From LAN to the Internet: Token Ring, Ethernet, and the Protocols That Won

Before the LAN: one cable, one conversation

Early computer networking was built around point-to-point or multidrop links between a mainframe and its terminals - reliable, but expensive to wire and impossible to scale past a handful of devices. As offices filled with independent minicomputers and, later, personal computers, the real problem shifted: how do dozens of machines share one physical cable without their signals colliding into unreadable noise?

Two very different answers emerged in the 1970s and 1980s, and for over a decade neither one was the obvious winner.

Two philosophies: token passing vs. listen-and-retry

IBM's Token Ring, standardized as IEEE 802.5, solved the collision problem by removing the possibility of collision entirely. A single small 'token' frame circulates continuously around a logical ring; a station may only transmit when it is holding the token, guaranteeing that exactly one device speaks at a time. It is deterministic and predictable under heavy load - properties IBM's mainframe-oriented enterprise customers valued highly.

Ethernet took the opposite bet. Its CSMA/CD access method (Carrier Sense Multiple Access with Collision Detection) simply lets any station transmit whenever it believes the wire is idle, and listens for a collision; if one occurs, both senders back off for a random interval and retry. It is probabilistic rather than guaranteed, but radically simpler and cheaper to implement in hardware - and under the light-to-moderate traffic loads of a typical office, it performed just as well in practice as the more elaborate token scheme.

The protocols that made packets talk

A shared cable only solves how frames move between two adjacent machines. Getting a message from a computer in Los Angeles to one in Boston, across networks owned by entirely different organizations, needed a second, independent layer of addressing and routing - which is exactly what the ARPANET research program was built to solve. Vint Cerf and Bob Kahn's 1974 paper introduced the idea of splitting that job into two protocols: IP, responsible for addressing packets and getting them from network to network, and TCP, responsible for reassembling them reliably and in order at the far end.

Both were formalized as RFC 791 and RFC 793 in September 1981, and on January 1, 1983 - a date engineers still refer to as 'flag day' - every host still connected to ARPANET was cut over from the older NCP protocol to TCP/IP simultaneously, essentially overnight. It is one of the largest coordinated protocol migrations in computing history, and it is the direct reason IP addressing and TCP's three-way handshake are still exactly how this toolkit's Ping, Traceroute, and every other tool talk to the network today.

The RFC (Request for Comments) process itself, still run by the IETF, deserves credit here too: unlike Token Ring's IBM-controlled standardization, RFCs are open documents anyone can propose, argue about in public, and implement - the process's own famous motto is 'rough consensus and running code'. That openness is a large part of why TCP/IP, rather than any of several competing proprietary networking stacks of the era (including IBM's own SNA), became the protocol every vendor eventually converged on.

Why Ethernet ultimately won

Through the mid-1980s, Token Ring and Ethernet coexisted as genuinely competing enterprise choices, with Token Ring often preferred in IBM-centric shops for its predictable behavior under load. The balance tipped decisively over the following decade for reasons that had less to do with raw technical merit and more to do with economics and pace of innovation: Ethernet's move to twisted-pair cabling with 10BASE-T in 1990 made wiring dramatically cheaper and easier than Token Ring's shielded cabling and specialized connectors, and the arrival of switching (replacing shared hubs with a dedicated point-to-point link per port, eliminating collisions altogether) removed CSMA/CD's biggest theoretical weakness.

The final blow was speed. Fast Ethernet's 100 Mbps in 1995 was already six times Token Ring's 16 Mbps ceiling, and by the time Gigabit Ethernet arrived in 1999 running over the exact same cabling already installed everywhere, there was no remaining reason to choose the more expensive, proprietary alternative. IBM itself quietly shifted its own recommendations to Ethernet during the 1990s; Token Ring survives today mostly in legacy industrial control systems, not new deployments.