IPv4 vs IPv6: The Road Ahead

jeffy124 writes "With the world moving towards having every device under the sun being Internet-connected, is the Internet going to be too large? This article off CNN.com examines this potential situation. They look into the problems of switching networks from IPv4 to IPv6, and the inclusion of inter-operability between the two. Benefits of moving to IPv6 are looked at, but so are the critics of it who point out that if we don't have a problem now, why fix it? While low of technical details, the story points out that not many systems out there currently support IPv6. "

Comments

Vital IPv6 links

[DaSyonic]

Some good MLP:

good IPv6 homepage

IPv6 HOWTO

IPv6 Standards

IPv6 Tutorial (PDF)

And the 6bone

Re:Not many systems support it?

[Frederic54]

every win32 system can have IPv6 for months using Trumpet TCP/IP stack, even win98, not "only" nt/2k that have ipv6 support from M$ also.

NAT doesn't solve the whole problem.

[David McBride]

Network Address Translation only provides one-way connectivity. It allows a system behind a NAT to establish connections from external sites and retrieve data.

What it *doesn't* allow is anyone out on the internet to go and connect to the machine behind the NAT, which is kinda essential for anything beyond web-browsing.

The internet is not just port 80. Many people treat it as such, and I hope they have fun. But don't delude yourself that you have a full internet connection, because you don't. You've just got a fancy TV with a few more channels.

NAT is a stop-gap measure at best. IPv6 is essential for allowing the internet to scale the way you want it to.

Think about it: it's not outrageous that MIT and similar institutions have class-A networks - it's outrageous that *you* don't. IPv6 can fix that.

Ask your ISP about their plans to upgrade to IPv6 - and what their IP allocation policy will be. If the ISP doesn't intend to give you lots of IPv6 addresses, start looking somewhere else.

Dynamic IP allocation sucks in the same way that NAT does. Many of the peer to peer projects nowadays, in order to keep functioning, have to build their own namespace and addressing structures just to work around it.

Re:More IP address !=more ease

[jd]

Doesn't work that way with IPv6. IPv6 uses a system whereby the provider announces a prefix to the device, and the device then attaches whatever number it likes to that prefix. If another device attaches to that first device, then that first device announces the prefix it got, PLUS the first 1-2 bytes of whatever it added, as the new prefix.

The theory behind all this is that you can then move a device from one network to another, without ever having to worry about routing problems, IP numbers colliding, or other such mundane trivia.

"Permanent" addresses, in this system, don't exist. They're all calculated.

How does this work, in practice?

Well, let's say that Joe Bloggs is connected to AOL. AOL decides that the backbone provider it uses can get stuffed, and switches. This changes all of AOL's addresses, and therefore Joe Bloggs' address.

However, because addresses have a lifetime attached to them, the old address remains active (although forwarded) for a finite length of time, although new connections to the old address are prohibited.

Because of this, it makes no sense for some central registry to store AOL's IP number. It can change once every 60 seconds, along with the IP address of everyone/everything connected via it.

The only person who can meaningfully store AOL's IP address becomes AOL, itself. Nobody else can possibly know it, with any reliability.

Normally, ISPs and large corporations aren't going to flip around like that. But they -can-. The protocol permits it. Because of that, and because uptime is increasingly important, they will then be able to shop for a secondary provider for a backup link, in case the first one dies.

In IPv4, a backup link via an alternative provider would be lethal. There would be no way to handle the changes in addressing, unless the entire ISP or company was behind a NAT system with High Availability at the IP level, which causes its own problems.

With IPv6, the change-over would take under 5 seconds for the whole of AOL. Nobody would notice the delay, nobody would get disconnected, and the whole setup is much simpler.

Re:More IP address !=more ease

[jd]

The people who designed IPv6 opted, right from the start, to make it as automatic as possible. Administrators should not be burdened with assigning IP addresses, making sure there's no conflict, or running DHCP and similar carp.

Further, they anticipated an increasing use of portable devices, such as laptops and hand-helds, which made it important to have Mobile IP a part of the protocol.

The result was the complete absence of any notion of static IP addresses. Addresses are assigned at connection time, and last until either they're revoked by the owner, or they time out. Once they reach that point, they are marked as expiring. A new address is then generated. The host machine is required to then notify ALL machines connected to it or that it connects to that the address is changing, and what that new address is. The remote machines then have a certain length of time (it's not long) to change over. During the change-over, if the host has moved, the old IP addresses are forwarded by intermediate routers to the new location.

In theory, this means that IPv6 has not just 2^128 addresses, but also a TOTALLY dynamic topology. (Mapping the Internet'll suddenly become a whole lot more interesting! :)

In turn, this means that you can have wireless IP and multiple providers, move from one zone to another, and be guaranteed you'll remain connected.

Further, because addressing follows an enforced heirarchy, router tables will NEVER need more than enough addresses to go one layer up or one layer down. For 99.999% of providers, this will mean an entire 512 entries, tops. Compare this with the millions of entries a typical router handles. Forwarding lag will be carved, sliced, diced and roasted.