[arin-ppml] Preemptive IPv6 assignment

[Christopher Morrow]

On Fri, Oct 8, 2010 at 10:28 PM, William Herrin <bill at herrin.us> wrote:
> On Fri, Oct 8, 2010 at 11:20 AM, Christopher Morrow
> <christopher.morrow at gmail.com> wrote:
>> according to the presos at ARIN this week... it's pretty simple to get
>> your for-real addresses. one would suggest a solid numbering plan, and
>> understanding of what technology you may need to employ for your
>> deployment are good to have in hand, though not required for initial
>> assignment.
>
> Sure. Getting IPv6 addresses is easy. Like getting a vaccine. You pay
> for it, get pricked, endure and it's all good.
>
> But do you only want to assure that -your- kids to get the vaccine? Or
> do you want -everybody- to get the vaccine so that the disease itself
> dies out and doesn't haunt you again year after year? If we've truly
> learned how to beat the disease, is it not worth it as a matter of
> public policy to preemptively provide everybody with the vaccine?
>
> We beat smallpox. Let's beat IPv4. ;-)

I'm not sure I buy the 'beat ipv4' ... I have a very good feeling that
it will outlive me in the live network. (albeit with dwindling traffic
rates)

>
>>> With the new tech out there like 6rd, what's your best guess as to
>>> what size a preemptive ISP allocation should be for those 2661
>>> ARIN-region ISPs (75%) that haven't yet gotten their IPv6 allocations?
>>
>> that probably depends a LOT on what sort of ISP these are, and what
>> their deployment today looks like (for v4 services). For instance, are
>> they a:
>>  o hosting company (seems like fairly simple 'drop a v6 addr on your
>> gear, slap one toward the customer-ip, done!)
>>  o transit/business ISP (put ipv6 on your devices, offer prefixes to
>> your customers, done!)
>>  o consumer ISP (cable/dsl/ftth/etc - see deployment situations for issues)
>
> Do you have any ideas on how to tell the difference in a relatively
> automated manner? I know how to test for whether you hold an

nope, but if the provider is prompted at next-interaction-with-RIR ...
it gets simpler, eh? Heck there's probably even a case to be made for
something like (taking my fav past job as an example):

701 - <some sized allocation, a /27 heather justified I believe)
702/3 - ditto to above
11486 (hosting) - one /32 per datacenter. It's plausible that there
could be 30-40k customers per DC (12 DC's in the US or so... so /28).
19262 - at 8.5m (and growing) customers - as much as a /24 for native,
plus probably a /24 for 6rd-ish things (this is with a /48 for native
customers, with a /56 it's far less, though regionalized /32's seems
reasonable still so about a /28)
vzw - 60m customers and growing - strictly on the /56 allocations 1 /28.

for the verizon companies (not counting private networks, one I know
of probably drops another /24 or so of space requirements) that's
about a /23, or on nibble boundaries a /20 ...I'm  sure heather  could
wrap some customer figures around that, but it's far beyond what you
could deduce externally/automatically.

> allocation, but I'm not sure how to test for what kind of allocation
> holder you are.
>
> If we can't tell the difference, does that negate the value in
> preemptive allocation? If it doesn't then let's make an educated guess
> as to what size is likely to be useful, recognizing that anyone who
> finds they received the wrong size is in no worse shape than they are
> without preemptive allocation.

I don't think you can... but APNIC's 'easy button' model isn't all bad.

-chris