Re: noid and applications (generic requirements from applications)

[Brian E Carpenter <brc@zurich.ibm.com>]

Juan Rodriguez Hervella wrote:
> 
> On Thursday 13 November 2003 14:36, Kurt Erik Lindqvist wrote:
> > On torsdag, nov 13, 2003, at 04:11 Europe/Stockholm, Iljitsch van
> >
> > Beijnum wrote:
> > > On 12-nov-03, at 20:45, Erik Nordmark wrote:
> > >> The only thing to consider is when AID(A) isn't working as a locator
> > >> when
> > >> C tries to contact it.
> > >>
> > >> There are two choices:
> > >> C can detect that the locator doesn't work and do a reverse+forward
> > >> DNS
> > >> lookup to get the whole locator set (Ls(A)).
> > >> or
> > >> The referal can include Ls(A) instead of just the AID(A); this
> > >> requires
> > >> modifying existing IPv6 appication protocols which do referal.
> > >
> > > I think it would be a good idea to look at a referral API. Such an API
> > > would allow a host to ask the system to create a neat little package
> > > containing all the pertinent locator and identifier information, then
> > > it can transmit this package to a correspondent and the correspondent
> > > can then unwrap it and use the content to do whatever it needed the
> > > referral for. The advantage of such a system is that it could allow
> > > applications that need referral to be completely independent of lower
> > > layer technologies, such as the multihoming solution we're going to
> > > implement, and the differences between IPv4 and IPv6. The downside
> > > would be that this referral package could get significantly larger
> > > than a simple address.
> >
> > Although I am tempted to say that a multi6 API with ULP hints, referral
> > package, etc is a good idea - I am not sure I think that is where we
> > should be right now. This will mean that applications will have to be
> > rewritten in order to make full use of the mulihoming solution. Maybe
> > this is another topic for Elliot's draft?
> >
> > - kurtis -
> 
> The Ilijtshc's comment raise the question whether multihoming should be
> solved at the IP layer. I'm sure that applications can find out a (better)
> multihoming solution.

There are in fact many applications that succesfully survive connectivity
changes, but this WG is chartered to work on *site* multihoming for IPv6
and that seems to imply layer 3 or layer 4 solutions.

    Brian