Chapter 08 - Networking: Diffusion 1972-1979
8.6 Token Ring and David Farber, UC Irvine and the NSF 1969-1974
In contrast to most of those who came to computer communications from 1965 to 1972, David Farber had an extensive background in both communications and computers. In 1956, Farber joined Bell Labs of AT&T working in communication systems. During his ten-year stay, he learned about computers and the needs of ‘real users’ as director of the computer center at Holmdel, N.J., and secretary to the IBM users’ group SHARE. In 1966, he joined RAND where he spent two years and came under the influence of Paul Baran’s work. Next he joined Scientific Data Systems (SDS), a Division of Xerox, and taught evening courses at UC Irvine. In late 1969, Farber accepted a two-year appointment as Acting Associate Professor UC Irvine (UCI) and given two years to earn a permanent appointment.
One question summarized Farber’s interest: “Could I use a whole bunch of these minicomputers together to form a more effective computation environment?” In early 1970, still undecided as to what would qualify as something important, he attended an IEEE conference in Georgia and heard Abramson’s first public presentation of the ALOHAnet. Capturing Farber’s attention, however, was a talk on communication rings by John Newhall and David Farmer of Bell Labs. He recalls:
What triggered me on, more than anything else, was the notion that there was technology for local networking. So I went back to Irvine and started thinking of a set of objectives, and remember, I have two years to come up with a project, to get it funded, and to get enough published on it, enough work done, so in two years I could go up for a permanent position. Good Trick! The things that drive technology are always fun. During that period, I wanted to see just how decentralized I could make an environment. I knew that I could certainly build something similar to the IBM token passing loops to communicate between processors, and I could certainly build a master/slave processor. I helped do that at SDS. And so the objectives of what became known as the Distributed Computer System, DCS, was to see if we could do total distribution. No vulnerable point of error. With both communications and processing and software that was completely decentralized. We certainly didn’t want to duplicate the central control box in the Newhall and Farmer ring.
Farber, Rusty Barbero, a young faculty member, and a couple of graduate students began “throwing ideas around;” ideas based on the token ring concepts elaborated by Newhall and Farmer. Two questions focused their inquiries: “Is it feasible to build a completely decentralized token ring?” and “Can we build a token ring, a communication system, that supports the type of software paradigms that we wanted?” Again Farber:
In the early days, you talk about specific machines that you communicate with. Well, there was no way we were going to do things different if user software knew about specific machines, even indirectly. So we thought of the idea of having software that would couple together by sending detached messages back and forth to each other. Now that’s to my knowledge, the first presentation of message based processing. Further, if there were going to be messages to pass back and forth, couldn’t we do it in such a way that the addresses were independent of the machines that the programs ran on? So we evolved this idea that we were going to have a process structured software system, with messages that were process addressed. So I want to send a message to somebody, to some program, once I knew its name I just sort of sent it to that program and the underlying communication system would take care of it for me.
Farber submitted a proposal for funding to the National Science Foundation (NSF) in 1970, asking for about $250,000 a year, a large grant. Farber recalls:
It got awarded, which surprised everybody including me, and so now the problem was to make it happen!
The Distributed Computer System would consist of a series of minicomputers, terminal controllers and File Machines interconnected together by a token ring network. Farber initially wanted to buy PDP-11s from Digital Equipment Corporation but, fortuitously, ended up using a PDP-11 clone from Lockheed Corporation called the SUE. Lockheed was in the core memory business and began selling computers as a way to put “some iron under their memories.” The SUE came with minimal software, forcing Farber’s team to write all the software they needed, including a compiler. If they had used the PDP-11s, they “might have ended up with a kludge because there was a fair amount of software for the 11.” They would connect each SUE to a shared twisted pair wire using a specially designed token ring circuit board called a ring interface. The inter-device communication architecture mimicked the Bell System digital T-1 specification with the modification of fixed-length messages with a bandwidth of 2.5 megabits per second.
A token ring network interconnects devices as node-to-node connections that close into a ring. Each node consists of a device, such as a computer, with a special printed circuit board connecting to the transmission media. (See Exhibit 6.3 Token Ring) In the case of the UCI network, the transmission media was twisted pair wire. The nodes, or stations, of the network are granted the right to communicate by passing a token, a unique bit pattern – usually eight-bits of ones – from node to node. If the node wants to communicate, it waits for the token to be passed to it which means it is authorized to communicate, then seizes the token and appends its message. Once sent, the token automatically passes to the next node, which repeats the process. If a node has nothing to send, it simply passes the token on to the next node. Since every node receives the token within a known time, bounded by whether all the other nodes have data to communicate or not, a token ring network is, by definition, deterministic.
In February 1973, Farber presented a paper at the International Conference of IEEE Computer Society even though the network did not yet work. He remembers: “A lot of the things back in those days were published before they were built. Dangerous game, but quite often, certainly necessary in the academic world.” NSF approved the contract renewal.
By late 1973, they completed the software needed to connect a computer to the network. Expecting to take a week or two to get a multinode network functioning, they had a three-computer network passing tokens in a day. As Farber remembers: “ It went very fast.”
By early 1974, they had a semi-stable network. Again Farber:
At that point it got very popular, and it was a very strange situation. People started hearing about it and we would get these phone calls saying: ‘We’d like to come down and see it.’ We’d say: ‘You mean you want to talk?’ They would respond: ‘No, we want to see it!’
Farber and his team at UCI proved that another way existed for linking distributed computers. However, their interest was distributed software systems and not advancing computer communications. So they did little to actively promote their token ring technology. But their success would not go unnoticed. In just a few years, their token ring technology -technology funded by the government and thus in the public domain - would be resurrected in response to market demands emanating from Xerox’s refusal to sell its local area networking technology, a technology developed by Robert Metcalfe at Xerox PARC.
Exhibit 8.6.0 Token Ring