From the terminal room next to his Pentagon office, Bob Taylor had a direct connection to several of the ARPA-funded computers around the country. Each terminal was connected to a single computer and Taylor needed to use a different login sequence and different commands on each mainframe. In 1966, it was the leading edge of computer networking, but Taylor was tired of changing seats and instructions every time he needed to communicate with another computer.

He composed an idea and walked to his boss's office, Charles Herzfeld, and gave him the pitch. Taylor explained the problem and described a vague solution about networking different computers together. Herzfeld liked the idea and said Taylor had one million dollars to make the idea work. When Taylor looked at his watch he noted that it only took twenty minutes to get the project funded.

One of the sayings at ARPA was "why don't we rely on the computer industry to do that?" instead of the government. So, Bob Taylor started writing a Request for Proposals titled "Cooperative Network of Time-Sharing Computers." He described the general idea, but he needed some help figuring out what they were asking contractors to do, exactly. The best person he knew who could help him was Larry Roberts, who was working at MIT's Lincoln Lab networking computers like the TX-2. Roberts had just built and tested the first transcontinental network between two computers, so he had as much experience as anyone in long-distance networks.

At first, Roberts had no interest in leaving MIT, but Taylor wouldn't take no for an answer. Since he was in charge of funding over half of the research at Lincoln Lab, he had some clout there. After over a year of asking, Charles Herzfeld called Roberts' boss and strongly suggested he help Roberts decide to take the job. The director of Lincoln Lab called Roberts into his office and made the suggestion that the position at ARPA might be a good career choice at that time. Roberts moved to ARPA in 1966 and began drafting the Request for Proposals that ARPA would send out to potential contractors.

At the next annual conference of ARPA-funded university projects, Roberts organized a meeting to talk about the project. Two important parts of the network were decided: that the network traffic between computers would be broken up into blocks (a packet-switched network), and that a separate computer would act as a gateway to the network for each node. This computer, named an Interface Message Processor (IMP), would be connected to the network and to a mainframe at the site. All the nodes would have nearly identical IMPs, creating a standard interface for the network between nodes.

Request for Proposals