3 Internet History: ARPANET and TCP/IP

3.10.1 The Problem of Survivable Communication

In the early 1960s, military planners faced a terrifying question: how do you maintain communication after a nuclear attack? The existing telephone system was centralized — if a few key switching centers were destroyed, the entire network could collapse.

Those switching centers were physical places, not just boxes in a diagram. A map of central-office sites shows how dense the public telephone network had become in cities and suburbs, while long-distance trunk maps show high-capacity routes concentrating traffic between regions. Packet switching did not merely replace a voice call with a data packet; it changed the geography of resilience. The design goal shifted from holding one reserved path through specific offices to letting many packets use available links and route around damaged or congested parts of the network.

Three researchers, working independently on three continents, arrived at the same revolutionary idea:

3.10.2 Circuit Switching vs Packet Switching

The telephone network used circuit switching — when you made a call, a dedicated physical circuit was established from your phone to the other person’s phone. This circuit was yours alone for the entire call, even during silences.

Packet switching was radically different:

1. Break the message into small packets (typically 1,000-1,500 bytes each)
2. Each packet travels independently — it can take any available route
3. Packets are reassembled at the destination — the receiver puts them back in order
4. The network is shared — many conversations use the same wires simultaneously

This is like the difference between hiring a private taxi (circuit switching) versus using a postal service (packet switching). The postal service is far more efficient because one mail truck carries hundreds of letters simultaneously.

Checkpoint: Packet Switching

You now know:

• The survivability problem was about avoiding a few centralized telephone switching centers.
• Packet switching breaks a message into independently routed packets, typically 1,000-1,500 bytes each.
• Baran’s quantitative example shows why redundancy matters: with 3 paths at 90% reliability, delivery rises to 99.9%.

3.11.1 The First Message

On October 29, 1969, at 10:30 PM, a 21-year-old UCLA graduate student named Charley Kline sat at an SDS Sigma 7 computer and attempted to log in to a computer at Stanford Research Institute (SRI), 350 miles away. He typed “L”… then “O”… then “G”… and the system crashed.

The first message ever sent over ARPANET was “LO” — an accidental abbreviation that would be poetic only in hindsight. An hour later, they tried again and successfully transmitted “LOGIN.”

The four original ARPANET nodes were:

The “IMP” (Interface Message Processor) was the ancestor of today’s router — a Honeywell DDP-516 minicomputer that weighed 900 pounds and cost $82,200 (about $680,000 in today’s money).

3.11.2 Growing the Network

ARPANET grew steadily through the 1970s:

• 1970: ALOHAnet in Hawaii pioneers wireless packet networking (ancestor of Wi-Fi)
• 1971: Ray Tomlinson sends the first email and picks ”@” to separate user from machine
• 1972: ARPANET demonstrated publicly at ICCC conference — 40 machines connected
• 1973: First international connections — University College London (UK) and NORSAR (Norwegian Seismic Array, Norway)
• 1974: Vint Cerf and Bob Kahn publish “A Protocol for Packet Network Intercommunication” — the TCP paper

That growth is easier to understand as a map than as a slogan. In December 1969, ARPANET was a four-node western U.S. experiment linking UCLA, SRI, UC Santa Barbara, and the University of Utah. By the mid-1970s, the map had spread across research sites, military labs, and international links. Later host-count curves make the same point on a log scale: the Internet grew by orders of magnitude because new hosts could join a shared packet network without redesigning the whole system around each pair of endpoints.

3.11.3 The Fathers of the Internet: Cerf and Kahn

Vint Cerf and Bob Kahn are often called the “Fathers of the Internet” for their 1974 paper defining TCP (Transmission Control Protocol). Their key insight was creating a universal protocol that could connect different types of networks — ARPANET, radio networks, satellite networks — into one seamless “inter-network” (which became shortened to “Internet”).

Their original TCP combined what we now know as TCP and IP into a single protocol. In 1978, they split it into two layers:

• IP (Internet Protocol): Handles addressing and routing — getting packets to the right destination
• TCP (Transmission Control Protocol): Handles reliability — making sure all packets arrive and are in order

This split was crucial. It meant that different applications could choose different transport protocols. Today, IoT devices often use UDP (User Datagram Protocol) instead of TCP because UDP is lighter weight — a choice made possible by this architectural decision from 1978.

3.12.1 Flag Day: January 1, 1983

The true birthday of the Internet is January 1, 1983 — “Flag Day” — when every machine on ARPANET simultaneously switched from the old NCP (Network Control Protocol) to TCP/IP. There was no gradual migration. At midnight, NCP stopped working and TCP/IP was the only option.

This was a massive coordinated effort. Every host on the network had to be ready. Buttons reading “I have TCP/IP — ask me how” were distributed at networking conferences. The transition went remarkably smoothly.

Quick Check: Test your understanding of the Internet’s “birthday”:

3.12.2 Domain Name System (1983-1984)

Before DNS, you had to know a computer’s numerical IP address to reach it. Paul Mockapetris invented the Domain Name System in 1983, allowing human-readable names like “usc.edu” instead of “128.125.253.136.”

For IoT deployments, DNS is not just a convenience. Resolver choice, cache TTLs, CNAME chains, and failover records determine how quickly devices recover when cloud endpoints move or regional services fail. A resilient design records which names the device resolves, how long it caches answers, what happens when DNS returns NXDOMAIN, and whether provisioning can update endpoint names without reflashing firmware.

Try the resolution workbench below to see what a resolver actually does with one of those names: trace a recursive client query through iterative root, TLD, and authoritative referrals, watch a TTL cache hit skip that trip entirely, and follow a CNAME chain or NXDOMAIN response.

The original top-level domains were:

The first .com domain ever registered was symbolics.com on March 15, 1985.

3.12.3 Key Developments

• 1983: BSD Unix includes TCP/IP — making it free and widely available
• 1984: ARPANET splits into ARPANET (research) and MILNET (military)
• 1986: NSFNET backbone created — connecting 5 supercomputing centers at 56 Kbps
• 1988: The Morris Worm — first major Internet security incident, infecting ~6,000 of the 60,000 connected computers
• 1989: Tim Berners-Lee proposes the World Wide Web at CERN