B O T T O M L I N E

Upside to IP Fees

With no more government subsidy, the InterNIC will start charging for IP addresses. But these fees may improve the Internet routing system.

by Joel Snyder

The instinct of ownership is fundamental in man's nature. --William James (1902) Things are going to change for ISPs and companies connected to the Internet. IP addresses, which had been available for free, are going the way of domain names and will involve a fee.

IP addresses--those little 32-bit numbers that live in every packet on the Internet--are vital to the operation of the Net. Every computer on the Net has to have a unique IP address, and the uniqueness of the address has to be coordinated by some central authority, usually called a registry. Even if you have a firewall, you need some small number of IP addresses to represent your network.

If you dial in via modem to connect to the Internet, your ISP assigns you an IP address for the duration of your session. When you disconnect, the address goes back into a pool for the next person who connects. Computers connected permanently to the Internet (via a T-1 line, for example) usually have permanent IP numbers.

IP addresses and domain names are different animals. IP addresses are embedded in your TCP/IP software. Domain names, like opus1.com or whitehouse.gov, are mostly for vanity purposes (a nice sounding e-mail address or URL) and may or may not map to one or more IP address. To keep IP addresses globally unique, there has to be a top-level registry in charge of making sure that no two organizations use the same address. In the Internet, that place is held by IANA, the Internet Assigned Number Authority. IANA doesn't actually give most organizations addresses. It delegates huge blocks of addresses to lower-level registries to pass out to Internet users in their geographic areas.

Addressing the Issue

In North America, the registry for IP addresses is the InterNIC, the same company in charge of passing out domain names. The InterNIC gives out addresses in blocks, the smallest of which is 256 addresses. This chunk of address space is often called a Class C IP Network Number, somewhat incorrectly because not every block of 256 addresses is a Class C network number. If you hang out in technical discussion groups, you'll see the blocks referred to as a "/24" because the first 24 bits of the 32-bit IP address are the network number, which leaves 8 bits for the host number, and 8 bits can hold 256 different hosts. But I digress. . . .

If you received a block of 256 addresses from the InterNIC tomorrow, it wouldn't help your company get on the Internet because what's most important in IP addresses is their "routability," and newly assigned blocks that small are not routable. It does you little good to have a globally unique IP address if everyone else on the Internet doesn't know how to get to you. Knowledge of the path, or route, to each IP address is stored in routing tables that are entered into enormous routers at the core of the Internet. Because these tables have ballooned with the growth of the Net, the folks who run the core routers will only store routes for new IP addresses in larger blocks, typically of at least 4,000 addresses.

But that's not a problem. You don't get your IP addresses from the InterNIC; you get them from whoever sells you Internet connectivity--typically your Internet service provider (ISP). The ISP gets a big block of addresses from the InterNIC and passes them out to its customers. Each block of IP addresses, no matter what the size, takes up the same amount of space in the core routers.

Here's an example. My company has two blocks of 256 addresses that we received about 10 years ago when we first connected to the Internet. Each block takes up an entry in the global routing tables, so we're using two entries. Suppose we were to give up those addresses and use some from the much larger block the InterNIC gave our ISP. We would still have the same quality of access as before; we would just be considered part of our ISP.

All our current addresses take up a single entry because all the core routers need to know how to get to our ISP. By giving up our two blocks, we would shrink the core routing tables by two entries. That doesn't sound like a lot, but multiply it by tens of thousands of inefficiently used blocks and there's potentially considerable savings.

Until the first of this year, all address blocks of all sizes were free. But no more. Sometime in 1997, there's going to be a fee. Actually, IP addresses have always cost money. Outside of North America, organizations that wanted to connect to the Internet have had to pay for the privilege of having IP addresses for several years. But the InterNIC, from which North American companies get IP blocks, has been subsidized by the U.S. government. All you had to do to get a block of addresses was ask.

With the U.S. government's subsidy going away, the InterNIC has started to charge for its services. First came domain name fees, instituted in late 1995. Now the InterNIC is proposing to spin off the IP address assigning process to another newly formed nonprofit company called ARIN (American Registry for Internet Numbers). ARIN is going to have to charge for addresses to stay in business.

I think it's a fantastic idea. Of course there are people who are upset about this, but that's always going to happen when something that used to be free suddenly costs money. The Internet has its share of crazies, and this proposal has brought them out in full force.

Unhappy Netters

The first week after ARIN was announced, the North American IP Registry (NAIPR) discussion list was flooded with folks threatening lawsuits, espousing conspiracy and economic theories that wouldn't pass muster in a kindergarten.

To put your five cents in, enter subscribe NAIPR in the body of a message and send it to listserv@lists.internic.net. One of the problems of the Internet is that a lot of people have rushed to join without really knowing what they're getting into. Because people often don't take the time to understand before acting, a lot of poor technical decisions have been made along the way. In some ways, the Internet provided no incentives to make the right decisions.

If IP address blocks had an annual cost associated with them, we would have incentives to make good technical decisions. For example, if you were an early Internet adopter you might have gotten a 256-address block 10 years ago. That block now represents an inefficient use of global routing resources. Starting soon, you will have to pay $2,500 per year to keep those addresses. (This is a proposed price and is subject to some adjustment before the new IP scheme goes into effect.)

However, if you were willing to renumber your systems and get a block of addresses from your ISP, your annual cost could drop to $250, $100, or less. Why? Because the more numbers you get, the less the per-number cost. The ISP buys a bigger block for $5,000 or $10,000 per year and has lots of smaller blocks to hand out.

If this cost structure provides incentive for small businesses to turn in their old number blocks, the size of the global routing tables (which have been growing constantly) will hold steady or even start to shrink. That's because all of the IP address blocks "downstream" from an ISP are aggregated into a single routing entry when they come from one larger block.

Of course, this scenario is not without a downside. If the business that turns in its old block of addresses wants to change providers, it will have to renumber all its machines because the smaller blocks are not portable across providers. But for a savings of $2,500 a year, a lot of businesses might be willing to go through a hellish long weekend.

There's also some discrimination against small ISPs. A typical ISP address allocation will cost about $5,000 a year, which is a lot to spread across 10 or 20 customers. But small ISPs all have to buy service from big ISPs--the Sprints, MCIs, and UUNets of the world. The big ISPs will have bigger blocks to split up, and the cost per address will decline. The smaller ISP will be linked more closely to the bigger ISP because if it wants to change providers it will not only have to renumber all its own machines, but all its customers' machines as well.

Beneficial Effects

I believe this kind of incentive will make the Internet a better, healthier, and faster network. Some people are disturbed by the idea of ARIN collecting the money and being insufficiently accountable to the North American community, but I'd almost be happy to see ARIN burn the money it collects. What I like about the plan is the economic incentive it gives to ISPs and Internet users to be more efficient. Where the money will go--well, we can work that out later.

Don't ignore the success Europe and Asia have had in charging for IP addresses. Through Western and Eastern Europe and parts of Africa, RIPE (Réseaux IP Européens, has handled allocations for several years with great success.

The InterNIC has deposited more than $10 million in an account from the monies it has collected for domain name registrations. Any ARIN excess funds could go in that account, which, hopefully, will be spent on some worthwhile infrastructure projects soon. The initial ARIN charter also mentions the possibility of dropping prices if too much money is collected. That may be OK, but I'd like to see a couple of years of relatively high prices to shock the Internet into better routing behavior first.

Even though it looks like the IP-fee proposal is going to increase the cost of the Internet, it could eventually cause costs to drop. ISPs that used to have to buy $80,000 routers to handle the enormous routing tables might be able to get by with $10,000 routers. So even if they have to pay $5,000 a year for IP addresses, they could save $70,000 by not having to buy bigger and faster equipment.

I'm not necessarily predicting that this will happen; I just want to point out that there are some potentially nice side effects when you introduce incentives to conserve routing-table entries. The Internet works as well as it does because the people who run it generally are cooperative and have had the best interests of the network at heart. With additional incentives like this, I believe things will only get better.


Joel Snyder is a senior partner at Opus One in Tucson, Ariz.
Reprinted from Internet World magazine Vol. 8 No. 2 (c) 1997 Mecklermedia Corporation. All rights reserved.

http://www.internet.com/

Internet.com home