Re: Architectural approaches to multi6

[RJ Atkinson <rja@extremenetworks.com>]

On Tuesday, Mar 25, 2003, at 10:07 America/Montreal, marcelo bagnulo 
wrote:
> I am wondering if these two approaches are the same or not:
>
> - The first case is where there is clear separation between locators 
> and identifiers. In this case, a binding between an identifier
> and one more locators is needed.
A binding is *always* needed between any identifier and its set of
locators, even when the identifier happens to be the permanent
home address (using MIPv6 as an example) and the other addresses
are the locators.  The *nature* of that binding changes depending
on implementation details, but the need for some binding does
not change.

> - The other case is where multiple addresses are used for reaching the
> other end. I am not sure that you need an identifier, all you need to
> know is that there is a set of possible locators for reaching the 
> other end of this communication. (Perhaps the identifier is the
> set of locators?)
If one considers MIPv6, there is only one IPv6 address that is
used for the TCP session state.  The address used in that way
is functionally an identifier.  The other addresses (i.e.
those not in the TCP session state) are functionally locators.

> Do you think that there is a difference between this two approaches?
At a high-level, no.  In gory implementation details,
of course yes.

As noted before, my goal (and I suspect tli's goal) pro tempore
is to discuss the high-level aspects only.  If/when/after some
consensus emerges at that level, it would then make sense
to dig one level deeper.  IMHO, it would not make sense
to dig deeply prematurely -- for fear that we'd be unable
to see the forest because of focus on individual trees.

Cheers,

Ran