Internet > History >

NSFNET -- National Science Foundation Network

"It was pretty much an unmanageable child," says Ellen Hoffman. "I think back then we all had hopes that it would grow the ways that it did, but none of us believed that it would grow so fast." Hans-Werner recalls working in an environment where "something that was brand-new and totally up-to-date in the morning was completely out of date by the afternoon. That happened pretty much on a day-by-day basis, because we moved so fast."

- Frazer, Karen; The NSFNET Phenomenon; 1987-1995.

The National Science Foundation's enlightened management of the NSFNET facilitated the Internet's first period of explosive public growth.

Starting in 1979, the National Science foundation (NSF) funded development of the CSNET to link computer science departments in universities not connected to the ARPANET, an experience that familiarized them with the significant benefits of internetworking. In 1983, the US military split off the network MILNET from the ARPANET, reducing their continued interest in the network, and freeing it to become primarily a non-classified research network.

In 1984, the NSF began construction of several regional supercomputing centers to provide very high-speed computing resources for the US research community. In 1985, with the CSNET growing rapidly, NSF hired Dennis Jennings to lead the establishment of the National Science Foundation Network (NSFNET) to link five of the university based super-computer centers to enable sharing of resources and information. Jennings made three critical decisions that shaped the subsequent development of NSFNET: that it would be a general-purpose research network, not limited to connection of the supercomputers; it would act as the backbone for connection of regional networks at each supercomputing site; and it would use the ARPANET's TCP/IP protocol.

The initial NSFNET consisted of a network backbone built with 56 kbps lines by a team from the University of Illinois National Center for Supercomputing Applications (NCSA) and the Cornell University Theory Center, because they were the biggest TCP/IP proponents, with help from Dave Mills of the University of Delaware and Hans-Werner Braun of Merit Networks Inc. While 56 kbps sounds awfully slow compared to today's Internet, the load on the early NSFNET was correspondingly less as well -- there was no multimedia yet, and simple wireframe and contour graphics were as complex as most communications got.

Ed Krol was the network manager for the NCSA when the contract was received to establish the NSFNET, and led the team in the network development. Among the first things they did was invite companies to a "bake off" to choose an appropriate NSFNET router, evaluated by a team consisting of Dave Mills, Braun, Scott Brim from Cornell University, Charley Kline and Krol from NCSA, Dave Farber (a grand old man of networking), and Jennings. The only viable commercial product was the Butterfly Gateway produced by BBN for the ARPANET (see RFC 898), but it was far too expensive. However, Dave Mills had the job of maintaining a machine called a Fuzzball which incorporated all of the approved Internet standards, so in another demonstration of the difference between engineering and science -- parameterization by cost -- they ended up choosing the Fuzzball for the initial NSFNET router.

Krol subcontracted the NSFNET operations center to Cornel University, run by Allison Brown and Scott Brim, and began ordering telephone lines and hardware to be drop-shipped to each site. They held quarterly meetings to work through the inevitable problems and issues and coordinate the evolution of the system. Krol recalls that during one of these meetings, Dave Farber mentioned that he had dinner with Len Bozak from Stanford University the previous night. Bozak had mentioned that he had plans to build a machine that would handle a wide range of network protocol needs, including TCP/IP, and might by useful on the NSFNET. Bozak's company was called Cisco, and later became the leading networking company in the world.

As part of the NSFNET contract, Krol authored the Hitchhiker's Guide to the Internet to provide a help manual for NSFNET users, providing one of the first comprehensive surveys of the Internet. In 1992, Krol published The Whole Internet User's Guide & Catalog, one of the first best-selling books about the Internet, and a publication that played a major role in popularizing the Internet beyond the research and academic communities.

In response to the network's rapid growth, in 1987 the NSF gave IBM, MCI, and Merit Network Inc. a contract to upgrade the NSFNET backbone. By July 1, 1988, a much faster 1.5 Mbps network had been established for communication between the original supercomputing centers, plus seven additional research networks: BARRNet, Merit, MIDnet, NCAR, NorthWestNet, SESQUINET, SURAnet, and Westnet. The upgraded NSFNET connected more than 170 TCP/IP enabled networks in all. Merit received a contract to manage the backbone, and traffic began to double approximately every seven months.

In 1990, the military sponsored ARPANET was officially dissolved, and responsibility for the remaining elements of the research network was passed to the NSFNET. The network continued to spread among research and academic institutions throughout the US, including connections to research networks in Canada and Europe, greatly extending the Internet's size and reach.

As the connected network grew, pressure began to build to allow commercial use of the network, which was prohibited by NSFNET management in order to maintain use of the bandwidth for research purposes. In response to the demand, a number of parallel networks were formed to allow commercial traffic, including the UUNET network ALTERNET, Performance Systems International (PSI) network PSINet, CERFNet, and NEARNet. Nearly all of the regional research networks eventually spun off commercial entities. As part of the same commercial dynamic, the umbrella organization Commercial Internet Exchange (CIX) Association was formed by CERFnet, PSINet, and AlterNet to promote commercial use of Internet networking.

In recognition of the fact that the network was growing beyond its research focus, in March, 1991, the NSFNET officially ushered in the next wave of Internet growth by modifying its Acceptable Use Policy to allow commercial use by "research arms of for-profit firms when engaged in open scholarly communication and research". Between growing connections to research networks and increasing commercial traffic, the growth of the NSFNET over the next few years was rapid:

  • By January, 1992, the NSFNET traffic exceeded 12 billion packets (1 trillion bytes) of traffic a month. By November the traffic had doubled, and NSFNET was connected to more than 7,500 networks, one third of which were outside the United States.
  • In December, 1992, the NSFNET backbone was completely converted to a T3 or 44.736 Mbps capacity, capable of transmitting 4 and a half million characters a second.
  • In 1994, the traffic on NSFNET broke the 10 trillion bytes a month level.

Starting in 1990, over the next few years the NSF conducted a series of workshops and studies to plan for transition of the NSFNET to private industry. The vehicle that evolved to support this new architecture was a set of Network Access Points that acted as connection points for the commercial backbones so that the network would remain connected at the top level once the NSFNET was retired. In February, 1994, the NSF awarded contracts for establishment of four NAPs operating at 155 Mbps -- one in New York operated by Sprint, one in Washington, D.C. operated by MFS, one in Chicago operated by Ameritech, and one in California operated by Pacific Bell.

Over the following year, all of the regional NSFNET networks migrated their connections to commercial network providers who were connected to one or more of the NAPs. On April 30, 1995, the NSFNET was officially dissolved, although, returning to its roots, the NSF retained a core research network for research only use called the Very High Speed Backbone Network Service (vBNS), which went on to form the basis for the Internet2 project. At its peak, the NSFNET connected more than 4,000 institutions and 50,000 networks across the Unites States, Canada, and Europe.

Resources. The following sites provide additional information about the NSFNET: