Keith;
GET A CLUE.
I did read the draft.
Huh? You obviously don't.
What's more, I have tried implementing multihoming
in a way very similar to what you are describing, which is what informed
my earlier comments.
I know "very similar" often means "completely different", especially
when you don't read the draft carefully.
Without routing support from the network, it cannot
be made to work well.
Wrong.
With the end to end principle, it is silly to distinguish routers and
nodes and give routers more intelligence.
With end to end multihoming, there is no reason that the global routing
table is large.
So, you can assume a host has full routing table.
However, unless you use host route, exsitence of a route entry means
that there is reachability to some part of an address range of the
entry but not necessarily to the target host.
Thus, routing systems may give you a hint but routing information
can not be fully relied upon.
All the possible addresses should be tried with certain time out.
Applications needing quicker recovery should use smaller time out,
of course.
They are all written in the draft.
READ THE DRAFT.
Especially if you're also trying to support mobility.
No.
My draft says nothing about mobility, because it is no difficult.
Masataka Ohta