Summary

  • The first NSFNET backbone joined six supercomputer sites over 56 kbps circuits; it was already under strain by early 1987.
  • A 1987 cooperative agreement led by Merit with IBM and MCI replaced that narrow service with a nationwide T1 backbone on 1 July 1988.

The network before the name became shorthand

NSFNET did not begin with the 1987 award to Merit. The National Science Foundation had funded five supercomputing centres—at Cornell, Illinois, Pittsburgh, San Diego and Princeton—and wanted researchers elsewhere to reach them. A separate initial backbone connected six sites, including the National Center for Atmospheric Research in Boulder. It used 56 kilobits per second leased circuits and Digital LSI-11 computers as routers. The topology was modest; its purpose was practical: give university networks a path to scarce national computing resources.

That distinction matters. NSFNET was a family of linked initiatives, not one switch that was thrown on a single date. Regional networks connected universities within states or neighbouring states. The first backbone linked the research centres. Together with other NSF-supported networking, they made remote access and exchange more useful, but they also created demand that the initial links could not absorb.

Congestion became a design signal

By early 1987, Merit’s institutional history says that both ARPANET and NSFNET were heavily used and performance had suffered. In June, NSF solicited proposals for an organization to build and operate a faster backbone linking the centres and regional networks, while providing information services to those networks. The solicitation specified TCP/IP. Merit lacked a nationwide operations structure and suitable backbone technology, so it assembled a partnership: Merit would lead project management, engineering and operations; IBM would supply switching and management technology; MCI would provide transmission circuits.

Michigan also pledged support.

NSF announced the cooperative agreement on 24 November 1987. Merit’s retrospective says the new service was due by 1 July 1988, and that it met that deadline: a T1 network at 1.5 Mbps, about 24 times the initial circuit rate and serving more sites. That is the cleanest boundary between the first network and the new backbone. The 1988 transition was not merely a capacity multiplier. It tied broader regional reach to a production operation with around-the-clock network management and an industry-academic division of work.

The available records do not provide a complete traffic series for the six-site phase or a minute-by-minute cutover log. Merit reports traffic growth above 20% per month after the T1 service expanded capacity; that is evidence of demand released by the upgrade, not a precise measure of the earlier network’s utilization. The technical paper by David Mills and Hans-Werner Braun describes the backbone’s robustness and congestion-control concerns, a reminder that bandwidth alone could not guarantee dependable service.

The historical hinge is thus the conversion of a research-access link into a managed national service. NSF’s centers supplied a reason to connect; the regional networks supplied users; congestion showed the first design’s limit; and the 1987 agreement joined public sponsorship to university coordination and corporate operating resources. The Internet’s later scale was not contained in the six-site network, but the operational problem it revealed—how to make shared connectivity usable when more institutions depend on it—was already visible.

Sources