Re: to be draft-ohta-multi6-8plus8-00.txt

[Masataka Ohta <mohta@necom830.hpcl.titech.ac.jp>]

Marcelo;

> If i understand correctly, you basically propose to:
>
> - Perform destiantion locator selection by injecting full global routing
> table into the hosts so that they have the required info to know which
> destiantions are reachable
> - Use OSPFv6 options to let the hosts to detect which cource address to use
> and so overcome the ingress filtering problems
> - Add an option to tcp so it can hanlde multiple addresses in a single
> connection
> - Use transport layer information to trigger a rehoming event (path failure
> detection)

Mostly so, though I think there can be RIPv6M6 and SCTPM6.

> My question is what do you need the 8+8 format to do this?

8+8 format is to let 8+8 transport layers disntinguish 8+8 addresses
and leagacy addresses, to let transport layers mostly ignore locators,
to have location independent IDs and to remove some IP layer tunneling
by simply rewriting locator part.

> I mean, wouldn't be possible to exactly the same using regular addresses and
> selecting one of them as identifiers to the transport and upper layers?

How can transport layer protocols detect that an end or its peer
is capable of M6?

How can you make IDs survive ISP changes?

How can you be sure to rewrite addresses at the IP layer, if
you are not sure that all the transport layer protocols can
still identify an end?

> In other words, as your architectural draft explains, your solutions is
> completelly end to end, so the ends can have the information about what
> locators are linked to an identifier and which is the identifier to be used
> for each connection, so why is required to reflect this loc/id split in the
> address itself?

There are elegant and not so elegant ways to implement an
architecture.

Have I answered your question?

							Masataka Ohta