How TCP/IP Won the Network Wars

The Department of Defense set the deadline years in advance: at midnight on what engineers called Flag Day, January 1, 1983, the ARPANET would stop speaking the old NCP protocol and start speaking TCP/IP. About 400 hosts had to be converted. T-shirts printed for the occasion read "I survived the TCP/IP transition."

The protocol itself was nearly a decade old by then. Vint Cerf and Bob Kahn laid out the design in a 1974 paper at IEEE. Their core insight was that a network of networks needed two separate jobs done by two separate layers. IP — the Internet Protocol — moves packets between machines without caring whether they arrive. TCP — Transmission Control Protocol — sits on top and handles the bookkeeping: reordering, retransmission, congestion control.

Splitting routing from reliability made the system astonishingly portable. Any underlying medium that can carry a packet works: copper wire, fiber, radio, satellite, even pigeons (a 2001 RFC formalized the joke). New applications can be invented above TCP without touching the network underneath. Email, the web, video calls, and ransomware all share the same plumbing.

The alternative was a stack called OSI, designed by committee at the International Organization for Standardization through the late 1970s. OSI was technically more complete and politically backed by European telecoms and the U.S. government. By the early 1990s, free TCP/IP implementations were already shipping in BSD Unix and being adopted by universities. OSI lost to running code.

The internet you use is the cumulative weight of one decision made in 1983, and the boring layered design that made it stick.