The NSFNET Backbone Project, 1987 – 1995
NSFNET: A Partnership for High-Speed Networking
NSFNET: Coming Together
In 1987, scientists and researchers in some universities and research centers already benefited from access to the first NSFNET, a 56 Kbps backbone network linking five NSF-sponsored supercomputer centers and the National Center for Atmospheric Research, or NCAR. But by mid- 1987, traffic on that backbone had increased to the point of overload. “People started to use it, and to like it, and to rely upon it. Their research was already critically dependent on the network being available,” recalls Hans-Werner Braun, co-Principal Investigator of the NSFNET project at Merit, Manager of Internet Engineering at Merit from 1987-1991, and a principal designer/engineer of the routers used on the initial 56K NSFNET backbone. In response to this need for adequate network service, the National Science Foundation issued a Project Solicitation for Management and Operation of the NSFNET Backbone Network (NSF 87-37) on June 15, 1987. But who would be served by the creation of an expanded national backbone network? And who could put such a project together?
The NSFNET partnership team–Merit Network, Inc., IBM, MCI, the State of Michigan, and the National Science Foundation–all brought unique strengths and resources to the project. Each played an important role in building the national backbone network service.
It may be difficult to believe today, but until recently–within the last two decades–most scientists and researchers were able to carry on their work without recourse to communicating via the marriage of computer and telephone lines known as computer networking. What was it like to be part of the research and education community in the United States before the NSFNET? Scientists had to travel to reach computing resources, from campus to campus or, in many cases, out of the country. Most people didn’t immediately share the results of their research with colleagues outside of their own departments; they would make a telephone call, mail a package, or arrange to meet in person. Researchers and scientists had little or no experience with sending a quick e-mail to notify a colleague of an interesting article, or downloading a data file from a fellow professor in another state, much less working simultaneously on a complex scientific problem in real-time.
Campus networks were increasingly common at major research universities, linking faculty with academic computing centers, but these closed networks were often like islands to one another. This began to change in the United States in the late sixties and early seventies when various wide-area computer networks, or WANs, began operation.
These WANs were primarily Federal research projects, the first of which was the ARPANET in 1969. An outgrowth of the Department of Defense’s Advanced Research Projects Agency, the ARPANET’s packet- switching scheme was meant to provide reliable communications in the face of nuclear attack. By the early 1980s, other examples of Federal networks included HEPnet and MFENET; however, the variety of different protocols used could impede efficient communication. A small but expanding part of the research and education community used e-mail via UUCP or had access to some kind of message-forwarding system such as BITNET, but access to Federal networking facilities such as the ARPANET was usually limited to computer science departments or restricted to researchers affiliated with laboratories funded by various Federal agencies such as the Department of Defense, the Department of Energy, and NASA.
While for these users, computer networking had become vital to their daily activities, the average professor of English or economics was unaware of the opportunities provided by information technology applications. Ellen Hoffman, manager of the Network Information Services Group at Merit from 1990 to 1993, reminds us:
“We forget that only a short time ago, the Internet was not a word that rolled off anyone’s tongue readily. It wasn’t so long ago that only the large research universities in the country were connected, and within those universities often there was only one department using the network. Even if there were 50 or 100 schools connected, most people on those campuses had no idea that there was an Internet connection to their campus.”
In the United States, “only about 50 percent of the large research universities had ARPANET connections,” estimates Mike Roberts, vice- president of EDUCOM, the nonprofit consortium formed to promote information technology applications in education.
Merit Network, Inc., the consortium of Michigan universities formed in November of 1966, provided both campus and statewide networking facilities. As a result, it knew, perhaps more than almost any other educational institution, that the lack of a reliable national backbone network was becoming a problem. By the late 1980s, approximately two- thirds of the state universities in Michigan were connected to MichNet, the statewide research and education network operated by Merit. Running its own protocols for much of that time, the connectivity to the outside world and to each other that MichNet provided instilled a broader understanding among Michigan colleges of the possibilities inherent in “internetworking” compared to most other states. Having seen the value of providing networked communications capabilities to members of the research and education community, Eric Aupperle, president of Merit Network, saw an opportunity to take the fruits of Merit’s experience with MichNet beyond Michigan, in an effort that would benefit not only the research and education community but the nation as a whole.
“In an environment of resource constraints, the higher education community has increased its investment in information technology to enhance the effectiveness of research and education. More and more of the important intellectual output of the higher education community first exists in digitally encoded form. The higher education community will benefit enormously from NSFNET and has a major interest in its development and operation … NSFNET should be managed by an organization with strong academic roots so that it will be able to anticipate the needs of the academic and research community and build an organization that fits the environment in which NSFNET will operate.” (Merit Proposal, submitted to NSF 14 August 1987)
As part of the research and education community, Merit could provide the foundation for a successful partnership between academia, industry, and government to construct a national backbone network.
Merit provided the glue that held the project together. That glue was composed of clear objectives, firm yet flexible management, superior staff and the maintenance of close ties to the community served by the NSFNET. The larger currents prompting the initiation of the NSFNET backbone project included evolving Federal networking programs such as the NSF-sponsored supercomputer centers and eventually the High Performance Computing and Communications (HPCC) Program; innovations in computers and networking technology; and the desire to extend networking throughout the research and education community. However, according to Merit, the immediate, overriding objective was simple: “generalized connectivity.” Expanding the 1986 NSFNET, scaling it to support the tremendous growth and interest by the community while recognizing the community’s increasing reliance on the network for their day-to-day tasks was Merit’s goal.
Another important function Merit provided was to mediate between industry, government and the research and education community. While the team partners exhibited an outstanding willingness to work together, nonetheless “you’re talking about very different cultures,” notes Ellen Hoffman. By having a representative of academia be responsible for the overall management of the project, the partnership was able to avoid competition between industry partners, as well as inspire greater confidence in the NSF and the research and education community that the needs of the community would be well served: “Merit was an excellent organization to have in that situation, because it had the interests of the community at heart,” says Jane Caviness, Program Director for the NSFNET 1987-1990. According to Priscilla Huston, Program Director for the NSFNET 1993-1995, “With the NSFNET program, NSF funded and encouraged Merit, but Merit facilitated the meetings, the open discussion, and got industry and others to be proactive.” But Merit couldn’t do it alone.
Composed of representatives from Michigan universities, the Board of Directors of Merit in 1987 was chaired by Doug Van Houweling, Vice- Provost of Information Technology at the University of Michigan. While working at Carnegie-Mellon University prior to arriving at the University of Michigan, Van Houweling had discussed with IBM an Ethernet-over-satellite communications system for higher education, and suggested that IBM and Carnegie-Mellon make a proposal to the National Science Foundation. Although that didn’t happen, as a result of this initial foray, Van Houweling “had been thinking about these issues” when the NSF released its solicitation in 1987:
“In my new role as Chairman of the Board of Directors at Merit, I went back to some of the same people at IBM and suggested this project to them. It took several meetings to get the right group of people in the room, but when we did, there was a lot of common interest.”
Van Houweling got in touch with Eric Aupperle and began a series of meetings with potential industry partners to respond to the solicitation.
Participants in some of these early meetings included Al Weis and Barry Appelman from IBM Research; Bob Mazza, Walter Wiebe, and Rick Boivie from IBM’s Academic Information Systems division; and Eric Aupperle and Hans-Werner Braun of Merit. Van Houweling explains the thinking in this way: “We felt we needed a university organization, and Merit certainly filled that role. We needed a software and equipment provider, and IBM certainly fulfilled that role. But we also needed a carrier for the communication,” so Al Weis at IBM talked to Dick Liebhaber, a vice-president at MCI, about the possibility of MCI becoming a partner in the proposal.
According to those involved in these initial talks, there were many reasons why IBM and MCI chose to become part of the Merit NSFNET partnership, including the desire to be part of a high-profile, high technology initiative benefiting the broader university community. However, from a business perspective, becoming involved with the NSFNET backbone service might provide opportunities for “technology transfer.” Broadly defined, technology transfer refers to the process by which the products of government-funded research and development move out of restricted use or research-only application to the private sector.
Executives involved with academic computing within IBM, a world leader in research and commercial products for systems and connectivity, were interested in increased experience and involvement with the Internet protocol TCP/IP, which by 1986 ran on a majority of campus networks. By developing and implementing routers, adapters, network management tools, and other equipment for the NSFNET backbone, IBM would be able learn a great deal, as Al Weis explains:
“IBM was unable to interconnect its large mainframes and some of its new workstations to all the research communities’ networks and get adequate performance, because those networks were TCP/IP networks. By working on the NSFNET backbone service, we learned a lot about TCP/IP and were able to address these needs common in many academic environments.”
MCI similarly grasped that data communications constituted an ever- growing part of the economy, and hoped to develop a reputation for data networking that would match its increasing clout in the long-distance telephone business. The divestiture of AT&T in 1984 in combination with continuing deregulation of the communications industry had created new markets for all kinds of information processing and telecommunications services. An increasing share of corporate, university and government agency budgets was devoted to telecommunications and data processing as the shift from a manufacturing to a service economy took hold in the mid-1980s. With over 34,000 global network miles and a coast-to-coast fiber optic network in 1987, MCI had the ability to provide high-speed circuitry for the NSFNET backbone service and, at the same time, learn about data networking at T1 and greater speeds. Al Weis recalls Dick Liebhaber, a senior vice-president at MCI involved with the NSFNET partnership from the very beginning, saying that MCI had always been known as a voice company, and if MCI was going to learn about data, there was no better opportunity than this project. In this way, two major U.S. corporations signed on to be part of the NSFNET backbone service.
When successful, this “transfer” of technology and know-how is an interactive process, with benefits flowing both ways: industry and academia alike benefit from the increased knowledge and (in the case of the NSFNET backbone) the resulting technology. On the NSFNET project, technology transfer was successful at many levels. Partnering with industry, according to Dale Johnson, the manager of Merit’s Network Operations Center (NOC), enabled Merit to “take technology that existed in the labs, and had been proven in the labs, and take procedures that had existed in the business sides of our partners, and bring them into the Internet and make them part of Internet culture.” The NSFNET project helped to distribute technical knowledge and skills among the Internet community, by training personnel in Internet engineering, routing, and network management. Finally, IBM and MCI also were able to gain valuable experience with high-speed networking technology, which later helped them to create commercial products and services based on that experience for the emerging markets for Internet services and products stimulated by the entire NSFNET program. According to Larry Bouman, MCI’s Senior Vice-President for Network Operations, “I don’t think we’d be where we are today if it weren’t for the NSFNET project. Providing the circuitry for the backbone service–and working with IBM and Merit to get it up and running–gave us experience that led to a productive new focus for the company.”
In return, the NSFNET backbone service’s corporate partners invested far more time, money and resources than was required by their joint study agreements with Merit. According to the proposal, IBM would provide the computing systems for the network, including hardware and software, and MCI would provide the underlying telecommunications circuits for the NSFNET. However, by all accounts, there was never a sense of meters ticking away for the partners’ contributions, and in fact IBM and MCI were generous with staff time and resources such as financial support or equipment. Harvey Fraser, Project Manager at IBM on-site at Merit for the NSFNET, puts it this way:
“The team worked well because we had resources, executive time, and the desire to make it a success, as well as a lot of good faith and good will. When issues came up, we evaluated them and, if we could, provided whatever was needed. For example, when we needed more routers, IBM got them for us. If we needed more lines for a test network, MCI got them for us.”
Harvey Fraser, Project Manager at IBM on-site at Merit for the NSFNET backbone service.
Mark Knopper, who succeeded Hans-Werner Braun as Manager of the Internet Engineering Group at Merit in 1991, also remembers that the behavior of NSFNET’s industry partners went above and beyond the call of duty:
“I’d go to meetings where the executives would say, ‘Well, what are the problems this week?’ and ‘What can we do to help?’ I would be sitting with vice-presidents and top managers of IBM and MCI and they would say, ‘What do you need? What can we do to fix things?’ Once they knew about a problem, they’d try to find a way to solve it.”
IBM and MCI realized that in working on a project like the NSFNET, which would necessitate the use of uncertain and unproved technology as well as the development of new operating and management systems, a certain amount of risk was involved. The answer was not to simply throw money at the project, but to make a number of different kinds of resources–people, technology, funding–available as needed, according to the judgment of each partner. This relationship of “good faith and good will” among the partners was built upon the high caliber of technical expertise going into the NSFNET project. IBM, MCI and Merit assembled a top-notch team for the NSFNET backbone service: “The quality of the technical contribution from people in Merit, IBM, MCI, all three, was really spectacular,” Doug Van Houweling remembers.
“IBM and MCI essentially donated a lot of resources and a lot of energy for the NSFNET backbone service for the good of the country. Working on a team like that, where people from IBM were typically interested in developing products and moving them into the marketplace, and MCI was typically interested only in selling circuits, it was personally satisfying to see people with different perspectives on what is important in their organizations and their businesses, joining their ‘vision’ of a national network, so that it could happen. If any one of the partners had balked, we probably couldn’t have done what we did.” Elise Gerich, Associate Director for National Networking, Manager of Internet Engineering, Merit Network.
IBM and MCI also brought certain established ways of doing business, “standard operating procedures” forged in the commercial realm that created a strongly production-oriented mind-set for the team. Because of this, according to Dale Johnson, the NSFNET project team “brought applied commercial operating procedures to the Internet. All of this was standard in the telephone companies and also at IBM.” Documentation, trouble tickets, escalation procedures, “follow- through,” and statistical reports were just some of the many procedures that were adopted by the NSFNET project and modified to be applicable to the Internet. Hans-Werner Braun explains what a difference these procedures made to the thoroughness of project team behavior:
“During one weekend all the routing daemons, all the code that handles routing in the nodes, crashed. I called the person who developed the routing software, Yakov Rekhter at IBM-Yorktown, at home on a Sunday and had a fix for the problem in about two hours. I think that’s pretty amazing. Also, with MCI, if there was any problem with the lines, they had a very aggressive escalation procedure. Within only a few hours if the problem wasn’t resolved it was supposed to be on the desk of a vice-president. It made a huge difference.”
These innovations, which were standard in industry but new to academic networking, would not have been worth much without high-level corporate backing behind them, and according to everyone involved with the NSFNET backbone service, IBM and MCI did indeed provide that commitment. All of the partners–NSF, Merit, IBM, and MCI– demonstrated an overwhelming desire to do whatever was needed to make the NSFNET a success. However, the involvement and dedication of technical and managerial personnel, at every level, at IBM and MCI was among the most important factors in the project’s success. With IBM, MCI and Merit, then, industry and academia were well represented for the NSFNET project. The final essential ingredient came from the United States government.