Biggest mistake for IPv6: It's not backwards compatible, developers admit

[Fred Baker]

This is a place where I find myself banging my head against the wall.

As I see it, there are six major proposals on the table.
    - provider-independent addressing (aka PI)
    - provider-dependent addressing (aka PA)
    - provider-dependent addressing with multiple prefix overlays in  
multihomed networks (shim6)
    - private addressing with NAT, converting to provider-dependent  
addressing at a DMZ
    - private addressing with Network Prefix Translation (GSE)
    - exchange-based addressing (aka Metropolitan addressing)

Each of those except straight PA give you ISP independence. Each of  
them has issues; folks don't like shim6 because it transfers the  
complexity to the edge networks, they don't like exchange-based  
addressing because it forces the transit contracts to be inverted to  
sender-pays, IPv6 was designed to solve a lot of the ills caused by  
NAT so re-introducing NAT is a step back into the mire, folks don't  
like GSE because they confuse it with NAT, and so on.

The issue with straight PI addressing is the issue that causes people  
to wonder about the size of the route table. If you have never heard  
the observation that "routing doesn't scale", I'm amazed. The thing  
that makes routing "not scale", and so drives memory volumes and their  
implied costs, both capex and opex, is that PI places a prefix on  
every thing that can be routed to (now on the order of 10^6, within  
the decade on the order of 10^7, per Marshall Eubanks' analysis at  
NANOG) as opposed to the number of entities that require routing to  
(autonomous systems or something like them, O(10^5)).

They say that insanity can be identified when someone applies the same  
algorithm to the same data and expects a different result. If we go PI  
- and yes, the RIRs appear to be headed down that path - I don't want  
to hear any complaints about the size of the route table or the costs  
it implies.

And I have to say that your assertion, that ways to provide ISP  
independence have not been provided, is problematic. Ways have indeed  
been provided, some of them (GSE) by the operator community. All of  
them with the exception of the one that got us into the current mess  
have been rejected out of hand without operational testing and without  
much thought as near as I can tell. That doesn't leave me very  
motivated to come up with yet another.

On Mar 30, 2009, at 9:48 AM, Udo Steinegger wrote:

>
> Am 30.03.2009 um 18:37 schrieb Fred Baker:
>
> Guys,
>
> all the arguments to this discussion are well and fine.
> But for the commercial non-ISP world, at least in old Europe, one of  
> the bigger Problems is to get
> the same level of provider independence in IPv6 (read: PI address  
> space), that they are used to in the IPv4 world.
>
> As long as this is not properly addressed, then people are very  
> reluctant to move towards IPv6 and stick with
> IPv4 until the last day.
>
> I know that this has to be discussed with other entities/bodies and  
> some folks do what they can to solve that issue,
> but in any case, I find that more important to solve rather than  
> discussing if we name the beast IPv6-"transition" or
> anything similar else.
>
> cheers
>
> Udo
>
>