Dublin
I arrived in Dublin on Monday. I had heard that it was a holiday, so I asked the cabbie what the occasion was. “Nobody knows,” he replied. And there goes your tip, I thought to myself.
Checking into my hotel, I caught a bus into town to find a true Irish pub. I must have walked five miles before I found one that was open.
“Never thought I’d have trouble finding an open pub in Dublin,” I remarked to the bartender.
“Ah, it’s the holiday, you know,” he replied.
“What’s the occasion?” I asked.
“Nobody knows,” he replied.
I left him a tip. Always tip your bartender is what my mother told me as I left home, and the advice has never failed to serve me.
The next morning, feeling a combination of Guinness and jet lag, I met Dennis Jennings, who picked me up in an ancient BMW and drove me across the street to University College, Dublin, where he runs the Computer Centre.
A tall, affable Irishman, Dennis set me down in his office, filled with mountains of paper and wall-to-wall binders on a wide range of networking topics.
“I understand you’re the father of NSFNET?” I asked.
He laughed.
“More like an uncle,” he said. For the rest of the morning he proceeded to tell me the story of how the NSFNET was born.
When all is said and done, Jennings certainly qualifies as the father (or at least one of several fathers) of the NSFNET. One must ask, of course, if the child turned out as hoped, but that is a different issue from its paternity.
Jennings started working on research networks in 1979 when he started the Irish Universities Network linking the two major Dublin schools. At the time. X.25 was considered a network and for two years, from 1979 to 1981, the Irish Universities Network consisted of X.25 and “higher level” protocols such as X.28 and X.29.
In 1982, he proposed a national star-based network centered around University College Dublin. After a year of discussion, negotiation, and consideration, the Higher Education Authority funded HEAnet.
The original HEAnet was based on the JANET Coloured Book protocols. When I visited Jennings at the end of October 1991, HEAnet 2 was beginning to be operational. Based on X.25 with a 64 kbps backbone and multiprotocol routers, the network was rapidly becoming TCP/IP-based.
In 1983, with funding for HEAnet approved, Jennings heard a rumor at Yale that IBM would be bringing BITNET to Europe. Although the European Academic Research Network (EARN) was going to be based on the big six European players, Jennings managed to turn that into seven countries.
The initial EARN meetings were chaotic and, as I found out over the next couple of days, Dennis Jennings is not shy about grabbing the whiteboard and jumping in. He did, and ended up becoming the first president of the EARN board.
The EARN position started bringing him to the continent on a frequent basis. On one trip, he attended a Landweber networking symposium. At the symposium, Jennings and his wife had dinner with many of the participants. Jennings' wife mentioned to Larry Landweber that Dennis had always wanted to work in the States. You know, one of those polite things you say over dinner.
“Would you like to be director of CSNET?” Landweber asked, not one to let such offers pass idly by.
Dennis had a trip scheduled to California anyway. While he was in California, he phoned Larry Landweber. Next thing Dennis Jennings knew, he was flying to Washington, D.C. to meet with the head of the CSNET board.
“By the way,” Larry added, “I’m also going to Washington. The NSF is considering a new initiative. Want to sit in?”
Landing in Washington after a red-eye flight, Dennis took a cab straight to the NSF, where he meekly took a seat in the back while network luminaries like Dave Farber, Ira Fuchs, and Vint Cerf came in. “Frankly,” Dennis explained, “I was a bit awed. After all, this was America.” Among some Europeans at the time, America had the reputation as some sort of super technology haven.
He didn’t stay awed for too long. In fact, he soon found himself at the whiteboard again.
Eventually, he was offered two jobs: Director of CSNET and the NSF Program Director in charge of networking. He played hard to get for a bit, until NSF upped its offer to include considerably more money.
Back in Dublin, he was sorting out his affairs when NSF called.
“Any chance you can be English?”
Asking an Irishman to be English is a bit like asking an IBM account executive to sell camels.
“I beg your pardon?”
In order to work for NSF, you had to be a citizen of an allied nation. Ireland didn’t cut it. Luckily, Dennis had been born in England. He got an English passport, and on January 2,1985 he started work at NSF.
At the time, NSF wasn’t thinking in terms of building a national network. It had funded four supercomputer centers and merely wanted to give scientists access to them. The original thinking, in fact, was to have four networks, one for each of the centers.
A technical advisory group was formed, chaired by Dave Farber and including such people as the ubiquitous Larry Landweber, Tony Lauck (head of the DECnet architecture group), Frank Kuo from SRI, and several others.
The first big issue to tackle was standards. The computer scientists quickly honed in on TCP/IP. Others were not so sure. At the time, there were several higher-performance proprietary networks such as DECnet and MFEnet (sardonically referred to by some as MuffyNet).
The argument was framed in terms of openness versus efficiency. Protocol suites like DECnet or MFEnet were thought to be significantly more efficient than TCP/IP. Another issue was freedom of choice. Some supercomputer center directors didn’t want to be constrained. Significant populations of users, especially in the physics community, preferred DECnet.
Eventually, TCP/IP won out. It was decided, however, that a network could start out as something else as long as it would migrate to TCP/IP. An interesting side note was the role of OSI. Everyone envisioned that OSI was five years away and would provide a migration path to a world of openness, truth, and beauty. More than five years later, OSI was still five years away and had become the canonical example of the sliding window.
Even more fundamental a choice than the protocol suite was the structure of the network. Jennings felt very strongly that this should be a general purpose internetwork. General-purpose meant that it was not specifically aimed at the current supercomputer user.
The argument was made that a general-purpose internetwork would best serve the goal of promoting science and engineering, the mandate of the NSF. The internetwork part of the structure was fundamental: this would be a network of networks. The lowest unit was the campus network.
From the present vantage point, this doesn’t sound like a radical requirement, but at the time there were many major universities without campus networks. As Jennings explained it, “networking was the private domain of individual researchers.” By refusing to extend the internetwork to the desk or the department, Jennings hoped to help fuel an expansion of the Internet.
In 1985, many of today’s mid-level networks got their start. Some, like the San Diego Supercomputer Center’s SDSCnet, were run by the centers. Others, like the Bay Area Regional Research network (BARRnet) were independently formed.
Looking back, one thing that Jennings regrets is that the NSF did not make a long-term funding plan explicit to these regional networks. Funding was based on a simple year-to-year allotment. A better model would have told the regionals that they would be funded at 100 percent for three years, with a two-year falloff. This is the same model, of course, that IBM used with BITNET and EARN.
Making the funding model explicit would have had two benefits. First, NSF would have been freed from continuing funding obligations with the regionals, allowing NSF to move on to other infrastructure projects. Second, and more important, the regionals would have had a very strong incentive to find alternative funding, from their members, consortiums, or the states in which they did business. The Jennings model of infrastructure is thus to have government start off the project and then have the users take it over. This leaves the government free to move on to the next building block.
In addition to funding the regionals, NSF ended up putting in a backbone, the NSFNET. In August of 1985, Jennings attended a two-day meeting in Boulder, Colorado with the directors of the four supercomputer centers and the director of the National Center for Atmospheric Research (NCAR).
Jennings presented his model of regional networks connected by a national backbone. The directors were not pleased. After all, this was a way for one of their local researchers to use the facilities of another center.
The next morning, however, the directors did an about face. True, their local researchers could use somebody else’s system, but the converse was also true. An Internet would make one center’s services available to the entire country. If you think your center is better than the others (and all the directors did), a backbone is a wonderful idea.
As Jennings related the story, not only did the directors change their minds, they started pounding the table asking why this wasn’t already in place. The group adjourned to a picnic table up on Table Mesa, in the shadow of the Rocky Mountains, and sketched out the backbone on a piece of paper. Jennings still has the piece of paper, dated September 17,1985. He had been on the job nine and a half months.
The backbone would consist of 56 kbps lines linking NCAR and the four centers. It was immediately labeled the “interim backbone” since everyone agreed a T1 network would soon be needed.
Events were moving very quickly. At times, Jennings was cautioned by the NSF bureaucracy that he was “sailing close to the wind.” For example, Bill Schrader and others at Cornell went ahead and ordered all the 56 kbps lines before NSF had formally approved the project. Cornell then took the lead in putting in the interim system.
In addition to lines, there was a crucial decision on the selection of routers. Major and minor router vendors were asked to make presentations. BBN had a wonderful system, but their routers cost U.S. $180,000 each. Cisco and Proteon came close, but didn’t have nationwide support structures in place.
Ultimately, attention focused on the Fuzzball, a device based on the PDP-11 and developed by David Mills. Everybody was reluctant to depend on a homegrown system developed and maintained by one person. As reluctant as anybody to have the national network based on the Fuzzball was David Mills himself.
David ended up spending the next two years maintaining and modifying his Fuzzball. Some of the changes were simple bug fixes, but others reflected some significant architectural flaws in the original TCP/IP.
For example, the TCP/IP protocol suite implied a single backbone: the ARPANET. Now, with the addition of the NSFNET, there were two backbones and the Exterior Gateway Protocol (EGP) broke. Mills was forced to periodically hack up his Fuzzballs to handle this, and other, situations.
Jennings felt that the decision to go with the Fuzzball was one of the major successes of the NSFNET project. It forced the Internet community into a formal examination of routing, leading to specifications of router requirements and the development of new protocols such as the Border Gateway Protocol (BGP).
Common, open specifications for routing in turn led to the immense growth of companies like Cisco, Proteon, and Wellfleet, as well as substantial product lines for larger companies such as DEC and Sun Microsystems. Low-cost routers allowed the creation of ad hoc internetworks, which in turn led to the creation of companies like PSI and AlterNet.
In March 1986, Dennis Jennings brought to a close his 15 months at NSF. Just 15 days later, the 56 kbps backbone went live. Jennings was replaced at NSF by Steve Wolff, who began the difficult job of formulating the solicitation for the T1 backbone and then later, a T3 backbone.
If Dennis Jennings is the uncle of NSFNET, then Steve Wolff would qualify as the nanny, expanding the network from an ad hoc, interim system into a mature, production network transferring over 14 billion packets per month. Wolff deserves a great deal of the credit for the rapid growth of the Internet by aggressively expanding the backbone, funding international links, and encouraging the growth of the regional and campus networks.
After a morning learning the history of NSFNET and a chance to view the tattered piece of paper that served as the original architecture document, we went out to lunch with the directors of the other two university computer centers. Over a glass of wine, the conversation turned to the question of European backbones. How would Dennis, if he had the power, accomplish the same thing in Europe that he did in the U.S.?
Dennis quoted a classic Irish proverb given when a tourist asks a farmer for directions: “I don’t know, but if I were going there, I certainly wouldn’t be starting from here.”
In the United States, a government agency, the NSF, was charged with promoting research and education. Within the European Commission, however, there was no such group. There was no mention of education in the Treaty of Rome, the enabling legislation for the European Community. Instead, funding for research networks was funneled through groups like the infamous 13th directorate.
DG XIII was responsible for promoting things like competitiveness and standards. In practice, this led to a strong focus on the PTTs and OSI. What it did not lead to was an operational, pan-European backbone, either commercial or research. Everybody was waiting for the Commission to act and the Commission was waiting for OSI to be commercially available.
After lunch, Dennis told me about an extraordinary meeting that was to be held the next day to try and bootstrap a European Backbone (EBONE). We quickly rearranged my plane tickets to route me to Amsterdam and Dennis drove me back to my hotel for a glass of Guiness.