RE: [Fwd: [Saad] Some initiating thoughts...]

Leslie,

> Leslie Daigle wrote:
> I posted this a few days ago -- before, I think, people had
> a chance to get subscribed

IMHO, this text is a good short assessment of the situation. Needs some
more work, but a good base.

> [draft-iab-addressing-2003815.txt]
> Although it is not strictly required, the IPv4 address
> architecture largely assumed a unique binding of a device
> interface to an IP address. Essentially this means a
> device interface existed as a member of a single
> addressing realm.

With IPv6, a large number of issues with SLs and scoping are a direct
consequence of the disappearance of this unique binding. Save for LLs
which are a different kind of animal because they are not routable, the
very assumption of multiple addresses per host has caused us a lot of
trouble.

IMHO the only way we could make any kind of scoping architecture work is
to accept the following restriction:

If scope is to be used, a host can have only two IPv6 addresses per
interface: a link-local and another one, which would be routable but
with possibly a have limited reachability depending on the scope. It
would then be a design element to decide to use either multiaddressing
or scoping.

The idea behind scoping is that it should be a largely-automatic access
control system, likely as a fail-safe for explicit filtering or
firewalling. This means that hosts would not need to be aware of the
scope except to make the distinction between LL and routable.

Note that, for a many enterprise operators, the "restriction" of having
only two addresses per host per interface which is very similar to the
existing IPv4 situation is a feature, not a bug. Regardless of issues
with scoping, multiple routable addresses per interface simply are too
much complication. I am not saying that multiadressing is bad, what I am
saying is that multiaddressing does not work in certain situations.

This list is focused on scoping. I am aware that some identifier/locator
solutions propose that the IPv6 address is used only as a locator, the
identifier being a different type of animal.

In the situations where scoping is desired, the way out IMHO is that the
host's routable address is the identifier (the reason being it would be
a lot simpler to scope than a flexible-shape identifier) and that there
is a shim layer between Transport and Network that does the identifier /
locator mapping.

Michel.

_______________________________________________
Saad mailing list
Saad@ietf.org
https://www1.ietf.org/mailman/listinfo/saad