Tuesday was spent hiding from the cold, doing research in the Hawk and Dove, one of a half-dozen Irish bars on Capitol Hill. After paying a call at the Office of Technology Assessment to brief a staff member on the Bruno project, I took the metro under the mall to the offices of the National Science Foundation.

Steve Goldstein, the manager of international programs for the NSFNET, had invited me to his home for dinner and we would join his carpool from the NSE. He had just come back from a WAIS symposium and brought me a set of the conference materials. We both agreed that WAIS could easily prove to be the Lotus 1-2-3 of the Internet, providing users with a simple, intuitive way to get real information.

Steve and his life-partner Isabelle were both members of a gourmet cooking club and my timing was lucky enough to get the leftovers. While we waited for Isabelle to get home, Steve heated up a pork roast stuffed with mushrooms and a dish of fennel bulbs in a delicate cheese sauce while we talked about the NSF and the NSFNET.

The NSF has taken its charter to help U.S. science and engineering as a broad mandate which ranges from hands-on science museums for children all the way to funding research into teraflop computers.

The NSFNET is a perfect example of the broad interpretation NSF takes of its mandate. Instead of dedicated networks for supercomputer centers, the agency built a system of general-purpose regional networks connected by a backbone‹a true infrastructure.

Steve Goldstein described the NSF funding philosophy as strategically placed "drops of oil," hoping to leverage comparatively small expenditures to achieve a broad impact. The drops of oil philosophy was necessary because the agency's budget was fairly small. Program directors have to husband their money carefully and overhead is kept down to a paltry four percent.

The international program of NSFNET, for example, makes do with a budget well under a million dollars per year, a minuscule amount when you consider how high telephone tariffs are and what an important role NSF played in building a global Internet.

To spread the money out, NSFNET usually funds half-circuits. Most international circuits are actually operated by the two PTTs involved, each making up the tariff for the outgoing line. Typically, NSF only picks up the cost for the U.S. half-circuit, which are often significantly lower in cost‹anywhere from 20 percent to factors of 2, 3, and even 4 times less. In some cases, the international partner will pick up the cost for both half-circuits. NORDUnet, for example, paid for the whole line to the U.S. for some time until the NSF was able to pitch in.

In the case of the "fat pipes," the high bandwidth circuits overseas, there is usually a consortium of agencies that pay. The link to the U.K., for example, has the NSF, DARPA, and NASA sharing the costs in one direction and the Joint Network Team and the Ministry of Defense sharing the other side.

In addition to sharing costs for general-purpose infrastructure, NSF is able to get a lot for its money because of simple competition. Sprint, for example, is the manager of international connection services for NSF, and won that contract because of a very aggressive bid response. Sprint didn't do this out of altruism‹the experience it gained with NSF was invaluable in capturing other international business.

NSF encourages regional solutions as a more efficient way to connect the world. To Europe, for example, the NSF helps fund the lines to INRIA in France and NORDUnet in Stockholm, both major gateways for their respective regions.

The agency also uses line sharing to buy larger pipes. The NSFNET backbone, for example, does not actually lease telephone lines. Instead, NSFNET contracts for a T3 service which runs on an underlying network owned by ANS. That underlying network carries traffic both for NSF and for ANS commercial customers, with packets from the two sources freely commingling.

Line sharing is applied to international links as well. NSF joined the EBONE consortium as an informal "supporting organization," the same category chosen by CERN and IBM. Supporting in this case meant that NSFNET and EBONE would be connected, but that NSF would not sign the memorandum of understanding.

Line sharing in EBONE meant that traffic from AlterNet and NSFNET might share the same lines. Government regulations prohibited NSF from funding commercial traffic, and this was the root cause of many of the appropriate use restrictions on the Internet. Allowing traffic from commercial services to share a link with NSFNET, provided the commercial service paid its own way, was an important step forward.

Acceptable use policies were set up in the era of the physical networks when NSF leased 56 kbps lines. The idea was that NSF should not be subsidizing traffic generated by those outside the community it meant to serve‹the scientific and engineering researchers.

This policy was getting harder and harder to enforce, particularly if the enforcement mechanism was to bar all traffic originating from certain broad groups from traversing key links. With NSFNET simply paying for service, it no longer really mattered if packets fraternized, so long as the educational and research communities got their money's worth.

Wednesday morning, I walked a few blocks north to Dupont Circle to the home of Bob Barad, founder and sole employee of Baobab Communications. Bob runs the Baobab bulletin board, a Fidonet node specializing in African development.

His house is full of old maps of Africa, and blankets, baskets, and masks collected during his two years in the Peace Corps in Sierra Leone and on many subsequent visits to the continent. After his two years in swamp rice development in Sierre Leone, he came back for a law degree, then returned to Togo on a Fulbright scholarship.

Following that he spent a couple of years at the World Bank, an experience he diplomatically called "an extension of my education." Bob started spending his spare time dialing into bulletin boards. He got an account on PeaceNet and found he could start talking to people in Africa. He was hooked.

Soon he had hung out his shingle as a freelance development consultant, a business that seems to draw heavily on his Fidonet expertise. He had quickly tired of being a user of other peoples' boards and wanted to set up his own node.

Bob had been corresponding with a digital nomad named Mike Jensen, a man who travels the world with a laptop, setting up Fidonet nodes. Jensen blew into town one day late in 1989, stayed at Bob's house for a day, and Baobab was in business.

Fidonet is a network which can be thought of as the PC equivalent of the UNIX USENET system. It provides mail, file transfer, and news feeds, all managed by an informal volunteer hierarchy of city, regional, and national hubs. The free software and the cheap platforms it runs on have attracted the attention of many human rights, environmental, and other public interest groups.

Being able to set up a network anyplace there was a PC had proved valuable in Bob's business. He had just completed a job putting in nodes between Chad, Mauritania, and a group in Arlington, Virginia. The Virginia group was a subcontractor on a U.S. AID program to provide a famine early warning system.

Satellite images of the weather are periodically broadcast by the European Space Agency. Data would be downloaded and processed, resulting in graphic images of the affected areas that would be placed in files totaling roughly 70 kbytes. These files would be transferred to floppy disks and sent by DHL to Chad and Mauritania for use by regional workers.

This was a situation ripe for automation. By elimination of the weekly DHL runs, the payback from the use of computers could be immediately quantified, a situation very different from those trying to justify the more nebulous benefits such as "increased interaction" resulting from general-purpose e-mail.

The Fidonet software was successfully installed and the first files transferred. The system worked so well that one site immediately requested the previous week's file, which had not yet arrived by DHL.

The transfer of data uses the zmodem protocol on a direct call between the African sites and Arlington. Direct polling to the end destination is not unusual in Fidonet. While it is possible to hand off messages to a local gateway for percolation through the network, high priority data usually cuts through the hierarchy. To send mail to Africa, for example, Bob placed a biweekly call to the London hub, a site only a hop or two away from the African destinations.

The number of bulletin boards like Baobab in operation all over the world is truly incredible. Merely listing the boards in ASCII format makes for a file of 1.44 Mbytes. One of the great challenges of the Internet will be to give those BBS systems an easy way to integrate into the global network, providing better connectivity than just mail delivery, but doing so in a way that is cheap and easy.

I returned to my hotel to check out, only to discover that the window to my car had been broken in the overnight parking garage. I filled out forms for the parking garage, the police, Hertz, and American Express in order to get my keys back and sit on a pile of broken glass, sucking wind on the drive to Dulles Airport.