Re: implications of 6to4 for v6coex

[Brian E Carpenter <brian.e.carpenter@gmail.com>]

On 2008-10-03 12:43, Nathan Ward wrote:
> On 3/10/2008, at 2:26 AM, Erik Kline wrote:
>> On Sun, Sep 21, 2008 at 7:12 PM, Brian E Carpenter
>> <brian.e.carpenter@gmail.com> wrote:
>>> On 2008-09-20 12:54, Nathan Ward wrote:
>>>> On 20/09/2008, at 12:15 PM, james woodyatt wrote:
>>>>> There is clearly something broken about 6to4, and I'm not sure how to
>>>>> fix it.  Can it be saved as a standard?
>>>
>>> Just to be clear, the issue is not with basic 6to4, which was intended
>>> to be implemented only in routers. It's with the "obvious" extension
>>> to allow host-based 6to4, which led fairly quickly to the notion
>>> of anycast relays and the resulting lack of both configuration and of
>>> a discovery protocol.
>>
>> Well it seems to be working well enough (or popular enough, anyway).
>> What if I said that 69.8% of IPv6-enabled users who visit google.com
>> have 6to4?

That's the least I'd expect, as an author of 6to4 ;-)

Any stats for Teredo?

> 
> 
> So, Brian disagrees with the host based 6to4 thing, ie. the way Windows
> Vista and XPSP2 do it.

No, I don't disagree with it, I just want to remind people that most
of the thinking about how 6to4 routes should be announced was based
on the router model, without considering either automatic discovery
or anycast.

> 
> I don't think that the anycast relays has anything to do with the host
> based stuff. It might have been related thought trains in peoples' minds
> in 2001, but the anycast relays are just as relevant to router-based
> 6to4 as they are to host-based 6to4.

Except that the idea was that router relationships for 6to4 would be
configured by mutually consenting operators; whereas the anycast model
implies zero-conf. There's a gap in the middle where a discovery
protocol could live.

> 
> For example, Airport Extreme, Linksys RVS4000, Cisco 8XX+config, etc.
> So, it's not just Windows.
> 
> I'm not so hot on 6to4 stacks in hosts either, but my problem is wider -
> it is more with automatically configured 6to4, which is not specific to
> hosts.

OK, but it's dominated by hosts, especially because of Vista.

    Brian

> 
> My feeling is still that this whole relay stealing thing is a non-issue
> in current implementations, I'll write up in a draft with some
> recommendations to remove that paragraph and bring it in line with most
> existing deployments.
> 
> 
> 
> Also, these stats are interesting, and are roughly the same as what I'd
> expect.
> So, my understanding is that this test is to a hostname with A and AAAA
> records, right? That would make sense, as Teredo will only be used by XP
> SP2 and Linux boxes who have Teredo enabled (Vista prefers A over AAAA
> if it only has Teredo IPv6 connectivity).
> 
> I don't think that you can accurately say that that you tested all IPv6
> enabled users, because you aren't testing whether they are Vista and
> have Teredo enabled. Point em at a hostname with only AAAA and my
> expectation is that you'll significantly see different results - close
> to what I see, which are like:
> - ~90% Teredo
> - ~7-8% 6to4
> - rest Native
> 
> -- 
> Nathan Ward
> 
> 
> 
> 
>