The following text is copyright 1993 by Network World, permission is hearby given for reproduction, as long as attribution is given and this notice is included.

The end of the world?

By: Scott Bradner

On and off for the last year or two there have been all sorts of "death by reason of success" stories about TCP/IP popping up all over the place. The basic theme of these stories is that TCP/IP has become too popular. The assignment of network numbers is proceeding at such a pace that the pool of one type of network addresses, known as Class B addresses, will be exhausted very soon. The first predictions, which surfaced almost two years ago, had the point of exhaustion occurring by now.

There are basically 3 types of network addresses used in TCP/IP: Class A network addresses which are capable of supporting 16 million hosts and are used for very large companies or even whole countries; Class B network addresses which are able to support 65 thousand hosts and are used in average sized companies and educational institutions; and Class C network addresss which can support 254 hosts and are used for small organizations or individual LANs.

The argument went that because of their size, most places required the use of one or more Class B network address. More than half of the potential Class B addresses had already assigned and the rate of assignment was increasing. Projecting the growth in demand showed that the supply would be exhausted in a year or two.

Somewhat forgotten by the popular press in this doomsday prediction was the fact that there were millions of Class C addresses available and organizations could use a whole bunch of Class Cs instead of a few Class Bs. However, there is unfortunately a real problem with this approach. The routers that hold the national and international backbone segments of the Internet together must maintain a listing of all reachable networks and a map showing how to get to each network. This map is known as a routing table. The routers exchange information and status messages to maintain the routing tables. Currently there are about 13,000 networks in the routing tables of the backbone routers. Increasing the size of these tables produces an unacceptable load on the routers themselves and increases the amount of overhead traffic required between the routers.

Things did seem rather bleak. A successor to TCP/IP would be required in the very near future. (Actually a successor to IP since TCP is not directly affected by the addressing problem.) A number of proposals were made to the IETF each vieing for the honor (if that be the right word) of being anointed the rightful heir. The pressure was high to make a quick choice since development time is necessary to implement and shakedown the required software. The IETF has been debating the proposals and debating what procedures should be followed in making such a far -reaching decision. One does have to be a bit careful when the result of your choice might have to be installed on a few million computers all over the world.

Working under a lot of pressure does not make a careful and deliberate decision any easier. However some of the pressure may now be off for a while. There is a relatively new proposal known as Classless Inter-Domain Routing (CIDR) which allows the aggregation of routing information. A company could be assigned 5,000 Class C network addresses but would only have to appear as a single entry in a routing table. The IAB, in RFC 1481, endorsed the CIDR architecture and its implementation .

With the assumption that CIDR will be deployed, multiple Class C network addresses are now being assigned in place of Class B network addresses. Projections based on the current assignment strategy and rate (the demand is still increasing) indicate that the supply of network addresses will be exhausted in about the year 2000. The use of CIDR does not remove the requirement to make a choice about a new IP but it can postpone the deadline for making such a choice. This will allow a more measured decision process and, one hopes, a better final product.

On a completely different topic. It has long been a tradition on the Internet mailing lists to end ones postings with a disclaimer. The idea is to be sure that no one confuses the author's opinions with those of the author's employers. This is something I should have been doing in this column from the start since Harvard is a large and varied place and it is quite unlikely that any opinion that I might put forth would be universally supported across the University.

I'm going to start a disclaimer of the month featurette using disclaimers gleaned from the Internet. One I will not be using is from Martin Cockerell of Whiteleaf Bucks in the U.K.; "Any views expressed are fully supported by the company. (*I own it*)"