[ppml] Policy Proposal: IPv6 Assignment Guidelines

[Owen DeLong]

On Aug 23, 2007, at 6:06 AM, <michael.dillon at bt.com> wrote:

>
>> I think your both naieve to think that your typical corporate
>> customer with a couple hundred nodes and no redundancy to the
>> Internet is going to put up with being told he has to
>> renumber his entire internal network when he decides his
>> current ISP is a chuckhead and decides to go to a competitor.
>
> Perhaps you haven't read RFC 4192? It is not necessary to renumber the
> entire internal network to change ISPs. A lot of the number changing
> happens automatically, because both ISPs assign a /48 prefix. All
> interface addresses are unchanged, only the prefix changes. And  
> this is
> accomplished in a two step process where you begin by adding a second
> prefix, which results in all interfaces having at least 2 IPv6
> addresses. The process is not entirely painless, but it is far less  
> work
> than with IPv4.
>
While the technical description is accurate, the consequences and
the relative pain computation are myth.

In any reasonably run enterprise, the majority of these addresses
are handed out via DHCP, so, renumbering end hosts for the
most part is irrelevant.  The ones that are tend to be servers
and can accommodate a second IPv4 address in most cases
and the same scenario can apply for the most part.

The real pain in renumbering, both IPv4 and IPv6 is the number
of configuration files where your addresses appear that you do
not control.  This includes vendor, client, and partner VPN setups,
holes punched in vendor, client, and partner firewalls, etc.

This pain is identical in IPv4 and IPv6.

Owen