National Networks

The Open Systems Interconnection (OSI) model

The OSI seven layer protocol model

The diagram shows a popular model that provided a framework for discussing networks. The model divides communication protocols into seven logical layers. Each layer serves the layer above it and is served by the layer below. In this model, Ethernet combines the physical and datalink layers, IP and X.25 are network protocols, TCP is a transport protocol, and FTP and Z39.50 are application protocols.

By the early 1980s the benefits of networking had become generally recognized, but building the networks proved to be challenging. The developments fell into two categories: the campus networks and the networks between universities. Together they developed into the Internet that we know today. If you read the conventional history of the Internet you could easily believe that the Internet protocols were so superior that it was inevitable that they would be used for both a national network and for campus networking. At the time it was far from inevitable.

Few people remember that the telecommunications companies and the computing industry fought for years to kill the Internet and its TCP/IP family of protocols. They recognized the potential market and the need for a shared set of protocols, but they wanted to control everything. They embarked on an intense lobbying effort to suppress any competition, using every political resource to stop alternatives, and were particular outspoken about wasting government money on TCP/IP developments. The telecommunications companies, who have made such enormous profits from the Internet, were the most opposed to its development.

Rather than adopt the Internet protocols, the industry set out on an ambitious program known as the Open Systems Interconnection (OSI). OSI was supported by all the major companies. Technically it had two main parts: standardization and creating commercial products. IBM alone spent hundreds of millions of dollars. It seemed inevitable that the national and international networks would be built to these standards. Parts of OSI, such as X.25, were successful, but as a whole it suffered from over complexity and eventually collapsed under its own weight.

Meanwhile universities had difficult decisions to make about what protocols to use on campus and for academic networks. For years, they wrote that their networking plan was to use TCP/IP "until OSI becomes available." Eventually they dropped the final clause. How did this happen?

The first national networks

When universities began to recognize the potential of a national network there were several possibilities. The dominant network between universities was Bitnet. This was a store-and-forward network that ran between IBM mainframes. It used software that IBM had developed for its own internal network. Bitnet provided the first widespread email service between universities and was used for most bulletin boards (called LISTSERVs) until the web became established in the 1990s. As late as 1986, there was no way to send an email message between Dartmouth and Carnegie Mellon except over Bitnet.

The first commercial network was Telenet, which offered X.25 services for a monthly fee. CSnet was an X.25 network developed by those computer science department who were excluded from the ARPAnet. At Dartmouth, we used Telenet for remote terminal access and CSnet for email.

The ARPAnet was a higher speed network developed by the Department of Defense's Advanced Research Project Agency (ARPA). The TCP/IP protocols were developed for the ARPAnet. A select group of universities were members of ARPAnet, but its use was restricted to people who were doing ARPA-related work. At Carnegie Mellon we were always careful to observe this restriction until I learned that MIT ignored it. They considered that everybody on their campus was doing ARPA-related work.

The success of TCP/IP

One reason that OSI failed was that it attempted to standardize everything. With little operational experience to guide them, the standardization groups added more and more features. The early Internet was much more pragmatic. Its motto was "rough consensus and running code." TCP/IP specified the network and transport protocols, but made no attempt to define the underlying network technology, while protocols for applications such as email were not standardized until there was practical experience with running code.

A key to the success of TCP/IP was the decision to include the protocols in the Berkeley Standard Distribution of Unix (BSD). This was also funded by ARPA. BSD 4.2, released in 1983, had an open source version of the full family of protocols. This code, which was soon ported to many other operating systems, was the basis for the first generation of the Internet.

In hindsight, the benefits to university computing are obvious. We could use the same protocols on campus and off-campus without the need for complex and inefficient gateways. Because all the versions were based on the same implementation, there were few compatibility problems. TCP/IP came from the computer science research community, which has a tradition of shared problem solving, and it was backed by a well-funded government agency with strong interests in its success.

The NSFnet

The crucial event was the 1985 decision by the National Science Foundation to establish five supercomputer centers for academic research, to connect them with a high-speed network, and to base it on the Internet protocols. Again this was not inevitable. The NSF's newly-formed networking group had a hard fight to stop the traditional disciplines from taking their money and to resist lobbying by commercial interests who were opposed to something that they did not control. It helped that the NSF's effort was led by an Irishman who could charm anybody. The NSFnet backbone was built by a consortium put together by Doug Van Howeling at the University of Michigan with generous support from IBM and MCI. IBM provided the routers and MCI supplied the T-1 connections, which at 1.5 Mbits per second were far beyond the speed of any previous data networks.

The NSF also funded an enlightened program to extend the NSFnet by creating regional networks. Our experience at Carnegie Mellon was typical. The computing directors of the major Pennsylvania universities formed a team led by Gary Auguston of Penn State. His design criteria were, "TCP/IP, T-1, free." With support from Bell Atlantic, we met with a senior member of the governor's staff. His first question was, "How will this get the governor re-elected?" His second was, "How can I help?" We then collectively applied to the NSF for a substantial grant. This grant supported the Pennsylvania regional network for long enough to become self sustaining.

Finally the NSF successfully transferred management of the network to the commercial sector in 1995. This was a shock to the universities, as henceforward we had to pay for our networking.