[arin-ppml] The role of NAT in IPv6

[Gams, Matthew D]

There would be no need for upper levels to help out. Only during the transition would there be IPv4 only clients. They would still talk to the globally unique IPv4 addresses. However, once upgraded to IPv-X they would have the ability to use the other level prefixes to communicate. Biggest IPv6 problem is the lack of compatibility with IPv4.

Anyway, since we have IPv6 my main point is that keeping your private network separate from the public is more a matter of minimizing changes to your internal LAN during whatever need comes up.



From: TJ [mailto:trejrco at gmail.com]
Sent: Thursday, April 15, 2010 11:10 AM
To: Gams, Matthew D
Cc: Gary T. Giesen; arin-ppml at arin.net
Subject: Re: [arin-ppml] The role of NAT in IPv6

> -----Original Message-----
> From: Gary T. Giesen [mailto:ggiesen at akn.ca<mailto:ggiesen at akn.ca>]
> Sent: Thursday, April 15, 2010 10:40 AM
> To: Gams, Matthew D
> Cc: 'arin-ppml at arin.net<mailto:arin-ppml at arin.net>'
> Subject: RE: [arin-ppml] The role of NAT in IPv6
>
> On Thu, 2010-04-15 at 11:21 -0400, Gams, Matthew D wrote:
Close is the key word. It is a flat address which is then being logically broken down. A true hierarchical model (closer to the OSI addressing) would have those layers built-in and be able to take them out when not needed. Also, the IPv6 model breaks down as you allow exceptions with bigger organizations getting direct allocations.

I like the level concept in IS-IS and wish IPv6 would have taken the concept and created something a bit different. Level-0, 1, 2, 3, and 4. We could have still ended up with 128-bit addressing if needed (or more) and been much more efficient. Organizations would still use their comfortable 32-bit IPv4 addresses and only the network "gods" would know anything about the larger space available. During transition RFC1918 would be kept intact but eventually DNS would respond with the updated prefixes to allow global routing of the new blocks and in theory the whole 32-bit address space could be used internally.

Routers would only know about the level are configured for with only Level-4 being the biggest back-bone routers that know the whole structure.

Oh well, maybe in another reality...  :)

That model works for edge nodes accessing centrally located/managed services, client-server style.
Scaling it to end-to-end / peer-to-peer operations is either really kludgey at Layer3 or relies increasingly on Layer7'ish helpers.
(In some cases that model breaks even sooner - there are networks where a total of 4.3Billion addresses may not be enough by the time you divie it up in a hierarchical fashion ... requiring yet more layers of duct tape and/or bailing wire.)

IMHO, if it is a network problem it should almost always be solved at the network layer.
Let's get that part fixed first, and let's see where we can make it with that infrastructure ...


/TJ
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <https://lists.arin.net/pipermail/arin-ppml/attachments/20100415/31fa1f16/attachment.htm>