Little Interest In Next-Gen Internet

Ant wrote in to mention a Computerworld article that is reporting on the slow acceptance of the IPv6 version of the internet. From the article: "Information Technology (IT) decision-makers, in U.S. businesses and government agencies, want better Internet security and easier network management. However, few see the next-generation Internet Protocol called IPv6 as helping them achieve their goals, according to a survey released Tuesday by Juniper Networks Inc."

Comments

Re:Give me an easy upgrade path

[MartinG]

What are you talking about?

I run ipv6 and ipv4 together on the same net all the time.

For ipv4 I have one static internet address on my router machine which provides NAT for all the other internal machines.

For ipv6, I have a /48 allocated to me so each machine has its own real ipv6 internet address and can talk directly to all other internet ipv6 addresses. (plus, I have over 281 trillion ip addresses spare for later)

Also the applications don't need writing for one or the other. By using the ipv6 API you automatically get support for ipv4 at the same time.

There is an easy upgrade path. What will really get folks upgrading is when more and more sites become ipv6 only. For example I am setting up a nature webcam site which will be ipv6 only for exactly this reason.

Re:Duh

[quantum bit]

Not picking on you in particular cHiphead, just happened to be a convenient place to post.

IPv6 = everybody gets their own ip address, walk up to a computer, swipe your worldID card, it switches to your personal ip.

Except it doesn't work like that. IPv6 uses a hierarchical routing model, much stricter even than IPv4 classful routing.

The IPv6 address you get assigned (each customer is supposed to get their own /48 subnet, we'll see) comes out of your ISPs pool, which is probably a /32 or something. The really big carriers (i.e. backbone) get /24s, and they allocate smaller subnets to smaller ISPs. The big carriers get their subnets assigned out of a pool, but of the remaining 24 bits, the first 3 specify the address type and the last 8 are reserved. That leaves 13 bits, so there can be a maximum of 8192 "big" carriers (called TLAs).

Now, unlike the current internet, ONLY TLA's exchange routing information with each other. Every single address withing a TLA's block MUST be routable from one of its peering points. Routing between the TLA blocks may only happen at those top-level points. Small netblocks are no longer portable, so when you change ISPs, you get new addresses. No exceptions -- doesn't matter how many you have. That also means if you want to have a redundant connection for your server (multiple ISPs), it has to have multiple IP addresses too. No more BGP tricks.

So you can't assign an IPv6 address to a person, as every terminal they use has to have a different address by definition of IPv6.

The other common misconception is that IPv6 has more addresses (2^128) than particules in the known universe. This isn't really true as the lower 64 bits are not routable. They're usually automatically derived from the 48-bit MAC address, but can be statically assigned if so desired. Even if you did statically assign them, all (2^64)-2 of them would have to be on the same (flat) subnet, which would be one huge honkin LAN.

So that really only leaves 2^64 routable networks, each of which MAY have a lot of machines but in practice probably won't have more than 100-200 max, and probably averaging much lower than that. If you take into account that the specification calls for each customer to be given their own 48-bit subnet (giving them 16 bits worth to route internally if they so desire), there isn't just a whole lot more room then IPv4 because so much is forced to sit unused. It is considerably more to work with yes, but not astronomically like many people seem to think.

Ok, sorry for the rant, but just trying to make sure reasonably accurate information gets posted somewhere :)