[ppml] IPv4 "Up For Grabs" proposal

[William Herrin]

On 7/11/07, Kevin Kargel <kkargel at polartel.com> wrote:
> Why is there such a big push to drop IPv4?  Is there a reason that v4
> and v6 can't operate concurrently in perpetuity?  Won't the customers go
> where the content is and the content go where the money is?

Kevin,

Others have offered excellent and concise answers to your three
questions. I'll attempt a longer one that hopefully clarifies more
than it muddies.

IPv4 causes a lot of grief for the operators of the "default-free
zone" or DFZ. The DFZ is the part of the Internet which has
authoritative knowledge of the direction in which to route any packet
legitimately on the Internet. It has no "default" route, no path to
"everything else."

Right now there are about 220,000 routes in the IPv4 DFZ. This puts a
good deal of strain on the system.

For one thing, every subprocessor on every router in the DFZ has to
have enough memory and horsepower to manage 220,000 routes.

For another, every time one link in the DFZ comes up or goes down,
routers potentially across the entire DFZ have to rearrange all
220,000 routes so that they follow the new best paths. While this
process completes there can be routing loops and dead zones where the
Internet is just plain broken. The more routes there are, the longer
it takes to complete.

As the crunch for IPv4 addresses starts to tighten, its likely that
large service providers will receive more small allocations instead of
fewer large ones. This exacerbates the problem: each allocation
consumes yet another route in the DFZ.

To address this, DFZ providers spend vast sums of money on routing
hardware and high-reliability core network links that rarely go down
yet they are still only able to do an adequate job of keeping the
Internet stable.

Because of the change in how IP addresses are justified and assigned,
the IPv6 DFZ has only a couple thousand routes and is expected to have
fewer than 100,000 routes at full deployment. This will make it
possible for folks on the DFZ to both spend less money -and- do a
better job of keeping the Internet stable.

The hitch is: until IPv4 goes away, you're not talking about 100,000
routes. You're talking about 100,000 IPv6 routes PLUS 220,000 IPv4
routes. So it gets worse before it gets better and until IPv4 goes
away, it doesn't get better.


So, how does end come? Surely companies don't just up and refuse to
provide IPv4! Right?

Right. But you don't have to be in the DFZ to provide IPv4 service.
You can use a default route to someone who is. That's the path to
IPv4's decline.

In addition to the redundancy/reliability advantage to participating
in the IPv4 DFZ, there is an economic advantage: DFZ participants can
peer with each other. Peering means you charge your customers to send
you packets but then trade them off to the destination network for
free. The destination accepts your packets for free. He'll charge his
customer to deliver them and would rather receive them for free than
pay someone for the privilege. Today, this cost advantage strongly
outweighs the costs associated with managing 220,000 routes.

As IPv6 use increases and IPv4 use correspondingly declines, these
advantages shrink until provider by provider, participation in the
IPv4 DFZ costs more than a default route would. Exit stage left.
They'll still announce their prefixes into the IPv4 DFZ but they'll
discard the routing table in favor of a default route.

Its the beginning of the end. As folks drop out of the IPv4 DFZ, the
reliability and efficiency of the IPv4 Internet will decline. Static
default routes break easily in non-trivial networks. That creates a
feedback loop encouraging more service migration to IPv6 which in turn
encourages more folks to drop out of the IPv4 DFZ.

Eventually, this destabilizes IPv4 enough that folks start to deploy
IPv6 tunnels to get the IPv4 packets where they need to go. With that
tunnelling in place and IPv4 traffic much lighter than it is today,
its suddenly very advantageous for folks still in the IPv4 DFZ to drop
the zone back to a single router inside the AS so that the IPv6 border
routers don't have to contend with IPv4 at all. IPv4 regains
stability, but the routing becomes opaque and very ineffecient.

IPv4 probably hangs on for quite a while in this marginalized state
but for all intents and purposes its no longer the protocol on the
Internet.

RIP.

Regards,
Bill Herrin



-- 
William D. Herrin                  herrin at dirtside.com  bill at herrin.us
3005 Crane Dr.                        Web: <http://bill.herrin.us/>
Falls Church, VA 22042-3004