The Problem of Being Permanent

By: Scott Bradner

Three predictions were made to the IETF back in 1990: 1/ the Internet would run out of some types of Internet Protocol (IP) addresses "soon", 2/ the entire 32-bit IPv4 address space would be exhausted soon after, and 3/ that the rate of growth in the size of the tables in the backbone routers was growing at a rate that could not be sustained for very long. The IPng process was started to address #2, and after quite a bit of discussion, the concept of Classless Inter-Domain Routing (CIDR) was approved to do two things: 1/ to permit the assignment of blocks of IP addresses sized to more closely match the needs of an organization so that addresses would be more efficiently used, and 2/ to enable the aggregation of routing information from multiple organizations so that the growth of the routing tables could be slowed. Note that IPng does not, by itself, change the dynamics of the routing table growth.

Using CIDR, IP addresses are no longer assigned directly to organizations in the old Class A, B, and C sized chunks, instead large blocks of addresses are assigned to network providers. These providers assign parts of these blocks to their customers. If the customer is another network provider, they in turn assign parts of their part to their customers. At the end of this tree, end-user organizations are assigned ranges of addresses which are sized to closely match the organization's actual requirements. This assignment process results in hierarchical addressing, where the topology of the underlying network is reflected in the IP addresses themselves. Hierarchical addressing permits the providers to aggregate the routing information and only advertise the whole chunk of addresses to the rest of the Internet, rather than having to advertise each individual organization's network addresses.

In theory, this would mean that the routing table in the backbone of the Internet would only have to have one entry for each of the major network providers. Things are not quite that way in reality since the Internet does not have a simple tree structure and the use of CIDR-style address assignments is quite recent The initial address assignments in the Internet was first come first served with no concern over network topology. In addition, many organizations have switched providers over the years, so even if they had been hierarchically assigned, the address structure would not reflect the network topology any longer. Thus, the ability to aggregate the routing information in the internet is currently somewhat limited and the routing tables are growing quite rapidly. But even with the current limitations, the use of CIDR has meant that the routing table in the backbone routers contains about 30,000 routes instead of the 65,000 that would have to be accommodated without CIDR.

At this point it seems that the best way to moderate the growth in the routing tables, which were growing faster than memory technology itself, is to request, but not require, that organizations renumber their networks to match their new place in the network topology when they connect to a new provider, ether by moving from an old provider or when they connect up the first time and are using address they were assigned in pre-CIDR days. A number of providers are already doing this with a lot of success.

For large networks renumbering can be quite a pain and an expense. The IETF is starting a new mailing list (pier-request@isi.edu to subscribe) to look at ways to make the process easier. Note that this means that an organization does not "own" its CIDR-style IP address, but since people do not use IP addresses directly, the fact of non-ownership should not be an issue, though the pain and cost of renumbering can be, but so would be an Internet that has outgrown the capability to be a connected whole.

Disclaimer: This can't be Harvard's opinion since whenever you have N Harvard people you have 1.3N opinions.