Internet pioneer: IPv6 transition needed

Network World, 10/25/99

The Internet engineering community is promoting a new version of the Internet Protocol - Version 6 - as the answer to the address shortage predicted for the current Version 4. IPv6 offers enough addresses that every computer, cell phone and set-top box can be hooked up to the 'Net. However, migrating a large network to IPv6 is so difficult that few organizations have committed to it. In an interview with Network World Executive Editor Doug Barney and Senior Editor Carolyn Duffy Marsan, Internet pioneer Vint Cerf argues that network managers need to start making the transition to IPv6 immediately.

More of this interview

Why is IPv6 the right direction for the Internet?

The Internet is growing very, very quickly, and we are very concerned about running out of address space in the Version 4 network, which has a 32-bit address field. Theoretically, Version 4 could support up to 4.2 billion devices, but the allocation of those addresses has not been very efficient. We tried to increase the efficiency with interdomain routing and allocation rules that go along with it. But the side effect of those rules is the proliferation of network address translation [NAT] boxes, which take a single Internet address and multiplex it among a bunch of different devices. It's a fairly ugly process from an architectural point of view, although it turns out to be very effective, and a lot of people are relying on it. But because NAT intervenes at the IP address level, it has some consequences for end-to-end security and integrity of the traffic. Many of us would just as soon solve the problem of address space by having a much larger address space to draw upon, and that's what IPv6 is all about. It has a 128-bit address field, and that allows for 10 to the 38th power possible addressable devices. We should be moving toward IPv6 promptly, and we need to start now because the transition is fairly complex.

Many people in the Internet community - including well-respected engineers and analysts - think that IPv6 is not a practical solution. What chances do you give IPv6 for succeeding?

It may very well be that the only way to get to Version 6 is for NAT boxes to convert Version 6 addresses to Version 4 addresses and back [for a while]. NAT boxes are turning out to be the path by which we get to Version 6. I challenge those who think we don't need to [move to Version 6] to come up with an alternative strategy that's achievable in the next several years.

What will it take for IPv6 to succeed?

All of the vendors of software in the edge devices have to believe that we need to [migrate to Version 6] and have to support it. One of the most prominent is Microsoft because of the huge number of devices that use Microsoft software to interact with the Internet.

All the router vendors have to pay attention to Version 6 so we can build a Version 6 core. Another set of players is the ISPs. A state of denial exists among some ISPs, who would just as soon not face this problem. Version 6 is workable, even if we have to make use of NAT devices in order to accommodate the mixture of Version 4 and Version 6 for a period of time in the network.

What if IPv6 fails to catch on?

We're still confronted with the problem of running out of Version 4 address space. What happens if there are so many NAT boxes that you can't uniquely identify them all with Version 4 addresses? Then we're back in the soup again. This is not a problem you can ignore.

If IPv6 is the right move, why don't users want it?

Most people who are doing applications haven't the foggiest idea of what the IP address space looks like and whether there's a risk of Version 4 vs. Version 6. If anybody should be paying attention to this, it's the ISPs. Most of them are betting the farm on NAT boxes in the near term.

What are the global ramifications of not adopting IPv6?

We can see the demand for hundreds of millions of devices on the 'Net already. Cell phones that are Internet-enabled. Cable set-top boxes and other appliances that become Internet- enabled. The people who are building [these devices] are getting IPv6 allocations. If we don't use IPv6, we'll have to use something else that gives us large address space. It took us quite a while to get to IPv6. There were a lot of debates, a lot of discussions.

The end result is a pretty strong design. So if we don't adopt it, we will wind up having to do yet another cycle of design and agreement. As we do that, the lifetime of the IPv4 address space is getting shorter and shorter. That's what I mean by being back in the soup.

The Electronic Privacy Information Center has raised some concerns about IPv6 because its 128-bit addresses include the user's unique Ethernet address. The center compares IPv6 to Intel's Pentium III chip. Are these privacy concerns valid?

I believe they're based on a misunderstanding. My experience is that devices that slip into a PC such as Ethernet cards, people swap these around. I don't see anything binding in the MAC address. The MAC address can be overriden by software. So there's too little certainty to associate a given MAC address with a given PC. If the MAC addresses are dynamically changed, that will remove a significant amount of concern. The trade-off is simpler configuration of the `Net. As opposed to having to call up a LAN administrator, you'd be able to just get on [with IPv6.]

What can we expect to see next from the IPv6 Forum?

My role continues to be as honorary chairman. I've committed to working with hardware and software vendors to try to outline where I see the network headed in terms of demand on address space and the need for dynamic configuration. As we start to see hundreds of millions of devices on the network, many of them simple consumer devices, the ability to configure automatically is critical. We can't afford to send trained teams of people out to install the IP-enabled washing machine. It's gotta be all automatic and all dynamic.

What role should ICANN play with regard to IPv6?

ICANN's job is to oversee the policies by which address and domain names are assigned. I don't see them as an advocate for one protocol over another. But I do see them as responsible for the address assignment process.