Re: stable addressing

[Iljitsch van Beijnum <iljitsch@muada.com>]

On 20-apr-04, at 13:11, marcelo bagnulo wrote:

> If the IPv6 multihoming solution doesn't provide provider 
> independence, it
> will be less attractive.

Exactly. And if multihoming in IPv6 is significantly less attractive 
people may opt to stay with IPv4.

> there are different renumbering scenarios here:
> - Renumbering in current IPv4 networks, The experience here says that 
> it is
> expensive and sites will try to avoid it.

Going through some renumbering right now... It's never any fun.

> - Renumbering in current IPv6 networks, considering stateless
> autoconfigurations, a fixed /48 etc. Perhaps it is still expensive but 
> it
> will be cheaper. Perhaps it is still too expensive.

Very much depends on the way things are deployed. When using stateless 
address configuration the actual renumbering is MUCH easier but editing 
all those text files where addresses show up isn't any better than with 
IPv4.

Note that renumbering gets worse as more systems are involved. In IPv4, 
it is common to start out with few addresses, then renumber once or 
twice and finally end up with a (somewhat) portable address block. 
Also, NAT limits the number of systems that receive a new address. With 
IPv6, it's entirely possible that netwoorks that are already very big 
need to renumber. This could be fairly traumatic.


> - Renumbering when there is a loc/id split solution available. I guess 
> we
> don't have much idea how this would be, but we guess that it will be 
> even
> cheaper.

This will depend on whether people will still find in necessary to 
filter on (locator) IP addresses. If people find it hard to leave NAT 
behind I'm afraid that address filtering won't disappear overnight 
either.

> As i understand this thread, perhaps it would be possible to design the
> solution to provide some sort of stable internal addressing for 
> multihomed
> sites, sort of GSE or MHAP like.

Yes.

> About the particular approach that you are considering below, imho a 
> very
> valuable feature of a solution would be that middle boxes are 
> stateless, so
> that packets belonging of a certain communication can flow through 
> different
> middle boxes. I don't really think that a solution with a per flow 
> state is
> a good approach.

Hm, we talked about the security issues with rehoming all sessions 
towards an IP address, our conclusion being that this is too dangerous 
without relatively strong authentication. However, this doesn't 
necessarily imply that the alternative must be working per session. 
That also applies here. I think it would be entirely possible to group 
sessions as per a creation/refresh date.

Also, there is still the option of having cryptographic hashes in the 
addresses. This would especially make sense with stable addresses. If 
we can find enough bits the hash can be in the higher part of the 
address so there is no need to make hosts aware of the existence of the 
hash.