10.Request for Proposals

Larry Roberts finished writing the Request for Proposals and sent it to 140 potential contractors in the summer of 1968. After a few months, about twelve came back to ARPA, including BBN's proposal. Two of the largest computer companies, IBM and Control Data Corporation (CDC), declined to make a bid, confident that packet-switching was an unworthy endeavor.

Roberts cut the number he considered appropriate down to four, including BBN and Raytheon. Raytheon looked like the early leader in the competition for the contract. They had more resources than BBN and they claimed they could make the network faster than the proposal required. However, BBN made the same claim and they supplied the details on how they were going to do it. BBN had spent several months and over $100,000 writing the proposal and Roberts felt BBN's proposal was a better plan.

BBN received word that they won the contract, and they were congratulated via telegram by Massachusetts' Senator Edward Kennedy for the winning the contract to build an "Interfaith Message Processor."

On the first day of 1969, Frank Heart collected his team together and started working on designing and programming the IMP. They chose the Honeywell DDP-516 for the computer they would modify into the IMP. The DDP-516 was one of the most powerful minicomputers on the market, and Heart liked it because it was built to military specifications - reinforced body and i-bolts on top.

The team's primary members included programmers Will Crowther and Dave Walden, BBN's star debugger Bernie Cosell, and Severo Ornstein, the geologist turned computer hardware specialist. Bob Kahn volunteered to write a specification to send to the participating centers detailing how to connect their computers to the IMP. They had about eight months to deliver the first IMP to UCLA on Labor Day.

Crowther and Walden spent several months writing code to send and receive packets over the network. They had to write the code in assembly language (what a computer can understand) and everything had to fit into the Honeywell's 12k of memory. To make the process more efficient, they wrote an assembler on BBN's PDP computer and transfered the compiled application via paper tape to a prototype Honeywell for testing each version.

When the first modified IMP (IMP-0) was delivered to BBN, Ben Barker (the young engineer assigned to testing it) discovered that the modifications were all wrong. In the 1960's changing the computer's configuration meant unwrapping and wrapping hundreds of tight bundles of wires. Barker spent a few months of long days reconfiguring the Honeywell.

IMP-1 was delivered to BBN just two weeks before Labor Day. They had sent instructions on the changes Barker made to the first IMP to Honeywell and they expected everything to work this time. However, when Barker powered the Honeywell up, nothing worked. He opened it up and discovered the same configuration the first IMP started with. However, this time he had detailed instructions on what to do, and he started working right away. He finished in just a few days, but he found a new problem.

When they tested the IMP it would consistently work for a while and then crash for no apparent reason. Most of the time it would go a day or two between crashes, but the IMP was supposed to work all the time or the ARPAnet wouldn't be practical. After a few days, Barker was convinced it was a synchronizer problem, an occasional mistiming in the CPU. It was one of the worst problems for a computer, and one of the hardest to fix. Heart had already arranged to have it shipped, so Barker and Ornstein raced against the clock to fix the problem.

11.Did You Get the "L"?

When planning began for the IMPs, four university research centers were chosen for the initial test sites. The decision on which university received an IMP was based on the specialties of each research center. Len Kleinrock, at UCLA, was one of the leading experts on packet-switching networks, so he would receive the first IMP and test the network as it was built and used. The second IMP would go to Stanford, where Doug Engelbart would manage the Network Information Center (NIC) providing a network home for ARPAnet documentation. Sutherland (the second director of IPTO), was researching computer graphics at the University of Utah, so the third IMP would go there. The fourth IMP would go to the University of California at Santa Barbara where research was conducted on interactive computer graphics.

Len Kleinrock's graduate students had found out about the problems BBN was having with the IMP, so they guessed that BBN would need to set the date back and give them more time to finish programming the software interface. However, on August 29th, the day before Labor Day, the IMP was delivered to the Stanford shipping dock as planned. Steve Crocker, the graduate student responsible for the host-to-IMP software, heard the news two days earlier and was a little surprised. He spent all night finishing the interface for the Sigma 7 mainframe.

On Labor Day, the IMP was carted up to Kleinrock's lab and connected to a power source by a BBN engineer. When it was powered up, it started working where it had left off back in Cambridge. Unlike the temporary memory used in today's computers, the IMP used core memory that didn't forget anything when it was powered off. When they connected the Sigma 7 to the IMP, the mainframe and the IMP communicated with each other just as planned.