There were some LANs in those days, but they had serious limitations, Metcalfe said. PARC used one, called the Data General MCA, between its Data General Nova minicomputers. But it could only connect 16 systems, and they all had to be in the same room. The cables were about 1.5 inches thick, he recalled.
At their desks, Metcalfe and his colleagues had ASCII text terminals that talked to a minicomputer at 300 bits per second. That wasn't enough speed to send files to a high-resolution, one-page-per-second laser printer. They were entering tomorrow's digital world with yesterday's connectivity. Though not everyone at PARC agreed, Metcalfe said, he insisted that networking was critical for the new computers.
"All that power would be for naught if they weren't connected," Metcalfe said.
The system Metcalfe and Boggs envisioned would leapfrog any LAN of the time.
"Our spec was, we wanted to connect 255 personal computers at a distance of a mile, at some number of hundreds of kilobits per second ... and we wanted to do it with a minimum of cabling, because the predecessor networks all had these rooms full of cables that we called 'rat's nests,'" Metcalfe said.
The network that Metcalfe and Boggs proposed met all those qualifications, including a top speed of 2.94Mbps. If you got that speed today, you might say you were getting 3Mbps, which would be decent performance on a 3G cellphone. But Metcalfe won't round up the number, even now. In 1973, he explains, the 60Kbps rounding error that would require was serious bandwidth in itself: more than the speed of a transcontinental Internet trunk. But compared with those 300bps desktop connections, the new network was an embarrassment of riches, about 10,000 times as fast.
In designing the network, they emphasized distributed computing, a new concept that was at the center of all of PARC's work at the time. It represented a shift away from the large, centralized time-sharing systems that dominated electronic computing since its inception in the 1950s. All the PCs on the Ethernet LAN shared one cable, and the algorithms that ran the network were distributed among them. Each participating computer got an add-on board in which Ethernet microcode was implemented on a standard microprocessor.
"It was just a cable running down the corridor and all the attached PCs would just plug into that cable near where they were," Metcalfe said.
Metcalfe and Boggs had the network running by November 1973. At first, PARC employees could order an Alto with or without Ethernet, but soon everyone depended on it, Metcalfe said. One reason was an application for testing each Alto's semiconductor memory, which was still an unproven technology. PARC scientist Chuck Thacker wrote a diagnostic routine that would run on the Alto and test its memory while the user was away. It sent the test results to a maintenance system over Ethernet.
Sign up for CIO Asia eNewsletters.