Keith Moore wrote:

> Some apps care about having consistent view of addressing across all
> locations in the network.

The trouble is that while filters exist this will NEVER be true, in the
general case.  Having multiple addresses per host confuses the issue as
well.

Similarly, the arguments about rescinding 'scope' are bogus.  Every
address is functionally scoped.  In an extreme model, 'scope' can even
vary dynamically as (say) links fail and recover.  What we really need to
consider is:

* Are there sensible ways of communicating 'scope' information of addresses
to applications? (I expect the answer is 'not easily' for the general case).

* Is there benefit in having classes of addresses with inherent scope
limitations (eg link-local, local)?

Note that any application CAN discover whether it is within the functional
scope of any given address - it tries to use it.  If not done cleverly, this
can cause significant delays and/or traffic, but the option is certainly
there.


In closing, three guidelines / work items:

* Applications should not assume that all addresses are equal.

* Applications should not assume that addresses can be forwarded at-will,
unless provided with additional (configured?) information that would support
this assumption.

* Applications that do not care to do their own address management need a
higher level abstraction to the current sockets and getaddrinfo APIs.

-- 
Andrew White
--------------------------------------------------------------------
IETF IPng Working Group Mailing List
IPng Home Page:                      http://playground.sun.com/ipng
FTP archive:                      ftp://playground.sun.com/pub/ipng
Direct all administrative requests to [EMAIL PROTECTED]
--------------------------------------------------------------------

Reply via email to