Re: "architectural change" [RE: Draft: PI addressing derived from ASnumbers]

["Mario Goebbels" <mario.goebbels@skynet.be>]

I have some general ideas about how someone could able routing -from/to PI
networks over the regular IPv6 network, using some additional logical data
structure (I tagged it Locator Cache in my idea), some extension header and
requires exchange of PI network information over some routing protocol to
fill that Locator Cache on routers. Clients are also supposed to make use of
that locator cache too in my idea. Due the fact I'm no expert the idea might
be pretty general, but maybe some aspects in it are worth a dime. If you
don't mind, I could write it down in some human readable form and post it
onto this mailing list.

-mg

> OK, maybe we need IPv6++ (although as Christian's message points
> out, the changes mainly wouldn't be in IPv6 itself).
>
> But that is R&D, and it will be some years before it ships in product.
> Over here, can we get on with something shorter term? What is an I6SP
> supposed to say to a customer who asks for IPv6 multihoming next week?
>
>     Brian
>
> Pekka Savola wrote:
> >
> > On Sat, 1 Feb 2003, Randy Bush wrote:
> > > > In the long term, i'm rather convinced that an architectural change
will
> > > > be required.  I can't see other scalable alternatives.
> > >
> > > this is also what i see, though i am anxiously awaiting enlightenment.
> >
> > Perhaps it was not clear what I meant with an "architectural change",
and
> > I should try to clarify.
> >
> > I didn't mean we need IPv7; rather, that the model of solving a problem
of
> > locator/identifier problem and dumb end-hosts with a global routing
> > solution.
> >
> > For example, I consider models where multihomed hosts make a much more
> > active role in the current "multihoming routing" problem an
architectural
> > change, as well as separating identifiers and locators as in LIN6 (or
even
> > more so, HIP), etc. as architectural changes.
> >
> > Certainly, there are ways to do multihoming with IPv6 today.  And there
> > will be ways.  But one point many, myself included, have tried to make
is
> > that doing it the way most have done it with IPv4 -- that is using BGP
> > with "PI addresses" -- is not scalable.  That is likely the only
solution
> > for a long time to satisfy all those requirements.
> >
> > For example, I personally have great faith in "multiple addresses from
> > multiple providers" -approach (though there are some small ways it will
> > have to be improved): it should give a sufficient amount of multihoming
to
> > most sites.  Also, multiple connections to a single ISP are also a way
to
> > get redundancy.
> >
> > The major things I see missing are:
> >  - ways to get upstream _independence_
> >  - scalable ways to protect against your upstream ISP failures (if
> > multiple addresses per node approach is not sufficient)
> >
> > .. but I'm not sure whether solutions for these are really even
required:
> > people just want them because they've always had them.