Ethernet: The Protocol That Wired the World and Won't Give Up
If you're reading this on a desktop computer, or on a laptop with a cable plugged into it, or on a device in a data center, there's a very good chance your network traffic is running on Ethernet. And if that cable is attached to the wall, the signals traveling through it are governed by a standard that's been essentially the same for 50 years — continuously updated and improved, but fundamentally the same clever idea that a researcher at Xerox PARC named Robert Metcalfe came up with in 1973.
Ethernet. The protocol that wired the world. Let's talk about it.
The Original Problem
In the early 1970s, Xerox PARC was a hotbed of computer science innovation — the team there invented things like the graphical user interface, the mouse, and laser printing, many of which would eventually appear in Apple and Microsoft products. They also had multiple computers they wanted to connect together.
The challenge: how do you let multiple computers share the same physical cable? If two computers transmit at the same time, their signals collide and both are garbled. You need some way to coordinate access to the shared medium.
Robert Metcalfe's solution was CSMA/CD (Carrier Sense Multiple Access with Collision Detection) — the algorithm at the heart of original Ethernet:
The name "Ethernet" came from the old physics concept of "luminiferous ether" — the hypothetical medium through which light supposedly traveled. Metcalfe used it as a metaphor for the cable medium through which data travels.
Ethernet Spreads: From Coax to Twisted Pair
Original Ethernet used thick coaxial cable ("thicknet") at 10 Mbps. Connecting to it required drilling into the cable and attaching a vampire tap — literally a spike that pierced through to the conductor. Not user-friendly.
Later came "thinnet" — thinner coaxial cable that was easier to work with, using BNC connectors. A chain of computers connected in a bus topology, each tapping into the cable.
The real breakthrough came with 10BASE-T in 1990: Ethernet over twisted-pair copper cable (the same basic cable used for telephone lines) connected to a central hub. This was the beginning of the star topology that dominates to this day — each device has its own cable running to a central device. It was dramatically easier to manage. If one cable failed, only one device was affected. You didn't have to tap a bus. You could run cables through walls to a wiring closet. Building wiring suddenly made sense.
The naming convention became standardized: [speed]BASE-[medium]. So 10BASE-T means 10 Mbps, baseband signaling, over twisted-pair. 100BASE-TX is Fast Ethernet at 100 Mbps. 1000BASE-T is Gigabit Ethernet over copper. 10GBASE-SR is 10 Gigabit over short-range fiber.
The Speed Race
Ethernet speeds have increased by roughly 10x every several years:
Each generation maintained backward compatibility — your 100 Mbps device can still connect to a Gigabit switch, it just negotiates down to 100 Mbps. This compatibility has been crucial to Ethernet's longevity.
The Ethernet Frame
Ethernet is a Layer 2 (Data Link) protocol. The fundamental unit of data in Ethernet is the frame — a structured package of bytes that contains the actual payload data along with the addressing and control information needed to deliver it across a local network.
A standard Ethernet frame has this structure:
Preamble (7 bytes): A specific pattern of alternating 1s and 0s that helps receiving hardware synchronize to the incoming signal.
Start Frame Delimiter (1 byte): Signals the start of the actual frame content.
Destination MAC Address (6 bytes): The MAC address of the intended recipient.
Source MAC Address (6 bytes): The MAC address of the sender.
EtherType / Length (2 bytes): Either indicates which Layer 3 protocol is encapsulated (EtherType, like 0x0800 for IPv4, 0x0806 for ARP) or the length of the payload.
Payload (46–1500 bytes): The actual data being carried (usually an IP packet).
Frame Check Sequence (4 bytes): A CRC (Cyclic Redundancy Check) value that allows the receiver to verify the frame hasn't been corrupted in transit.
The minimum frame size of 64 bytes (payload minimum of 46 bytes plus headers) was chosen to ensure collision detection works reliably at 10 Mbps on the maximum allowed cable length. The maximum payload of 1500 bytes is the standard MTU (Maximum Transmission Unit) that most networks use today.
Jumbo Frames: In data centers, many networks support jumbo frames with payloads up to 9000 bytes. Larger frames reduce the overhead of framing per byte of payload and reduce CPU interrupts for high-throughput applications. Jumbo frames require all network devices in the path to support them.
CSMA/CD and Its Obsolescence
CSMA/CD — the original collision detection mechanism — was essential for original Ethernet's shared-bus topology. But as we discussed in the switches section, modern Ethernet networks use switches rather than hubs, and each device has a private full-duplex connection to the switch. With no shared collision domain, there are no collisions, and CSMA/CD is irrelevant.
In fact, modern full-duplex Ethernet — where both sides can transmit simultaneously — technically doesn't need CSMA/CD at all. The 802.3 standard still defines CSMA/CD for backward compatibility, but every modern wired connection runs full-duplex and never invokes it.
Auto-Negotiation
When you plug an Ethernet cable between two devices, they don't just start transmitting at some assumed speed. They first go through auto-negotiation — a process defined in IEEE 802.3ab — where they exchange information about their capabilities and agree on the highest common speed and duplex mode.
This is what makes your Gigabit NIC automatically negotiate 1 Gbps with a Gigabit switch, or fall back to 100 Mbps with an older switch. And it's why "duplex mismatch" — when one side is set to full-duplex manually and the other auto-negotiates to half-duplex — is a classic source of terrible Ethernet performance. Always use auto-negotiation on both ends.
Power over Ethernet (PoE)
PoE is a technology that delivers electrical power through Ethernet cables alongside data. A PoE switch can power IP cameras, WiFi access points, VoIP phones, and IoT devices through the same cable that carries the network connection — no separate power supply needed.
The IEEE 802.3af standard (PoE) delivers up to 15.4W per port. 802.3at (PoE+) delivers up to 30W. 802.3bt (PoE++) delivers up to 100W — enough for a monitor or a thin client PC.
PoE has become enormously important for simplifying the deployment of wireless access points, which now typically get all their power and connectivity through a single cable run. It's also critical for IoT deployments where running separate power to every sensor or camera would be impractical.
Ethernet in Data Centers: Spine-Leaf Architecture
Traditional enterprise networks were built in a hierarchical three-tier model: access switches at the edge, distribution switches in the middle, core switches at the top. This works well for typical office traffic patterns where most communication is between internal clients and central servers.
But modern data centers have different traffic patterns. Applications are distributed across many servers that need to communicate intensively with each other (east-west traffic), not just with external clients (north-south traffic). The old hierarchical model doesn't handle this efficiently.
The modern data center architecture is the spine-leaf (or Clos) network:
This creates a flat, non-blocking fabric where any server can communicate with any other server with exactly two hops (source leaf → spine → destination leaf) and with predictable, consistent latency. The bandwidth between any two servers is the same regardless of which servers they are. This uniformity is ideal for distributed applications.
High-speed Ethernet (25 Gbps, 100 Gbps, 400 Gbps) runs on the connections between spines and leaves, while 10 or 25 Gbps typically connects to individual servers.
The Cable Standards: Categories and Capabilities
The "twisted pair" in twisted-pair Ethernet refers to the way the individual copper wires inside the cable are twisted together in pairs. The twisting dramatically reduces electromagnetic interference by canceling out noise (what's picked up on one twist is cancelled by the next).
Cable categories determine what speeds the cable can support:
Cat5e (Enhanced Category 5): Supports Gigabit Ethernet (1000BASE-T) at up to 100 meters. Very widely deployed. Still fine for most applications.
Cat6: Supports 1 Gbps at 100m and 10 Gbps up to 55m. Thicker than Cat5e due to a plastic separator between pairs, reducing crosstalk.
Cat6A (Augmented Category 6): Supports 10 Gbps at 100m. Required for 10 Gbps over full cable runs. Significantly thicker and stiffer than Cat6.
Cat7 and Cat8: Higher performance categories for specialized applications. Cat8 supports 25/40 Gbps at up to 30m, designed for data center use.
For new installations today, Cat6A is generally recommended — it supports 10 Gbps at full length and is future-proof for several generations of Ethernet speed improvements.
Fiber Optic Ethernet
Copper Ethernet has limits in terms of distance. Gigabit Ethernet runs 100m on copper. 10 Gbps runs 55m on Cat6. For longer distances — between buildings, between floors, between data center rows — fiber optic cables are used.
Fiber sends data as light pulses through glass or plastic fibers, eliminating electromagnetic interference entirely and allowing far longer runs.
Single-mode fiber (SMF): A very thin core (9 microns) that carries light in a single mode, allowing very long distances — tens or hundreds of kilometers. Used for campus backbones, metro networks, and long-haul telecommunications.
Multi-mode fiber (MMF): A larger core (50 or 62.5 microns) that's cheaper and easier to terminate, but limited to shorter distances (typically up to 400m for 10G, 100m or less for 100G). Used for within-building and short data-center runs.
The connectors used with fiber (LC, SC, MPO) are different from the RJ45 connectors used for copper — they're more fragile and require more careful handling.
Ethernet's Remarkable Longevity
Why has Ethernet lasted 50 years while other networking technologies have come and gone? Token Ring, ATM, FDDI — all had their adherents and have faded.
Several factors:
The prognosis for Ethernet is more of the same: continued speed increases (terabit Ethernet is coming), expansion into new domains (automotive Ethernet for in-vehicle networks, industrial Ethernet for factory automation), and a continuing role as the backbone of nearly all serious wired networking on earth.
The cable running from your router to your computer is doing what Robert Metcalfe envisioned in 1973. That's a remarkable technological legacy — the kind that shapes the world so thoroughly that people stop noticing it's there at all.