The History of Ethernet

The History of Ethernet

Ethernet is the foundational technology for the vast majority of wired local area networks (LANs) and has become a dominant force in metropolitan and wide area networking. Its history is a chronicle of pragmatic engineering, fierce standardization battles, and relentless adaptation. What began as an experimental system to connect some of the world's first personal computers at a research lab grew into a global standard, scaling in speed by a factor of over a million in five decades. Its success was not preordained. Ethernet triumphed over technically sophisticated rivals through a combination of simplicity, low cost, and an evolutionary path that allowed it to adapt to new challenges.

The story of Ethernet starts with a problem of packet collisions in a Hawaiian radio network, moves to a coaxial cable in a Palo Alto laboratory, and expands to fill buildings with twisted-pair wiring. It then evolves into the fiber-optic backbone of data centers and global carrier infrastructure. The core principle of a shared medium with a simple access protocol, Carrier Sense Multiple Access with Collision Detection (CSMA/CD), defined its early years. However, the technology’s longevity is owed to its ability to shed this original design in favor of a switched, full-duplex architecture that eliminated collisions entirely, paving the way for the extreme speeds that power the modern internet.

From a 2.94 megabit-per-second prototype in 1973 to standardized 800 gigabit-per-second systems today, Ethernet's progression demonstrates a powerful cycle of innovation. Each increase in speed was driven by new application demands and enabled by advances in silicon, cabling, and signal processing. Its open standardization process, initially driven by a consortium and later by the Institute of Electrical and Electronics Engineers (IEEE), ensured interoperability and a competitive marketplace, making it the ubiquitous plumbing of digital communication.

Precursors and The ALOHAnet Inspiration (1968 to 1972)

Before local area networking became common, wide area networks (WANs) like the ARPANET^[1] demonstrated the power of packet switching for connecting disparate computer systems over long distances. However, connecting computers within a single building or campus presented a different set of challenges. The primary technical inspiration for Ethernet came from an innovative solution to this problem developed at the University of Hawaii.

In 1971, a team led by Norman Abramson^[2] developed ALOHAnet^[3], a pioneering computer networking system that used packet radio to connect terminals across the Hawaiian Islands. Because all terminals used the same radio frequency to communicate with a central computer, a key problem arose: what happens when two terminals transmit at the same time? Their signals would interfere with each other, resulting in a corrupted, unusable message. This event was termed a collision.

ALOHAnet's solution was elegant in its simplicity. A terminal would transmit its packet of data whenever it was ready. After transmitting, it would wait for a small acknowledgement packet from the receiving station. If this acknowledgement was not received within a short timeout period, the terminal would assume a collision had occurred. It would then wait a randomly determined amount of time and retransmit the packet. This random backoff was crucial to prevent the two colliding stations from immediately colliding again. This method, known as pure ALOHA, was a foundational concept in random-access protocols. While functional, it had a maximum theoretical channel utilization of only 18.4%^[4], as many transmissions were lost to collisions.

The Invention at Xerox PARC (1973 to 1979)

The next crucial step took place at the Xerox Palo Alto Research Center (PARC)^[5], a facility renowned for its groundbreaking contributions to computing. In the early 1970s, PARC engineers developed the Xerox Alto^[6], the first computer to integrate a bitmapped display, a graphical user interface (GUI), and a mouse. These were effectively the first personal computers, and researchers needed a way to connect them to each other and to shared resources, most notably the Xerox 9700^[7] laser printer. While IBM's 3800 was the first laser printer installed in data centers (1976), the first true commercial laser printer was the Xerox 9700, released in 1977. This followed Xerox's own pioneering non-impact printer, the Xerox 1200^[8], which used xerography without a laser and was released in 1973.

Robert Metcalfe^[9], a researcher whose PhD work at Harvard had focused on analyzing the performance of ALOHAnet, joined PARC in 1972. Tasked with creating the network for the Alto, he collaborated with David Boggs^[10], another researcher at the lab. They sought to adapt the random-access principles of ALOHAnet for a wired medium. On May 22, 1973, Metcalfe wrote a memo that outlined this new system, which he named "Ethernet," a reference to the disproven concept of the luminiferous aether as a passive, omnipresent medium for propagating electromagnetic waves.

CSMA/CD: The Core Innovation

Metcalfe and Boggs's key innovation was to improve upon the simple ALOHA protocol. They developed a more sophisticated access method called Carrier Sense Multiple Access with Collision Detection (CSMA/CD^[11]). This protocol added two critical rules:

1. Carrier Sense (CS): Before transmitting, a station must listen to the network medium (the "ether"). If it detects a carrier signal, meaning another station is already transmitting, it must wait until the medium is quiet. This "listen before talk" rule dramatically reduced the number of collisions compared to pure ALOHA.
2. Collision Detection (CD): A station must listen to the medium while it is transmitting. If the signal it hears on the line does not match the signal it is sending, it knows a collision has occurred. Upon detecting a collision, the station immediately stops transmitting and sends a brief "jam signal" to ensure all other stations are aware of the collision. It then waits a random amount of time, determined by a binary exponential backoff algorithm^[12], before attempting to transmit again.

The first prototype, known as the Alto Aloha Network, went live in 1973. It operated at a speed of 2.94 megabits per second (Mbps) over a thick coaxial cable backbone. This specific speed was not arbitrary; it was derived from the system clock of the Xerox Alto's processor. The concepts and experimental results were formally published by Metcalfe and Boggs in their seminal 1976 paper, "Ethernet: Distributed Packet Switching for Local Computer Networks^[13]."

Standardization and The DIX Consortium (1980 to 1985)

In 1979, Robert Metcalfe left Xerox to found 3Com^[14], a company focused on manufacturing computer networking equipment. Metcalfe understood that for Ethernet to succeed commercially, it needed to be an open, multi-vendor standard, not a proprietary Xerox technology. He successfully persuaded Digital Equipment Corporation (DEC)^[15], Intel^[16], and Xerox^[17] to collaborate on creating a public specification.

This "DIX" consortium published the first version of the Ethernet standard, known as the "Ethernet Blue Book," on September 30, 1980^[18]. It specified a 10 Mbps system running over a thick, yellow coaxial cable, a design that would later be formalized as 10BASE5. The DIX group then submitted its specification to the Institute of Electrical and Electronics Engineers (IEEE)^[19] for consideration as an official industry standard.

Within the IEEE's Project 802 LAN/MAN Standards Committee, Ethernet faced stiff competition. IBM was heavily promoting its own technology, Token Ring^[20] (later standardized as IEEE 802.5), which used a deterministic token-passing scheme that prevented collisions entirely. Proponents of Token Ring argued that its deterministic nature made it more reliable for time-sensitive applications. After considerable debate, the IEEE decided to create separate standards for the different access methods.

The Ethernet-based standard was designated IEEE 802.3. The first version, IEEE 802.3-1983^[21], was heavily based on the DIX specification but introduced one small, significant change to the frame structure. The DIX frame used a field after the source and destination MAC addresses to indicate the protocol type (e.g., IP, AppleTalk). The IEEE 802.3 frame changed this to a length field. This created a minor incompatibility between the original "Ethernet II" (DIX) frames and the new "802.3" frames. In practice, most network software was quickly adapted to check the value in this field and interpret it as either a type or a length, allowing both frame formats to coexist on the same network.

Thicknet and Thinnet

The first official standard, 10BASE5^[22], used a thick, rigid coaxial cable often called "Thicknet" or "yellow cable." It could support segments up to 500 meters long. Transceivers were attached to the cable using a device called a vampire tap, which pierced the cable's insulation to make contact with the inner conductor. This topology was robust but expensive and difficult to install.

To make Ethernet more accessible, a new standard was developed. In 1985, the IEEE published the 10BASE2^[23] standard, known as "Thinnet" or "Cheapernet." It used a more flexible and less expensive coaxial cable (RG-58) similar to that used for cable television. Computers were connected in a daisy-chain fashion using BNC T-connectors. While cheaper, this bus topology had a significant weakness: a single loose connector or break in the cable could bring down the entire network segment, making troubleshooting difficult.

The Rise of Twisted Pair and The Switch (1986 to 1995)

The physical and logistical problems of coaxial bus topologies were a major barrier to widespread Ethernet adoption. The solution came from a seemingly mundane source: the unshielded twisted-pair (UTP) wiring already installed in many office buildings for telephone systems.

A startup named SynOptics Communications pioneered the concept of running Ethernet over UTP. The key was to abandon the bus topology in favor of a physical star topology. In this design, each computer was connected via its own dedicated cable to a central device called a hub or concentrator. The hub, a simple electronic device, received a signal from one port and repeated it out to all other ports. Electrically, the network still behaved like a single shared bus, a single collision domain, but the physical wiring was far more resilient. A problem with one cable would only affect one computer, not the entire network.

This approach was standardized in 1990 as IEEE 802.3i^[24], which defined the 10BASE-T^[25] specification. 10BASE-T was a transformative technology. It was inexpensive, easy to install using standard telephone-style connectors (RJ45), and simple to troubleshoot. This combination made Ethernet the dominant choice for new LAN installations, rapidly displacing Token Ring and other rivals.

The Ethernet Switch

While hubs made Ethernet more reliable, they did not solve the performance problem of collisions. As more devices were added to a hub-based network, the shared 10 Mbps bandwidth became a bottleneck. The solution was the Ethernet switch, a device invented by the company Kalpana^[26] in 1990.

A switch is a far more intelligent device than a hub. While a hub is a Layer 1 (Physical Layer) device that simply repeats electrical signals, a switch is a Layer 2 (Data Link Layer) device. It inspects the destination MAC address^[27] of each incoming frame. The switch maintains a table of which MAC addresses are connected to which of its ports. It then forwards the frame only to the port connected to the destination device, rather than broadcasting it everywhere like a hub.

This had two profound consequences. First, it dramatically reduced network congestion. Second, because each port on a switch is an isolated segment, it effectively becomes its own collision domain. When a computer is connected to a dedicated switch port, collisions are impossible. This allowed for the introduction of full-duplex operation, where a device can send and receive data simultaneously, effectively doubling the potential throughput of a 10 Mbps link to 20 Mbps. With the advent of affordable switches, the original CSMA/CD protocol, while still part of the standard for backward compatibility with hubs, became a historical footnote in most modern networks.

The Need for Speed: Fast and Gigabit Ethernet (1995 to 2002)

With Ethernet's dominance in the LAN established, the engineering focus shifted to increasing its speed to support more demanding applications and growing numbers of users. The next logical step was a tenfold increase to 100 Mbps.

In the mid-1990s, two main proposals competed to become the 100 Mbps standard. One was 100VG-AnyLAN, a different protocol supported by HP and AT&T. The other, promoted by a vendor consortium including 3Com and Intel, was 100BASE-T^[28], which retained the CSMA/CD mechanism for backward compatibility and was a more direct evolution of 10BASE-T. Due to its simplicity and clear migration path, 100BASE-T won the standards battle and was formalized as IEEE 802.3u^[29] in 1995. The most popular variant, 100BASE-TX, required higher-quality Category 5^[30] UTP cabling.

Almost immediately after Fast Ethernet was standardized, work began on the next tenfold leap: Gigabit Ethernet. The primary drivers were the need for faster connections for servers and for high-speed backbone links between switches. The technical challenge was immense: pushing 1,000 Mbps over the same Category 5 UTP cabling that many organizations had just installed for 100BASE-TX.

The solution, standardized in 1999 as IEEE 802.3ab^[31], was 1000BASE-T^[32]. It was a marvel of signal processing. To achieve the required data rate, it used all four pairs of wires in the cable simultaneously and in full-duplex. This required sophisticated hybrid circuits to separate transmitted and received signals on the same wires, as well as powerful digital signal processors (DSPs) to cancel out echo and crosstalk. Alongside the copper standard, fiber optic versions like 1000BASE-SX and 1000BASE-LX were standardized for longer-distance backbone applications.

Ethernet Beyond the LAN (2002 to Present)

By the early 2000s, Ethernet was the undisputed king of the LAN. Its next conquest was the carrier network, a domain traditionally dominated by proprietary, circuit-switched technologies like SONET/SDH. The standardization of 10 Gigabit Ethernet (10GbE^[33]) in IEEE 802.3ae-2002^[34] was the turning point. This standard initially focused exclusively on fiber optics and notably removed the CSMA/CD protocol from the specification for these speeds, as it was only intended for full-duplex, switched links.

Ethernet's packet-based architecture and commodity hardware made it a much simpler and more cost-effective alternative to SONET/SDH for building Metropolitan Area Networks (MANs) and Wide Area Networks (WANs). This gave rise to the concept of Metro Ethernet or Carrier Ethernet. To make Ethernet suitable for carrier-grade services, the IEEE developed a suite of enhancements to provide features like service-level agreements and network resilience. These included standards for VLAN stacking (Q-in-Q, IEEE 802.1ad^[35]) and Provider Backbone Bridges (IEEE 802.1ah^[36]), which allowed service providers to build scalable, multi-tenant networks using Ethernet as the underlying transport.

Parallel to this expansion in speed and scope, another innovation greatly expanded Ethernet's utility. Power over Ethernet (PoE)^[37] allows for low-voltage DC power to be delivered to devices over the same UTP cable used for data. The first standard, IEEE 802.3af^[38], was released in 2003 and could deliver up to 15.4 watts of power. This was a catalyst for the growth of Voice over IP (VoIP) phones, wireless access points, and IP security cameras, as it eliminated the need for a separate power outlet. Subsequent standards like PoE+ (IEEE 802.3at-2009^[39]) and PoE++ (IEEE 802.3bt-2018^[40]) have increased the available power to support more demanding devices.

The Terabit Era and The Future (2010 to Today)

The rise of cloud computing, social media, video streaming, and big data created an insatiable demand for network bandwidth, particularly within the massive data centers that power these services. The pace of Ethernet speed increases accelerated dramatically.

In 2010, the IEEE standardized both 40 GbE and 100 GbE in a single stroke with IEEE 802.3ba^[41]. These standards were built by bonding multiple 10 Gbps or 25 Gbps lanes together. A key pivot occurred with the introduction of a native 25 GbE standard, IEEE 802.3by^[42], in 2016. Data center operators realized that a single 25 Gbps lane for server connections was a more cost-effective upgrade from 10 GbE than the four-lane 40 GbE. This also provided a more efficient building block for 100 GbE (4 x 25 Gbps).

This multi-lane approach, combined with more advanced modulation techniques like four-level Pulse Amplitude Modulation (PAM4^[43]), which encodes two bits per symbol instead of one, has fueled further speed increases. The IEEE published standards for 200 GbE and 400 GbE (IEEE 802.3bs^[44]) in 2017. As of 2024, the work of the IEEE 802.3df task force concluded with the approval of IEEE Std 802.3df-2024 on February 16, 2024, which includes specifications for 400 Gb/s and 800 Gb/s operation, while a separate project, IEEE P802.3dj, is developing the 1.6 Tb/s standard, driven by the needs of artificial intelligence and machine learning workloads.

Today, Ethernet is the universal foundation of wired networking. Its original CSMA/CD protocol is a historical artifact, but its frame format and MAC addressing scheme persist across all speeds. Looking forward, the Ethernet Alliance^[45] roadmaps point toward even higher speeds for data centers. At the same time, a new frontier is opening at the other end of the spectrum with Single-Pair Ethernet (SPE). Standards like 10BASE-T1L^[46] are designed to replace legacy fieldbus protocols in industrial automation, automotive systems, and building management. This effort aims to bring the unified, IP-based architecture of Ethernet to the very edge of the network, completing a fifty-year journey from a single lab cable to a technology that connects nearly everything.