Brian E Carpenter allegedly wrote on 03 19 2009 1:48 PM:
I recently had this exchange with Dan Wing on the BEHAVE list:
... it seems to me
that we might consider defining a generic 'referral object', containing
more information than just an address, that could be passed among
application entities. It could contain TLVs that would provide the
semantics of the referred address as well as the raw address bits.
Does this seem worth exploring?
Yes, this could form the basis for very useful guidance to application
designers. ICE does something like this already, but abstracting its
functions up several levels, as you propose, would be useful.
Referrals are hard to figure out. I think the tradeoffs are worth
serious study.
First party referrals like this are interesting but not so hard as third
party referrals (Tom tells Dick how to reach Harry). They assume that
Tom knows Dick's and Harry's shared scope. Can we assume we have some
kind of globally unique identifiers that are globally unique? So far
that has been extremely hard to pull off. I have a suspicion that for
referrals you want to go all the way back to domain names and URIs.
Also, the referral problem is multilayer (you want to be able to refer
to processes in addition to machines) but only at the layer that needs
that information. The question, which is something like the e2e
argument, is whether to bother providing details in network-layer
referrals if they will be trumped by higher-layer needs.
But in any case we should do a major investigation of referrals in the
light of location/identification separation and the fracturing of the
Internet into various scopes.
Scott
_______________________________________________
Ietf mailing list
Ietf(_at_)ietf(_dot_)org
https://www.ietf.org/mailman/listinfo/ietf