Christian speaks
Why should be deploy IPv6?
IPv6 is the same service with more addressess. There is a better way to do options and a flow label, but basically IPv6 is the same as IPv4.
Bellcore is projecting 100 million by 2000.
Most of the Fortune 100 companies did not get addresses when they were plentiful. So, they have to do NAT and use RFC 1918 addresses.
NAT is not a full replacement for having real address space.
Right now, the ratio of modems to users is 1 to 5 or 10. When ADSL and cable modems happen, the ratio is 1 to 1 or > 1 to 1.
ISPs want to multihome. Users want to multihome. Some of the reasons are good and others are not. CIDR does not generally work in this kind of world. Ipv4 does not have an answer here.
IPv6 can provide all this. The addressing hierarchy follows the basic architecture of the Internet. There is a lot of flexibility is how the address hierarchy can be configured.
To deal with multihoming, the end systems would have multiple addresses: one for the local link, one for the site and one for each provider to which the site is connected.
Router renumbering
Address Configuration
DHCP can also be used.
DNS has a new record called an AAAA record.
This record has a level of indirection that will permit the first X bits of the address to be replaced with the prefix of the provider.
There are also the issues of dealing with access-lists and such. These need to be addressed by the router vendors as they do IPv6.
The transition -- parallel stacks in the end systems or use an IPv6 NAT to Ipv4.
The use of which stack can be done by activated by neighbor discovery.
The current backbone is done with tunnels in 6BONE.
The big winners in using IPv6 will be large users and new ISPs.
Yakov speaks
IPv4 is ready now.
There is a much larger address space in IPv6, but increasing the address space isn't enough to support the growth. The addresses are globally unique, but are they globally routable?
The routing strategy in IPv6 is just like CIDR. Anyone who says anything else is confused.
The internet routing table is growing linearly over the last four years. CIDR does work.
Just under 21% of the total Ipv4 space is used in the globally routing system. This is a compliment to the good work by the IP allocation registries.
Renumbering is a fact, but it is totally unrealistic to expect people to do this voluntarily.
Are addresses "owned" or "lended" in IPv6?
Host autoconfiguration is a good thing. However, renumbering and enterprise is more that just autoconfiguation.
IP security is required IPv6 and optional in Ipv4. You can get it in Ipv4 today.
What about QoS?
There is flow lables and priorities in IPv4.
If you need this, you can get it in IPv4.
What about Multicast?
There is a problem with the scalability of multicast routing. IPv6 does not make this better. In fact, it make make it worse (more addresses).
What about NAT?
Can be used in both Ipv4 and IPv6 for basically the same purposes.
Yakov notes that the claim for the lack scalable Multihoming in IPv4 is wrong.
There is a problem with Non-distruptive Multihoming in both protocols. The NAT can help do this (re talk yesterday).
A NAT is not a perfect solution:
Are NATs and IPv4 a better solution than IPv6?
IPv6 nees to fill its incomplete parts.
IPv6 needs to work to fully uncover is potentials.
CIDR and NATs are here to say. ALG (Application Layer Gateways) are also here to stay.
Both Christian and Yakov agree that "The Internet will become more and more like the telephone network."
Christian says that the TLAs will be allocated by the address registries. Yakov notes that the proposed allocations hierarchies are unreasonable.
Guy from APNIC is now seeing linear address growth. They are projecting that their five class A's are expected to last until 2005. They are also seeing a large number of NATs at their customers, because that's what the customers want. He does believe that the IPv6 allocation mechanism will get rid of some of the historical artifacts in the IPv4 environment.
Yakov says that he is aware of the historical issues but the current routing system does work.
Vadim asserts that having multiple addresses in the host is problematical without some knowledge of the topology. Christian says that the host will know the address of the next hop for the prefix.
Editor's Note: There was a bit more here, but it was difficult to understand some of the conversation because of my inability to parse the accents.
Christian notes that deploying that IPv6 is important to making it work better.
Christian also notes that Yakov is supporting lots of little hacks to accommodate changes. He sees NATs and ALGs as being hacks. Yakov argues that it's important to do a cost-benefit analysis.
Someone askes Yakov when Cisco will return their address space one they have moved their network behind a NAT. Yakov says that if someone has a specific offer, they should make it.
What is the killer app for IPv6? Christian believes that IP telephony has the potential to do that. Host autoconfiguration is another important feature that large enterprises could make good use of.