[ppml] 2005-1:Multi-national Business Enablement

[Paul Vixie]

i've been watching this debate, both on the current thread and in years past,
and while i watch, i've tried to see if there's a grand unifying principle.

multihoming is the right technical basis for needing one's own address space,
and i do think that a technical basis is better than any other kind of basis.

but what kind of multihoming?  buying two DS3's from two transit providers is
technically "multihoming" but it's not sufficient cause to allocate a slot in
the global routing table-- no matter what size the address block might be.

i think i'm ready to make a proposal.  let's make it be all about peering,
and leave the only-buys-diverse-transit folks to the mercy of their ISP's.

we can argue about how many peers you have to have, in how many exchange
points, and whether buying transit reduces or increases the number of peers
you have to have.  we can argue about how often you have to re-show proof
that you're still peering, and how long is your grace period if you fall out
of compliance.

i feel VERY sure that bechtel, ford, daimler-chrysler, and the other large
companies with large networks, are all capable of renting rack space at PAIX
or Equinix or whatever, buying some routers, renting some BGP engineers, and
meeting this requirement.  especially if they know they'll get a /32 out of
it.  occasionally you'll run into smaller companies who can also do this, but
i think it's unlikely to cause a "gold rush" toward the peering centers.