RE: Doors for IPv6 connectivity [RE: comparaison grids presented in meeting]

["Pascal Thubert (pthubert)" <pthubert@cisco.com>]

Hi Pekka:

I realize you already spent quite some time on this, on both the public
and private sides. I do appreciate the quality and the consciousness of
your work.
 
> 
> On Mon, 8 Mar 2004, Pascal Thubert (pthubert) wrote:
> > MIPv6 (or Nemo for that matter) does the bulk of the work, and Doors
is
> > just opportunistically using the states that MIPv6 installs at the
HA to
> > store the UDP encapsulation parameters. Doors goes through all forms
of
> > NATs, AFAIK. Actually, in the case of a nested Nemo, a single PAT
state
> > in the network covers the full nested cloud.
> >
> > I'm afraid that the draft has expired but it can still be found at
the
> > nemo site:
> >
http://www.mobilenetworks.org/nemo/drafts/draft-thubert-nemo-ipv4-traver
> > sal-01.txt
> 
> Actually, I think this is conceptually similar to
> http://www.watersprings.org/pub/id/draft-soliman-v4v6-mipv4-00.txt --
> just adding some IPv4 glue in the MIPv6 HA implementations to create a
> tunnel.
> 
Right. I pointed that to Hesham at the WG and he very well knows this.
So does Ryuji (Wakikawa) who also works on something similar. Hesham's
approach for MIP6 has 2 aspects, the careof can be v4 and the payload in
the MIP6 tunnel can be v4. While I agree on the latter (for the time
being because maybe one day the HA will not have IPv4 connectivity), I
think (as you seem to do as well) that changing MIP6 to traverse IPv4 is
unnecessary, and doors is an illustration of that. As opposed to using
an IPv4 careof:

- Doors builds an IPv6 careof and leaves MIP6 unchanged
- Doors goes through NATs (which is just a MUST in my mind)

> My kneejerk reaction to that was, why not co-locate a tunnel
> server/broker implementation with the HA, and get around the issue
> without modifying MIPv6 (to include IPv4 specific components).  After
> all, IPv6 connectivity in IPv4 networks (and NAT traversal) is a more
> generic problem than that.
> 
:)

> Actually, it's even better not to co-locate the v4 tunnel endpoint
> with IPv6 HA -- that way, if a tunnel endpoint close by, closer than
> the HA, there are some route optimization benefits.
> 
True, while harder to achieve than the NAT traversal. Maybe something to
address in our Nemo RO taxonomy draft.

> (This is the "mobility thoughts" topic which was discussed on Thursday
> session, and next a problem statement/scenario -like draft is probably
> going to be written.)
> 
> Pascal, is there something obvious I missed when I glanced through
> your idea?
> 
I do not think so but I'd like to point out this: the cool thing with
doors is that the security and state maintenance are inherited from
MIP6; makes it a lightweight draft (and implementation) for quite a huge
value. In other words, if your stack has MIP6 and you have a HA to
register to, then Doors may be the easiest way to go through IPv4 and
NATs. Thus it makes sense to me to include it in the grid, with the
conscious dependency on MIP6 success. 

What do you think?

Pascal