.
| Standard | Approved | Power sent | Power the device can use |
|---|---|---|---|
| 802.3af | 2003 | 15.4 W | 12.95 W |
| 802.3at | 2009 | 30 W | 25.5 W |
| 802.3bt | 2018 | 60 W / 90 W | 51 W / 71.3 W |
The figures for the sending side and the device differ because some of the electricity turns into heat in the thin wires. A PoE switch checks that the device at the other end supports PoE before it sends power. So with standard PoE, plugging in a computer that does not support it will not damage the computer.
Why is it steadier than Wi-Fi?
With Wi-Fi, nearby devices take turns using the radio waves on the same path, called a channel. Radio from a neighbor's Wi-Fi or a microwave oven can also mix in. So speed and delay change with how crowded the air is and how far away you are (see "How Does Wi-Fi Work?").
On a wired LAN connected to a switch, each cable serves one device alone and carries traffic both ways at once. There is no waiting for a turn as with radio, and differential signaling resists noise. So speed and delay stay consistent. That is why people prefer wired connections for online gaming and for moving large files. The trade-off is that it works only where the cable reaches.
The chips at work inside Ethernet
Behind the LAN port of a computer or router are two kinds of circuits with different jobs. One turns 0s and 1s into electrical signals for the cable, and turns incoming signals back into 0s and 1s. It also cleans up the shape of signals that have weakened along the cable. This is called the PHY (the physical-layer circuit).
The other packs data into frames, and adds and checks the check value. This is called the MAC (the same MAC as in "MAC address"). The MAC and the PHY may be combined into one small chip. Or the MAC may be built into the computer's main chip.
At the heart of a data center switch is a special chip whose only job is to sort frames. For example, one product that began shipping in 2025 handles a total of 102.4 trillion bits per second on a single chip. Along the chip's edges sit many circuits that combine data into a single fast signal for sending. These circuits are called SerDes. The big picture of data centers is explained in "What Is a Data Center?"
Common misconceptions
Myth: Ethernet and the internet are the same thing
In fact, Ethernet is a set of rules for carrying data to the next device on a cable. The internet is the whole of the world's networks, linked together by IP. The names sound alike, so people often mix them up. A wired LAN at home covers only the stretch up to the entrance to the internet (the router). Some LANs, such as networks used only inside a company, are not connected to the internet at all.
Myth: A Category 8 cable will make your home internet faster
In fact, the top speed is usually set by your fiber plan and by the ports on your router and computer. Take Japan as an example. The "FLET'S Hikari Cross" service from NTT East and NTT West offers up to about 10 Gb/s (since 2020). Category 6A can carry 10 Gb/s up to 100 m. Category 8 is a standard for data center wiring of up to 30 m.
Myth: A network cable can be as long as you like
In fact, the network-cable standard allows up to 100 m in total. Beyond that, the signal gets weak, so the link may fail or slow down. To reach farther, put a switch partway, or use optical fiber.
By the numbers
| Item | Figure | Compared with |
|---|---|---|
| The first Ethernet (1973) | 2.94 Mb/s | The 800 Gb/s standard (2024) is about 270,000 times faster |
| Contents of one frame | Up to 1,500 bytes | About 12 microseconds to send at 1 Gb/s |
| Maximum network-cable length | 100 m | 30 m for Category 8 at 25 or 40 Gb/s |
| Power delivered to a device by PoE | Up to 71.3 W (802.3bt) | About 5.5 times the first standard, 802.3af |
"About 12 microseconds" is the time it takes to send 1,500 bytes (12,000 bits) at 1 billion bits per second. A microsecond is one millionth of a second.
Learn more
- Compare in tables: Hub vs Switch vs Router, Single-Mode vs Multimode Fiber, CAN vs LIN vs Automotive Ethernet
- Try the numbers: data transfer time calculator, CRC calculator
- Terms for the next step: Ethernet (key points of the standard), ARP (finds a MAC address from an IP address)
- Going further: VLAN (divides one switch into separate LANs), MTU (the largest size that can be sent in one go)
- Protecting your home network: how a router sorts traffic is explained in "What Is a Firewall?"