The next version of IP is making slow progress

The internet is a great place for history. For example, search for "IETF RFC 791" and you will find the original specification document for version 4 of Internet Protocol, as edited by John Postel, who shepherded the internet through its formative years.

The internet is a great place for history. For example, search for "IETF RFC 791" and you will find the original specification document for version 4 of Internet Protocol, as edited by John Postel, who shepherded the internet through its formative years.

IPv4 is still going strong, but as with many other things internet-related, new standards threaten to supersede it. IPv6 has been looming in the shadows for 10 years now, but has yet to step in to the spotlight. Forget about IPv5; it was just a streaming protocol that was not used commercially.

One of the biggest drivers for IPv6 has been the impending shortage of available IP addresses. IPv4, the address space currently used by most devices on the internet, is just 32 bits long, which limits it to a maximum of about 4.3 billion addresses.

You would think that was more than enough, and at present it is, but the addresses are expected to run out at some point in the next 15 years, as more people continue to request them, and as devices become increasingly internet-enabled.

The squeeze on IPv4 addresses is increasing due to rapid development in Asia. Until 2001, North America obtained more IP addresses than anywhere else in the world, but in 2002 Asia pulled ahead.

Since then, more IPv4 addresses have been registered through the Asia Pacific Network Information Centre (Apnic) than any of the other five global regional internet registries.

Now, according to Apnic's annual report, China is rapidly overtaking Japan as the major consumer of IP addresses in the region. As China and its billion-strong population modernises, it is easy to predict rapid uptake of mobile phones, PCs and voice over IP, all of which will require an increasing number of IP addresses.

Enter IPv6. From its inception in the mid-1990s, the supposed successor to IPv4 was designed to address the capacity problem. Instead of 32 bits, an IPv6 address has 128 bits, creating 340 undecillion addresses. If you wanted to give an IP address to every blade of grass on the planet, you would still have enough left over to keep the world in unique internet addresses for a very long time.

Although the larger address space is seen as the biggest driver, there are other reasons to adopt IPv6. For example, according to Patrick Grossetete, manager of product management at network supplier Cisco, auto-configuration is a huge advantage.

IPv4 uses Dynamic Host Configuration Protocol (DHCP) to allocate addresses to new devices joining the network, which means network managers must define address ranges for the DHCP servers to work.

IPv6 devices will obtain their own addresses, listen for the router on the network, and couple with it. It represents true plug-and-play on the network without any back-end configurations, said Grossetete.

This would be useful in the enterprise, but to hammer the point home, he described a future home networking scenario.

"You have many suppliers coming in from Japan or Korea with LCD screens, DVD recorders and game stations that include an Ethernet or Wi-Fi connection," he said. "With IPv4 you are asking the end-user to manage a DHCP router, and most people do not know what DHCP is."

IPv6 sounds like a networking Utopia. So why is nobody using it? "In the enterprise there is not much uptake at all," said Alan Miller, vice-president of product management at Allied Telesyn, which makes core networking equipment for telcos.

The problem is that enterprises have other things to spend their money on rather than universal upgrades to their entire network infrastructure, just to remove the need for DHCP configuration.

"There is not a great deal of service being offered by service providers either," said Miller. He added that when they climb on board, service providers will probably be the initial drivers for IPv6 because the increased address capacity will be beneficial to them.

The world's telcos may not be jumping in with both feet, but there is some burgeoning service provider activity in IPv6. BT Retail is already running an IPv6 exchange, said Steve Masters, general manager for IP infrastructure.

There are 60 customers accessing the IPv6 exchange through a tunnelled network, and 15 more accessing it natively with IPv6. Masters' best guess is that the tunnelling companies are corporations accessing the exchange via a service provider that is not IPv6-compliant, meaning they have to tunnel in through an IPv4 system. The 15 native connections are service providers.

Still, aside from such pockets of activity, there is little real-world use of IPv6. Apart from the fact that there are still at least eight years' worth of unallocated IPv4 addresses, network address translation (Nat) has also alleviated the problem to some extent.

Nat uses a machine at the perimeter of the network, which represents the whole network with a single IP address. Every IP-enabled device behind the Nat machine has a private IP address that cannot be seen from outside the network, making Nat not only a means of sharing an IP address, but also a security measure.

With companies the world over assigning hundreds of private IP addresses to a single public IP address, the problem of address shortages begins to look more manageable, further weakening the case for IPv6.

However, Nat has its problems, said Paul Mockapetris, chief scientist of domain name system server software supplier Nominum. By blocking individual IP addresses from the network, Nat makes it difficult for direct peer-to-peer networking, which affects various applications, including some forms of VoIP.

"VoIP suffers when it goes through the Nat boxes. So if what you had was IPv6 with unlimited address space and you had some substitute for the NAT boxes, it would make IP telephony easier," he said. "But that is a steady pressure, and people find workarounds."

Such workarounds include application layer gateways that help to support different protocols, but, like NAT, they are quick and dirty fixes, conceptually speaking. They create problems while solving others, said Rob Turner, product marketing manager for enterprise Ethernet switching at Nortel.

"Today you might have a gateway that is doing address translation for different protocols across the Nat. But that is a single point of failure," he said, adding that application layer gateways also have to be updated as protocols change.

IPv6 theoretically removes such problems, but there are significant barriers to migration. "Most internet clients cannot talk to servers with IPv6 addresses," said Ben Robb, lead technical consultant at internet consultancy cScape.

"Until there are lots of clients that are able to communicate using IPv6, there will not be enough of an audience out there for servers to warrant upgrading."

To be fair, Windows XP Service Pack 2 does include support for IPv6, nevertheless, companies will surely find elements of their network unable to support the new protocol.

Migration strategies vary. Some advocate running a dual IPv4/IPv6 stack to comply with IETF RFC 2893, which defines standards for such equipment.

This still requires equipment to be updated, and is likely to introduce more complexity into the network, but it at least allows for a slower roll-out with piecemeal equipment upgrades over time.

Other methods are tunnelling, where you encapsulate an IPv6 packet into an IPv4 network, and translation, which uses a device to translate between the two protocols on the network.

Whichever equipment you choose, beware of potential flaws, warned Turner. Nortel implements its IPv6 support in hardware, but many suppliers do it in software on generic chips, rather than using custom fabricated Asics (application-specific integrated circuits) to support the technology at line speeds, he said.

Even if the device does claim hardware support for IPv6, be sure that it works properly. Robb pointed out a post from a disgruntled Cisco Catalyst 3750 switch user on the O'Reilly network, who claimed his unit offered "unadvertised features" with regard to IPv6, including the inability to use policy routing with IPv6, and the inability to boot the router with IPv6 interfaces configured.

"We would like to offer IPv6 natively (non-tunnelled) for the university, but certain suppliers definitely do not make it easy," posted Charles Schluting, a network engineer at Portland University in Oregon, who also writes technical books on networking for O'Reilly. Cisco refused to comment.

You will have to wait a few years before IPv6 becomes prevalent, but some developments are gradually moving it along. For example, the US government has mandated a move to the protocol for all federal agencies by 2008. It is unlikely to affect most UK companies yet, but if nothing else, it is good to know what is coming down the pipe.


Read more on Voice networking and VoIP