making multihoming transparent to the application within the IP level
is a hard problem, but is often _besides_ the point - failover and
fault tolerant applications WANT to know in a timely and explicut
manner when a link or interface fails - they do not want transparency
- some networks preople think that the applicatio nbuilders do (and
some naive applioaction writers do) but in any sophosticated realworld
(tm) fault tolerant application, transparency is not what is
wanted...so in that sense, the use of smarts in the appkcation layer
are exactly whats wanted and the problem of trying to do multihomeing
(and multi-provider multihoming) while maintaining address (and other
state) aggregation is a snare, edlusion and chimera...
seems like an exaggeration. FT apps do want to be explicitly informed
about network failures so they can adapt, and would like the ability
to make their own decisions about which links to use (given multiple
choices). but most apps are not FT, and to those that aren't FT,
address stability under changing network conditions is a boon.