Chapter 08 - Networking: Diffusion 1972-1979
8.8 Massachusetts Institute of Technology 1974 - 1977
In 1974, the only effective means computer scientists at MIT had to interconnect their computers were two IMPs, an increasingly unacceptable solution, especially given what was known about Metcalfe’s work, enviously nicknamed “ALOHAnet on a wire.” Thus MIT scientists were challenged to either build or buy local area networks.
The Artificial Intelligence (AI) Laboratory acted first. Taxed by their own research agenda, they had little desire to develop their own local area networking technology and approached PARC about buying Ethernet products. Xerox declined the opportunity. Xerox viewed Ethernet as proprietary and integral to their office systems and not a product to be sold separately. Professor Jerry Saltzer, one of the first to become involved in network development for the Laboratory for Computer Science (LCS), remembers:
Xerox had invented and built the first version of the Ethernet, but still considered it proprietary and would not allow anyone to use the internal knowledge. The fact that it worked was sufficient for someone else to say: ‘Well, in that case, we’ll build one too.’ So the AI Laboratory built Chaosnet. The only reason it was invented was because we couldn’t use Ethernet. Chaosnet was essentially another Ethernet that had slight differences, but the differences aren’t important enough to worry about.
In 1975, the LCS became serious about local area networking. Michael Dertouzos, the head of LCS, Saltzer and Dave Clark, a research associate, directed Ken Pogran to investigate available local area networking technologies. DARPA, learning of LCS’s interests, lobbied MIT to import the token ring technology developed at UC Irvine by Farber, and not create a new technology.
Pogran visited PARC and spent a day being briefed by a clean-shaven Metcalfe, a business-like contrast to the Metcalfe he had known during the earlier Arpanet days. Pogran then visited Farber and learned of his plans to create a second generation token ring employing graduate students, not to study token ring per se, but needing an improved token ring to do research in distributed systems.
On his return to MIT, Pogran presented his findings to Dertouzos, Saltzer and Clark. They decided against Chaosnet, concerned that future product support and development they knew they would need would be a low priority for the AI Lab; a serious problem since LCS did not have a hardware development capability as the AI Lab did. Since Xerox wouldn’t share its technology, their only alternatives were either to create a new technology or import UC Irvine’s token ring technology. LCS chose token ring even knowing graduate students were developing the version of the product they would use.
LCS began planning what Dertouzos called the “76 Net.” Pogran began as project manager but soon took on direct engineering tasks that no one at UC Irvine had either the time or competence for. Despite his efforts, Pogran could not meet the original schedule, finally delivering working LNI’s (Local Network Interfaces) in 1977 supporting a 1 megabit per second token ring network.
That same year, 1977, Saltzer, took his sabbatical from MIT, and spent a year working for IBM:
I was working in White Plains, but I traveled quite a bit, and one of my stops was in Zurich. I discovered that they were interested in token rings so we proceeded to compare notes, and I basically supplied them with every piece of information we had. The Zurich laboratory had as its charter to do communications, and it was trying to figure out if it could do something useful in the area of local area network communications. At this time, the only local area network in all of Europe was the Cambridge Ring at Cambridge University, U. K. Ethernet was something being done 5,000 miles farther away than a token ring. Xerox was also a competing company and it was hard to get inside to look at what they were doing.”
Saltzer’s opinions, based on his early experience with token ring at LCS, reinforced IBM’s historical preference for deterministic, synchronous protocols. The initial diffusion of Farber’s and UC Irvine’s token ring technology to IBM had occurred months earlier with Phillipe Janson. Clark adds:
Phillipe Janson, after receiving his PhD from MIT in 1976, joined IBM in Zurich. It’s my belief that when Zurich was looking to build a network, and they didn’t want to build an Ethernet because they had a strong dose of ‘not invented here,’ and Janson was the one that suggested that they ought to build rings based on his experience at MIT.
LCS Network represented an incremental innovation on the token ring technology imported from UC Irvine. For example, concerned with finding wiring faults, the LCS Network connected each network node through a central “wiring closet,” giving rise to a star topology that would become standard for future token ring networks.36
Another important contribution was the symmetry of the starter token. See Clark, Pogran and Reed, “An Introduction to Local Area Networks,” Proceedings of the IEEE, Vol. 66, No. 11, Nov. 1978