Future Internet Structures

By: Scott Bradner

Just to prove that the press can do the reader's bidding, at least every now and then, this column is in response to a reader's request. In an email message, Michael Moravan noted that there had not been much attention given to current status of the 'new NSFNET'. He asked if I'd bring people up to date.

This column will focus on the U.S. part of the Internet. This is a bit misleading since, at the current rate of growth, the non-U.S. Internet will surpass the U.S. part in the near future. However, the U.S. Internet is complex enough as it is.

Currently Internet connectivity within the U.S. is provided by approximately two dozen larger providers and a fast-growing number of smaller ones (note that the terms larger and smaller are a bit vague). A number of the larger ones started life with the support of the National Science Foundation to provide Internet service to the Research and Education (R&E) community and are commonly known as regional networks. The rest of the large providers are commercial service offerings and have never had an R&E focus.

The regional networks currently make use of the NSF-provided NSFNET for their inter-network connections, where the users of those connections agree to restrict their usage of the NSFNET to R&E traffic or traffic in support of R&E activities. The commercial networks and those regionals that have expanded beyond their R&E base make use of the Commercial Internet Exchange (CIX) and MAE-east as interconnect points. The CIX is a multi-port router located in the San Francisco bay area and MAE-east is a metropolitan area Ethernet in the Washington DC area.

NSF is about to get out of the general R&E backbone network business. It is shutting down the NSFNET. The NSF is contracting for a new network service, the Very High Speed Backbone Network Service (vBNS) but since the usage of that will be very restricted, the vBNS will play no substantial role in the Internet of tomorrow.

Since the NSFNET will no longer be available and the vBNS will be out of bounds, there will have to be some other way for the regional networks to exchange traffic. NSF has proposed establishing a number of traffic exchange points around the country. NSF has called these points Network Access Points (NAPs) and has designated four specific ones, near Washington DC, near New York City, near Chicago and near San Francisco.

In order to obtain interconnectivity the regional networks can make their own connections to a NAP or they can contract for a connection from some other network service provider (NSP). Of course, in order to ensure full connectivity they would need to attach to all the NAPs or purchase service from an NSP that did. As far as I know, all of the regionals with the exception of one have chosen to purchase service from NSPs.

Since NSF was providing the connection to the NSFNET at no cost to the regionals and no one has found any NSP willing to do the same, the regionals will suddenly be faced with costs that they previously did not have. In order to ease the transition, NSF is offering to help fund each regionals's NSP connection, but only to the extent that it is used for R&E traffic and at a decreasing percentage of that until the support vanishes after four years.

There is one additional part of the NSF plan. Since the task of figuring out just how to get all that data traffic from here to there will be an increasingly semi-impossible task as the Internet increases in its complexity, it would be far better to have a single organization responsible for providing routing information at the NAPs, the DMZs of the new Internet. NSF is designating a Routing Authority to assume this role.

The process of migrating to the new structure is starting. NSF is due to make awards to the regional networks very soon. You can expect to see a number of announcements in the next few weeks about who is buying what from whom.

There are some unknowns at this point. The most troubling is just how strong the interconnections between the commercial networks, the international links and the NAP-based part of the Internet will be. It is to the benefit of all that this work well but we might be in for a few interesting detours on the information highway.