ARPANET: The First Network
ARPANET was the first packet-switching network and the direct ancestor of today's internet. Created by ARPA (Advanced Research Projects Agency) in 1969, it proved that computers could communicate without a central hub—and survive a nuclear attack.
The Problem ARPANET Solved
In the 1960s, telephone networks used circuit switching—creating a dedicated physical path between two points for the duration of a call. This was inefficient for computer communication, which comes in bursts. J.C.R. Licklider and Robert Taylor at ARPA envisioned a network where computers could share resources and communicate via packet switching.
Packet Switching: The Key Innovation
Unlike circuit switching, packet switching breaks data into small chunks (packets) that travel independently through the network. Each packet contains its destination address and can take different routes. This made the network robust—if one path failed, packets could be rerouted automatically.
The IMP: Interface Message Processor
Each ARPANET node was connected to an IMP (Interface Message Processor) — a specialized packet-switching minicomputer built by Honeywell (DDP-516) under contract from BBN (Bolt, Beranek and Newman). The IMP design was specified inBBN Report 1822("A Host IMP Protocol"), which defined how hosts connected to the IMP and how IMPs forwarded packets.
The DDP-516 was a 16-bit minicomputer with 40KB of core memory and dual disk drives. Each IMP had four serial interfaces for connecting to other IMPs (50 kbps dedicated lines) and eight asynchronous ports for host computers. The IMP software handled routing, flow control, error detection, and retransmission — tasks now performed by network interface cards and routers. IMPs communicated using a custom protocol defined in BBN Report 1822, with 1024-bit packets (including 96-bit headers containing destination IMP, host, and link numbers).
The First Four Nodes
On October 29, 1969, the first message was sent from UCLA to Stanford Research Institute (SRI). The planned message was "LOGIN" but the system crashed after "LO"—resulting in the world's first network transmission being "LO". By the end of the month, four nodes were operational:
- UCLA (Sept 1): UCLA's Sigma 7 host computer, connected to IMP #1. Professor Leonard Kleinrock's team managed the Network Measurement Center.
- SRI (Oct 1): Stanford Research Institute's SDS 940 (sometimes referenced as Sigma), IMP #2. Doug Engelbart's NLS system ran here — the same system from "The Mother of All Demos."
- UCSB (Oct 15): University of California Santa Barbara, IMP #3. Connected to their IBM 360/75 mainframe running the Culler-Fried Interactive Mathematics Center.
- University of Utah (Dec): Utah's DEC PDP-10, IMP #4. Ivan Sutherland's graphics research group ran early CAD and graphics applications over the network.
The network grew rapidly: 15 nodes by December 1969, 37 by 1971, and over 60 by 1973.
NCP: The First Network Protocol
ARPANET initially used NCP (Network Control Program) to coordinate communication. NCP operated at the host level, providing simplex (one-way) byte-stream connections between processes on different hosts. It used 8-bit port numbers, limiting each host to a maximum of 256 simultaneous connections. NCP handled connection setup, flow control (using IMP-level "allocate" and "return" messages), and teardown — but left error recovery and reliability to the application layer.
NCP was later replaced by TCP/IP in 1983—the event called "Flag Day" that marked the birth of the modern internet.
Email Changes Everything
While ARPANET was designed for terminal-to-terminal communication, Ray Tomlinson's 1971 invention of email transformed it into a communications medium. Tomlinson chose the @ symbol to separate the user name from the host name in addresses (e.g., user@host) — a convention that persists in every email address today. His first emails were sent between two machines in the same room at BBN, but the concept quickly spread. By 1973, email accounted for 75% of all ARPANET traffic.
Flag Day: The TCP/IP Transition
On January 1, 1983, ARPANET switched from NCP to TCP/IP in a coordinated event known as "Flag Day." This was mandated by ARPA and required all hosts to run TCP/IP software simultaneously. The transition was carefully planned over months, with a "crossover" period where both protocols were available. After Flag Day, TCP/IP became the standard protocol suite. This date is often considered the true birth of the internet as we know it.