Inside the Pentagon office of Bob Taylor, head of the Information Processing Techniques Office at ARPA, three separate computer terminals occupied the room. Each terminal connected directly to a different funded mainframe: the System Development Corporation Q-32 in Santa Monica, Project Genie at the University of California, Berkeley, and Multics at MIT. Accessing each machine required a completely different set of user commands. In February 1966, Taylor asked ARPA Director Charles Herzfeld for funding to link these isolated systems into a single network. Herzfeld redirected $1 million from a ballistic missile defense program to initiate the project.
Architectural Trade-offs and Minicomputers
In January 1967, ARPA hired Larry Roberts to serve as program manager and turn Taylor's vision into a working network architecture. Roberts initially proposed connecting all mainframe host computers directly to one another. However, computer science researchers resisted the idea, reluctant to surrender valuable processing capacity on their primary mainframes to manage complex network traffic overhead.
Following an ARPA technical design session in April 1967, computer designer Wesley Clark proposed a critical structural shift: using dedicated minicomputers as intermediate gateways at each location. According to ARPA history, these machines, dubbed Interface Message Processors (IMPs), would relieve host mainframes of routing duties. Roberts revised the plan to use IMPs and incorporated packet switching principles developed by Donald Davies and Paul Baran, raising planned line speeds from 2.4 kbit/s to 50 kbit/s.
Contracting BBN and the IMP Hardware
In mid-1968, ARPA issued a Request for Quotation to 140 potential bidders for the network hardware contract. Most computing companies dismissed the proposal as impractical, resulting in only 12 formal submissions. In January 1969, ARPA awarded the contract to Bolt Beranek & Newman (BBN) in Cambridge, Massachusetts.
Led by Frank Heart with theoretical network design by Bob Kahn, BBN had nine months to build the first functional hardware nodes. BBN constructed the IMPs using a ruggedized Honeywell DDP-516 minicomputer fitted with 24KB of expandable magnetic-core memory. Each IMP supported up to four local hosts and routed data between remote sites over leased 50 kbit/s telephone lines.
The First Transmission at Boelter Hall
The first connection test occurred on October 29, 1969, at 10:30 PM PST. In Boelter Hall room 3420 at UCLA, student programmer Charley Kline attempted to log into an SDS 940 mainframe at Stanford Research Institute (SRI) from UCLA’s SDS Sigma 7 machine. Kline typed the letters of the command "login" one by one over the line while talking with SRI programmer Bill Duvall. The host system crashed after receiving the first two letters, successfully transmitting only "lo". After Duvall adjusted machine parameters, Kline logged in successfully about an hour later.
By December 5, 1969, the full initial four-node network was operational, interconnecting UCLA, SRI, UC Santa Barbara, and the University of Utah. Establishing this hardware framework demonstrated that dedicated minicomputers using packet switching could bridge incompatible mainframe architectures. This physical topology established the standard for distributed network control, creating the structural foundation that eventually expanded into the modern global Internet.
AI-generated editorial illustration; not a documentary photograph.
Source: ARPANET
SIGNAL TITANS
Leave a Reply