[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

Re: identity/location separation


On Friday, Nov 22, 2002, at 09:47 America/Montreal, Iljitsch van Beijnum wrote:
If we do
locator/identifer seperation, being reachable is still a good idea but
if a set of locators isn't: too bad. We don't flood this information
throughout the network and we also don't have to repair it using
backdoor routes or tunnels. So routing becomes much less dynamic.
Given that there is no specific proposal for locator/identifier
separation, the above it totally bogus. Routing is no less dynamic
than at present. Somehow the concept must be unclear, so I'll try
to explain a bit.

It is better to think about the separation this way:
Currently we overload the "address" with both "identity of end system
(or its interface)" and "location in the topology". The proposal
removes that overloading. The "routing system" continues to propagate
the location information (i.e. topology). However, a host can move
around the topology without being forced to change its identity.
And there clearly will need to be an identity-->location mapping scheme,
for which the existing DNS system is an existence proof that such
name to location mappings are feasible to implement and deploy.

In one sense, Mobile IPv6 is already doing all this. Consider that the
"Home Address" is functionally the identifier and the "Mobile Address"
is functionally the locator of a mobile node in a Mobile IPv6 deployment.

So the proposal to separate the two concepts and remove overloading is
neither new (e.g. Nimrod, Mobile IPv6), nor impossible to implement
(e.g. DNS FQDN-->IPaddr translation, Mobile IPv6's new Destination Option
carrying the identity).

A useful new property of such a schema is that mobility of nodes becomes
a first-order property of network, not a step-child.

Ran