IPv6 PI allocation

[Iljitsch van Beijnum]

On 18-mei-2007, at 15:06, Niels Bakker wrote:

>> Let me know when the I* community is ready for something better.

> Sure thing, Iljitsch.  Have fun on your mountain top watching the  
> ants run around looking for your mystical solution.

Mythical, you mean?

Coming up with a solution isn't the problem. I can point you to one  
or two drafts I've written myself; there are plenty of others.

The problem is:

1. Not making decisions that turn out to be bad ones later

2. Accepting consequences

For instance, shim6 was (is) as struggle. It would have been a lot  
easier if the IPv6 socket API hadn't required IPv6 applications to  
know about IPv6 addresses the same way IPv4 applications must know  
about IPv4 addresses. With IPv4, this was unavoidable, because the  
socket API came first and the DNS came later. With IPv6, this wasn't  
the case so rather than (or, at least, in addition to) the same  
address-based API that was used with IPv4 there should have been a  
connectbyname() call or something like that.

I'll spare you a diatribe about which consequences must be accepted  
exactly, but one thing is for sure: in order to make progress, it's  
often necessary to get rid of existing features that get in the way  
of future developments. That can be unpleasant for the people who  
rely on these features, but that's the cost of progress.

I'm not talking about giving up multihoming, BTW. Multihoming is  
important and worth some pain to be able to support in a scalable way.