NSFNET

~ the obit ~

Turkey's first internet connection came up on 12 April 1993, a 64 kbit/s leased line running from the METU computer center in Ankara to Washington. What sat on the far end of that line was NSFNET, and it had two years left. I got online in 1998 on a modem at home, three years after the switch-off, so the backbone my own country first plugged into is one I never touched and never heard named at the time.

It was a National Science Foundation program that ran from 1985 to 1995. The problem it was funded to solve was expensive computers rather than communication: university researchers needed to reach five supercomputer centers and could not. Dennis Jennings made the decision in 1985 that the thing would run TCP/IP, which is most of why the acronym ended up mattering. The first backbone went live in 1986 on 56 kbit/s leased lines, routed by PDP-11 minicomputers.

The upgrades now read like a countdown. Merit Network at the University of Michigan won the operating contract in November 1987 with IBM and MCI behind it, and by July 1988 thirteen nodes were running at 1.5 Mbit/s. The 45 Mbit/s version finished in November 1991. Traffic doubled roughly every seven months for years, which is the kind of curve that ends with somebody calling a meeting about money.

~ the rap sheet ~

~ the rule about what your packets were for ~

NSFNET had an acceptable use policy, and from here it is the strangest thing about it. Public money paid for the wires, so what moved on them had to be research and education. For-profit activity was out, and so was extensive use for private business. Somebody had written down what your traffic was allowed to be about.

It was never enforceable in any deep sense, but it shaped the market that grew around the network. Commercial providers built their own backbones to carry what NSFNET would not, and in 1991 several of them, PSINet among them, set up the Commercial Internet Exchange so they could hand traffic to each other without a federal wire in the path. The policy loosened in March 1991 to cover the research arms of companies, and on 23 October 1992 Congress passed a law letting NSF connect to commercial networks outright.

~ the part where it becomes a business ~

In September 1990 Merit set up a nonprofit called Advanced Network and Services to run the backbone, with IBM and MCI in it, and in 1991 that nonprofit grew a for-profit subsidiary to sell capacity on the same infrastructure. A network the public had paid for now had a sales arm. Congressional hearings followed, and the complaint in them was that the government had built an advantage and then handed it to one company.

NSF's answer was to stop owning a backbone at all. The replacement was four network access points where any network could meet any other, plus a routing arbiter to keep the tables honest. The science that had started the whole program moved onto a smaller research-only network called vBNS, which is the part of NSFNET that went on to become Internet2.

~ the last day ~

The backbone service was decommissioned on 30 April 1995. At the end it reached more than 4,000 institutions and something over 50,000 networks. Traffic had been running past ten trillion bytes a month since late 1994. Nothing went down. The load had been draining onto commercial routes for years and the shutdown was closer to a change of billing address than an unplugging.

What went with it was an argument that has not come back. For ten years the main road of the internet was public property with a rule about what you were allowed to carry on it. After April 1995 there was no such road, and the question of what the wires are for stopped being a question anyone had standing to ask.

@}-->--

              )
             (_)
            .-'-.
            |   |
            |   |
            |   |
            |   |
          __|   |__   .-.
       .-'  |   |  `-:   :
      :     `---'     :-'
       `-._       _.-'
           '""""""

« previous  ·  random  ·  next »