RE: PReserving ids when changing providers (was:RE: to be draft-ohta-multi6-8plus8-00.txt)

["Christian Huitema" <huitema@windows.microsoft.com>]

> > Probably locators will be hardcoded in less places than today, so
they
> will
> > be simpler to renumber and IDs will be hardcoded in some of the
places
> that
> > ip are used today and perhaps in many others, since they will be
used
> for
> > recognizing endpoints in apps and filters, acls and so on
> 
> If so,it's good. But, for example, SIP hardcodes locators instead of
DNS
> names. As per discussions in ietf-aulli, apps folks attributes it to
lack
> of access to naming service and reduced delay (in terms of lookup
> latency.)
> So, in future, we can expect more such locator hardcoding at ULP.

It is not exactly "hardcoding". SIP payloads carry the IP addresses to
be used for initiating media sessions, but the protocol also has
provisions for changing these addresses if the SIP agent's location
happens to change. It is much easier to move a session to a new location
by telling the peer about the new location than by updating a name
service and hoping that the peer will find out. It is also much more in
line with end-to-end argument.

In any case, there is more than performance at stake. We must also to
consider privacy. A SIP agent will only reveal its current location to
the peers with whom it agrees to establish a session. This is very
different from publishing a mapping between identifier and location in a
globally accessible name service.

Not to say that SIP has got everything right. The payload format only
allows for one IP address per media stream, which is problematic when
the host is multi-homed or multi-addressed. 

-- Christian Huitema