Re: Missing DNS reqt? [was Re: (multi6) requirements draftcomments]

[David Conrad <david.conrad@nominum.com>]

Hi,

On 1/2/02 6:02 PM, "Alex Zinin" <azinin@nexsi.com> wrote:
> If you make EID (your PI address) "addressable" and expect it to be
> usable by the endpoint to tell its routing names, then EID needs to
> be routable

Not necessarily.  If you have some mechanism that maps between the EID and
the locator, you can route on the locator independently of the routability
of the EID.  Numerous mechanisms can be imagined which could provide such a
mapping service (DNS among them).
 
> Regarding the question of whether EID->routing name mapping should
> be done through DNS together with DN->EID mapping or through a
> separate packet exchange. I think it is important to keep in mind
> node mobility, where the node's routing name changes dynamically.
> The relatively static nature of [at least] today DNS does not
> converge well with possible dynamics of the routing name.

My feeling is that any mobility solution that relies on the variability of
an EID is doomed given current name to address translation technology.  Not
only do you have to worry about stuff like DNS TTLs, but you must deal with
the fact that applications themselves cache addresses (that is, application
data structures know what an address is).

If, however, the EID is separated from the locator, you can vary the locator
according to topology changes (be they provider driven or the fact you've
roamed between one cell tower to another) without changing the bucket o'
bits applications have memcpy'd from the struct hostent returned by
gethostbyname().  In this model, DNS TTLs become relevant (assuming the
locators are looked up via DNS), but that is relatively easy to deal with.

Rgds,
-drc