[e2e] node addresses vs. interface addresses

[Christian Huitema]

> % In fact, if the multi-homed site is small enough, it is possible to
> % provide each host/interface with as many addresses as the site has
> % providers, and then use the host multi-homing technology to sort out
> % which address is used when. This is clearly a trade-off: 
> lower impact on
> % the global routing fabric versus more complex site 
> management. I would
> % expect small sites such as home networks to lean towards "host
> % multihoming", while large sites such as large corporations 
> will simply
> % pay their way into global routing.
> % 
> % -- Christian Huitema
> 
> 
> 	This seems counter-intuitive. The normal arguments I've heard
> 	indicate that small sites have less "clue" and so can't/won't
> 	deal with "complex site management" while large sites can
> 	afford and will spend to have a rich set of options.

The management action, at least for IPv6, consists solely of having
routers advertizing multiple network prefixes from which nodes can
automatically configure servers. In a small network, specially in a
single subnet site, this can be completely automatized -- no manual
intervention required. In a large network, you need to propagate the
announcements to all the routers on the site, and get some
synchronization between these routers, e.g. to simultaneously mark as
deprecated the prefixes that happen to be out-of-service. 


> 	The normal example is DNS.  Small sites are generally considered
> 	to not have the time/patience/expertise to administer 
> DNS locally.

Not a lot to do with DNS. 

> 	If v6 multi-homing technologies are robust enough for a small
> 	site, why would a large corporation "pay their way" when
> 	the host-based multihoming solutions work so well?

Tradeoff between managing multiple prefixes in routers and pressuring a
provider. It is all depending on contract negotiation, business
relations, clout, etc. Small sites don't stand a chance. 

-- Christian Huitema