[ppml] Solicing comments: IPv4 to IPv6 Migration Incentive Address Space

[William Herrin]

On 6/26/07, Scott Leibrand <sleibrand at internap.com> wrote:
> A couple of critiques of this proposal, which may prevent it from
> gaining consensus:

Thanks Scott, I appreciate the comments.

>     * Anyone with an IPv4 assignment from ARIN can request an IPv6
>       assignment under existing policy, and they will likely receive it
>       with no trouble whatsoever.  (We just requested ours, and it was
>       remarkably easy, since we didn't have to justify the size of our
>       allocation like we do for IPv4.)

I agree. Yet IPv6 uptake increases at a pace which won't meet the 10%
deployment mark before IPv4 exhaustion.

Three years from now, that will be a real problem for Internap. You
sell a high value service with cleverly optimized routes. Without IPv6
in place, no more IPv4 addresses means no new customers despite your
technology. Cogent gets them instead because this week Cogent is the
one with spare IP addresses. Next week it'll be someone else, Level 3
maybe, but not you. You're stuck until a customer leaves and forfeits
his IP addresses.

That's why I've proposed an -Incentive- space, not merely a migration
space. We have maybe three years to get IPv6 to a point where dialup
users, dsl users and folks with wifi laptops at Starbucks receive IPv6
addresses automatically alongside the IPv4 addresses. That's three
years to get IPv6 in the hands of the end users and every network
admin in the path from us to them. If we can't bring Mohammad to the
mountain, maybe its in our interest for the mountain to make a trip.


>     * While reservation of a /32 is trivial space-wise, reservation of
>       that many distinct IPv6 netblocks, with the expectation that they
>       can all be routed, will accelerate routing table growth.

I think that's a legitimate concern that deserves investigation. In
the worst case scenario, 100% of the organizations who announce IPv4
space would announce all of the IPv6 blocks they receive under this
proposal. Since IPv6 addresses are 4 times the size of IPv4 addresses,
that would increase the memory demand on routers by a factor of 4. A
factor of 5 if you consider that they also have to maintain the IPv4
table.

That yields two questions: 1. Is anything approaching the worst case
scenario a plausible outcome, and 2. What would be the impact of such
an increase?

The impact of the worst case scenario is nasty but survivable. Some
folks are still running on 256 mb routers. That dies. So do 512mb
routers and 1 gb becomes tight. On the other hand, servers come with 8
gb these days and expand to 32 gb without too much difficulty or cost.
That has inverted from the early days when your cx-rp1 had 64mb and
your sparc web server had 32. Granted this is an oversimplification:
routers have multiple processors and multiple task-specific sets of
memory. Nevertheless, its reasonable to expect that router vendors
could make an 8-factor jump in routing table capacity in the next
generation of products without unduly increasing the cost.

So the next question is: is it real? I don't think so. I'd need more
data (and maybe the folks at ARIN can help out here) but I suspect
that something on the order of 75% of the route slots are filled by 5%
of the organizations. I'll bet that the much of the same 5% will
prefer a single contiguous block of IPv6 space and will neither need
nor choose to claim the space under this proposal.

Does anyone have hard numbers on how many discontiguous v4 allocations
are held by the top 1%, 5% and 10% of ARIN registrants?

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