Click the comments link on any story to see comments or add your own.
Subscribe to this blog
04 Jul 2011
Every packet of data sent over the Internet is sent from one IP address to another. The IP addresses in the Internet serve somewhat the same function as phone numbers in the US phone system, fixed length numeric identifiers where the first part tells what network the address is on. Since the dawn of the Internet in the early 1980s, the IP addresses in use have been IPv4, 32 bit addresses which means there are about 4 billion of them. Unless you've been living under a rock, you've doubtless seen reports that the supply of IPv4 addresses is running out. Earlier this month IANA, the master allocation authority, handed out the last so-called /8, a large chunk of 16 million addresses, to one of the regional address registries, and sometime months or perhaps a few years after that, the registries will hand out the last pieces of their chunks. Then what?
The conventional wisdom is that everyone needs to support IPv6, a mostly compatible upgrade to IPv4 with much larger addresses, by the time the v4 space runs out. But I'm not so sure, particularly for e-mail.
There's two unanswered questions here. One is is how hard it will be for new or expanding networks to get IPv4 address space. The other is how important IPv6 addresses will be to be able to reach the rest of the net. The conventional answers are very hard and very important, but I think the real answers to both, for the next several years, at least, is not very. Today I'll opine about getting IPv4 address space, tomorrow about addressing and reachability.
There has always been a rule that, approximately, nobody other than perhaps IANA "owns" IPv4 address space, and that it's allocated in the best interests of the Internet community. This means that applicants have to justify requests for more space by showing responsible use of previous allocations and a reasonable plan for what they want. It also means that in theory one could not buy or sell address space. In practice, the rules have routinely been circumvented. It's a continuing scandal that eastern European spam gangs run fake address space registries, and even within the US, dodgy ISPs are still willing to burn address space for snoeshoe spammers (ones that change addresses to evade blacklists) quite cheaply. Even under the nominal rules you could always sell an address block by wrapping a dead company around it and calling it a merger.
The rules also said that you were supposed to give back address space you weren't using, again with mixed results, e.g., Stanford and CMU were issued /8 networks with 16 million addresses, far more than they needed, and renumbered and returned them. But MIT also got a /8 and didn't, so they still have theirs. Now that everyone knows that address space is running out, nobody's going to hand anything back without getting something in return, and the registries are grumpily acknowledging that since there's going to be a market in address space whether they like it or not, there's nothing to be gained by refusing to admit it.
Nobody knows what the going rate for IPv4 space will be, but of course here in blog-land, that doesn't keep us from telling you anyway. One theory is that the address space is like choice urban real estate, a precious resource that rarely comes on the market and costs a fortune when it does. But if you look at how much of the allocated space is in use, by checking how much of it has routes announced to allow Internet traffic, about half of the allocated space is lying fallow. This isn't surprising -- back in the 1980s /8 networks of 16 million addresses were handed out to a lot of organizations that are using only a small fraction of what they've got, e.g. Eli Lilly and Halliburton. My guess for the price of address space is about a dollar per address, with large contiguous chunks being worth somewhat more, perhaps twice that for a really big one. MIT has 16 million addresses, but has nowhere near a million computers. The addresses they use are spread all over the 16 million addresses. It would cost something to go through and renumber down into the first million addresses (a /12), but if they could then sell the other 15 /12s for, say, $2 million each, after a while they'll do it. This situtation is typical throughout the net, with address space used sloppily since there was no reason not to. (Indeed, the more space you used, the easier it has been to persuade the registries that you needed more.) With a financial incentive to use space more efficiently, we'll see vast amounts of space freed up.
Another important issue beyond that of simply getting space is routing it. For an address to be usable on the Net, its ISP needs to tell the world how to route traffic to it. The "defaultless core" routers need to have a route entry for every address in use in the entire Net. A single route entry can cover anywhere from a /24 with 256 addresses to a /8 with 16 million addresses. If people subdivide up their address space, e.g., MIT sells off their address space in /12 chunks, there will be 16 routes where before there was just one. It's true, the core routers have a limited number of route entries, and it would be bad if they filled up. But there's a weekly survey of routes known as the CIDR report which counts both the number of active routes, and what the number of routes would be if each network combined overlapping or adjacent routes down to the minumum needed. The current size of the route table is about 300,000 routes, and the minimized number would be about 150,000. That is, half of the route table entries could be freed up right now if networks were less sloppy in their route management.
The obvious way to clean up the route table would be to charge networks for the number of routes they announce, but that turns out to have logistical problems. While it's easy to tell who's announcing what routes, it's much harder to tell who's running defaultless routers, and to figure out a settlement system to move money from the former to the latter. So far, the route table has been kept under control by rules of thumb, notably that few networks will accept a route for a chunk smaller than /24, and by hardware vendors building routers with ever larger route tables. Eventually the number of routes may outstrip the ability of routers to handle them, but the fact that there's currently 50% slop suggests that time won't be here any time soon.
But no matter what you do, if you have a network with a thousand users or a thousand servers, you need a thousand addresses which means before long ISPs won't be able to provide IPv4 connections to new users, right? Maybe, maybe not. In the next message we'll look at just how much public address space a network needs to serve its users.
My other sites
© 2005-2015 John R. Levine.
CAN SPAM address harvesting notice: the operator of this website will not give, sell, or otherwise transfer addresses maintained by this website to any other party for the purposes of initiating, or enabling others to initiate, electronic mail messages.