[ih] Multi-homing naming Was: Re: History question -- multi-homing

[Karl Auerbach]

On 9/12/26 2:06 AM, vinton cerf via Internet-history wrote:
> It would be an interesting exercise to posit a different identification
> scheme that would scale and also permit multi-homing. Has someone already
> worked that out? Would it require a flat ID space?

Back in the 1970's my (now gone) friend (and Best Man at my wedding) 
Frank Heinrich (student of Dave Farber and who worked on DCS) taught me 
that we ought not to think of application-to-application associations as 
having invariant locations for the participant entities.  The idea was 
that on a network, application level stuff can move - that the binding 
of applications to network addresses ought not to be considered as 
immutable.

Others seem to have had similar thoughts - I, however, being slow on the 
uptake, never really understood the reasons why ISO/OSI defined "titles" 
for things.

It wasn't until the web came along, and more recently use of web 
protocols as vehicles for service APIs, i.e. "cloud computing", that I 
began to realize that we now live on a network in which Frank's wisdom 
has come true.

In today's network world, at the application layer, we have entities 
that split/partition, move, join, vanish, and appear.

That dynamic has created the need for not only what ISO/OSI called 
"titles" but also something more.  We need more because after a 
client-application association is created, the client or the application 
could split/partition.  Yet because of things like state and caching, we 
want to have handles that let us preserve that association relationship 
without abandoning all of the context (or relying on troublesome things 
like "cookies".)

To answer Vint's question: I think that our experience with DNS has 
informed us that caching is not only important, it is necessary.  And 
that structure in identifiers is needed in order to do effective caching 
of what are very large (and growing), dynamic, name spaces and mappings.

Personally, I am dubious whether DNS is dynamic enough for this. (And I 
certainly want to avoid the wars, particularly trademark wars, that have 
burdened DNS administration - so my preferred format for elements of 
these new kinds of names are tokens that do not carry human semantics.)

(I have come to also believe that on today's net we often need to find 
instances of resources by attributes rather than names.  This suggests, 
to me, that we need something parallel to DNS that does lookups based on 
some sort of flexible and extensible set of attributes - sort of like 
the core of a search engine, but better packaged as a network service 
for software to use, and more efficient and dynamic.)

I wrote about this way back in 2010:

On Entity Associations In A Cloud Network

https://www.cavebear.com/archive/public/cloud-entities.pdf

         --karl--