Making the change
At a technical level, the first prerequisite is support for IPv6 in operating systems and network hardware.
Operating systems with support for IPv6 include AIX, FreeBSD, HP-UX, Linux, Mac OS X, NetBSD, OpenVMS, Solaris, Tru64, and Windows (.NET Server 2003, XP and CE .NET). This takes care of most popular platforms, providing you are running a reasonably up-to-date version. Router vendors already supporting IPv6 include 3Com, Cisco, Ericsson, Extreme, Hitachi, and Nortel.
The ability of IPv6 devices to autoconfigure their addresses in the event of an ISP change is useful but not sufficient for enterprise users, who will need routers and firewalls to autoconfigure in a similar way in order to minimise the amount of administrative effort.
Any software that uses hard-coded IP addresses will need to be changed manually.
In fact, the software side may be the bottleneck. If a program stores internally an IP address for any purpose, the code and the data storage will need to be changed as IPv6 addresses take up more space.
It’s Y2K revisited, but without the hard-and-fast deadline. Similarly, routines that format or check the format of IP addresses as strings will need modification.
Danny Thomas of ITS, University of Queensland, has pointed out that it makes sense for applications to be created in such a way that a single version supports IPv6 and IPv4. The new protocol would then be automatically adopted as soon as it is available on the host running the program.
Three main approaches take care of the transition from IPv4 to IPv6: dual stack, tunnelling, and translation. A dual stack handles both IPv4 and IPv6, selecting the appropriate protocol according to the result of a DNS lookup: if it is an IPv4 address, it uses IPv4; if it is an IPv6 address, it uses IPv6.
Tunnelling provides a way of connecting pockets of IPv6 through the existing IPv4 Internet. IPv6 packets are encapsulated into IPv4 packets for carriage. This is the technique originally used by the experimental 6bone network, but it is gradually migrating to native IPv6 links.
According to Cisco, enterprise users can begin their evaluation and assessment of IPv6. It suggests the main reasons for a relatively early evaluation include planning for mobile IP applications (which may perform and scale better with IPv6), address space expansion to support IP telephony, and to take advantage of security and QoS features that are not easily applied in IPv4 networks using techniques such as address conversion, pooling, or temporary allocation.
But be warned: “It’s not a trivial task to upgrade an enterprise network to IPv6,” says Simon Newstead, systems engineering manager, Australia and New Zealand at Juniper Networks.
Two approaches are available for these trials. An organisation can set up an IPv6 domain and link it to one of the experimental networks such as the 6bone, or set up two or more IPv6 domains and link them via the existing IPv4 infrastructure.
Tasks include setting up (or gaining access to) a DNS with IPv4 and IPv6 support, a protocol translation mechanism such as NAT-PT to allow communication between IPv6-only and IPv4-only hosts, and configuring the routers appropriately.
Several different approaches exist. Cisco suggests small enterprise networks—most Australian enterprises would fall into this category—would use dual stack backbones. The advantages are that it is easy to implement for small campus networks with a mixture of IPv4 and IPv6 applications, but the downside is that managing both protocols is complex and the upgrade effort becomes significant as network size grows. However, the company does not recommend an overall upgrade to a dual-stack network at this stage, though it is a “valid deployment strategy” where IPv4 and IPv6 support is required.
Tunnelling to connect IPv6 islands is a low-cost, low-risk strategy, according to Cisco, although management is again relatively complex.
When these connections go beyond the building or campus, “it’s important to look at what carrier services are offered—ideally, you want a direct IPv6 connection,” says Newstead.
Although IPv6 has quality of service (QoS) features, network managers will still need a way to ensure that particular applications or users are given priority. For example, Packeteer is already planning to add IPv6 products to its range of packet-shaping devices. Senior technologist Mike Morford says the company is already doing IPv6 packet classification and has completed its tagging architecture. The final step will be native IPv6 support. “That’s not a big deal,” he says, “we’re largely waiting for a market driver” before launching products. IPv6 support could be delivered as a software update for existing devices, he says. “It’s obvious to us that IPv6 is happening . . . we’d be excited to see it take off.”
Timetable
IPv6 addresses were officially released in 1999, and the first one was allocated to ESnet, a network operated by the US Department of Energy. ESnet is one of the first large-scale (as in hundreds of thousands of users) networks to run dual stack IPv4 and IPv6.
“By working with Juniper Networks, IPv4 and IPv6 now coexist on the same routers in our network, which reduces operational expenditures and enables us to easily migrate to IPv6 services as required,” says Michael Collins, network engineering services group leader for ESnet.
The move to the new protocol has begun in a small way, with mainly experimental pockets of implementation in various parts of the world including Australia. (We were unable to identify any substantial sites that could serve as a case study.) The shortage of IP addresses is likely to hit first in some parts of Asia (including Japan, where the government has set a 2005 deadline for the switch to IPv6) and parts of Europe.
Newstead says the first large-scale rollouts of IPv6 in the Asia Pacific region are beginning, driven largely by demand for mobile services in Japan and China. Japanese telco NTT has deployed IPv6, and offers customers direct and tunnelled connections.
Australia and New Zealand are “probably lagging a little bit behind, but we don’t have the drivers”. For example, Telstra was able to secure a large block of IPv4 addresses, he says, and a trend among corporate users to employ IP VPN services from carriers makes it easier for those organisations to stay with IPv4. The next two to three years will see “fairly sizable production deployments” of IPv6 in Australia, he predicts, but warns, “It’s not a trivial task to upgrade an enterprise network to IPv6.”
Roland Chia, national business manager—networking at Dimension Data agrees. “There’s no business justification or drivers” pushing organisations towards IPv6, he says. Converting to the newer protocol means “a major organisational expense” at a time when most businesses are very concerned with costs and are looking for quick paybacks.
About the only things likely to change that outlook would be carriers forcing the use of IPv6 for mobile devices, or a sudden boom in “always on” Internet connectivity for embedded systems and electronic devices. Otherwise, the latest estimates are that we won’t run out of IPv4 addresses until 2007, he says.
If that’s the case, there’s time for the economy to get out of the doldrums before organisations have to invest in IPv6.
Super address
An IPv4 address is usually expressed as, for example:
192.168.0.1
whereas an IPv6 address looks like
3ffe:0501:0008:1234:0260:97ff:fe40:efab.
Any preceding zeros can be omitted, for example:
3ffe:501:8:1234:260:97ff:fe40:efab.
Two consecutive colons can be used not more than once in an address to represent sufficient zeros to fill out the address:
3ffe:501::efab
would expand to
3ffe:0501:0000:0000:0000:0000:0000:efab.
IPv4 addresses can be made compatible with IPv6 by prefixing them with 80 zero bits, for example:
::192.168.0.1.
Note that the familiar “dotted decimal” format can be used instead of converting to hexadecimal.
However, the common use of host names such as www.zdnet.com.au rather than IP addresses means that the change from IPv4 to IPv6 will be invisible to most users.
