> From: Steve Deering <deering(_at_)cisco(_dot_)com>
This is kind of a ways from my original point, which was simply griping about
this continued irritating claim that "there's no fully worked out example of
separating location and identity", but what the heck...
> Way back in June of 1992 on the big-internet mailing list, I pointed
> out ... that the reuse of IP addresses for identifiers in that case was
> an important part of making it work.
It's used to perform the rendezvous (and error recovering in existing
connections after movement), yes. However:
- i) there are other ways to do the rendezvous/error-recovery (depending
on which namespaces you have, and what bindings between them exist)
- ii) tying the rendezvous to the original topological location of the host
does have potential failure modes (e.g. if that part of the network goes
unreachable, the mobile host may become unreachable even if it's still online
elsewhere);
- iii) using the location namespace as identifiers places constraints on that
namespace (especially when designs place that name in every packet), which I
think is a really big problem.
There may be more, that's off the top of my head.
> You, on the other hand, insisted that the identifier be taken from a
> different, flat space, which would then *not* be amenable to using the
> same binding mechanism as mobile IP
I gather you're speaking of this message:
>> Date: Fri, 26 Jun 92 15:37:14 -0400
>> From: jnc(_at_)ginger(_dot_)lcs(_dot_)mit(_dot_)edu (Noel Chiappa)
>> Message-Id:
<9206261937(_dot_)AA08754(_at_)ginger(_dot_)lcs(_dot_)mit(_dot_)edu>
>> Subject: Re: Mobile Hosts and IDs
Although that message nowhere mentions the use of a "flat" (i.e. non-
directoried) space of identifiers, it does speak of using a different,
non-location namespace for identification.
Also, at that point I was thinking of a "shortish, fixed-length" space for
identifiers, since I was thinking that they would be placed in every packet
(e.g. for robustness, as well as multiplexing) and for various efficiency
reasons I felt that those two constraints would be needed to produce good
efficiency (surely a goal dear to you :-) for such a use.
(As a result of discussions in the NSRG, I'm now thinking about other
namespaces for identifications, ones which don't have those constraints, and
result in more complex mechanisms, but that original outline is still viable.)
> the one mechanism that .. looked like it would actually work and scale
> up
Actually, I've heard that some people (not academics, BTW, it was a large
commercial entity) are unhappy with IPv6 mobility because they think it won't
scale. I'm not sure of the exact scaling problem they are concerned about (I
got this second-hand, and it was a while ago), so alas I can't give you more
details.
Noel