[ppml] 2005-1 status
sleibrand at internap.com
Mon Jan 23 19:47:43 EST 2006
On 01/23/06 at 4:25pm -0800, Tony Li <li.tony at comcast.net> wrote:
> > I'd like to know that it is an assignment policy problem exclusively
> > before we ascribe an assignment policy solution.
> I believe that this hits the nail squarely on the head. The real issue
> at hand is that we still do not have consensus on an acceptable routing
> architecture for IPv6. Until we have consensus on a multi-homing
> solution, we do not have a tractable technical solution for anyone to be
> forming address allocation policies around. And it seems highly
> unlikely that better policy is going to compensate for the lack of a
> technical solution.
I see two camps on the multi-homing solution, and I think they might both
be right in some ways.
Some network operators wonder what's wrong with continuing to multi-home
the same way that has worked reasonably well for so many years. They look
at Moore's Law as solving the memory usage problems introduced by routing
table growth. They see a host-based solution as unworkable for large
networks with sometimes complex traffic engineering requirements that
shim6 hasn't even attempted to address yet.
Many others see the need for a new multihoming model for a new and
somewhat different Internet Protocol. They see a need for host and
site multihoming that doesn't require coordination with transit providers,
doesn't require global routing table resources to be used when a site
multihomes, and in general provides more flexibility for end users.
I started out in the first camp, but recently have been reading the shim6
Internet-Drafts and am starting to see the two approaches as
complementary. For sites with hundreds, thousands or more diverse hosts
who know and run BGP, IPv6 PI space seems the way to go. For multihomed
end users, small sites not running BGP but who'd like to multihome, and
sites who'd like their hosts to be able to reroute traffic when they see
performance degradation, shim6 will provide a useful alternative.
I think we can support both camps. Those who already have IPv4 PI space,
who meet the requirements to get it, or who meet similar requirements
adjusted for IPv6 concepts should be able to get IPv6 PI space. Smaller
users who might have a /24 from their ISP announced with BGP to two or
more upstreams can do the same with their IPv6 PA space, and at the same
time roll out shim6 as appropriate to improve failover. Smaller users yet
who currently have no good multihoming options will soon have an option
with IPv6, as long as we don't throw the gates wide open for PI space
assignments and discourage shim6 adoption.
More information about the ARIN-PPML