[address-policy-wg] IPv6 PI request is turned down for my multihomed hosting facility - Why?

Hi,

On Wed, Mar 30, 2011 at 11:08:06AM +0200, Joao Damas wrote:
> this is really strange and weird. I have always thought that RIPE policies aimed at good address stewardship (so far this has been linked to conservation, under IPv4) and some consideration to route table size contention (promote aggregation).

Conservation is not one of the primary goals for IPv6.

Routing table size (=aggregation) *is* - which is why the consensus of 
this group, some years ago, was "be restrictive on PI".

Now, this particular problem ("datacenter business") has been showing up
a few times in the last 1.5 years, and in the end, it boils down to a
number of considerations that are NOT easily solved.

 - the distinction between PA and PI is, conceptually, 

    "PA is for ISPs that want to number (their) customers" 

    "PI is for customers that want to number their own(!) network in an 
     independent way"

   --> and this is why the chunk size is much bigger for PA, because
   it's meant to give an ISP a chunk that is so large that they will not
   need to come back for more space in the near future (-> which is good
   for routing aggregation), even taking lots of customers with a /48-each
   into account.  PI is a /48, because "that's enough for one customer".


 - PI has been positioned as "cheap", so it's not a major hurdle for
   small "non-isp" companies that want/need independent address space

 - PA is "expensive", because it's only given to LIRs, and all the LIRs
   around share the expense of having the RIPE NCC do their job

   --> so, if a large number of ISPs change to run their IPv6 business on 
   PI space, and stop paying LIR fees, all *other* LIRs have to pay *more*
   (the NCC's expenses pretty much stay the same, and get distributed to
   less paying members).  So from a fairness point of view, there's good
   reason to ask everybody who is running an ISP-like business to become
   a LIR, and pay their share, staggered by "ISP size".


 - there's another twist to it: we want to encourage access providers
   (DSL, cable, ...) to give their customers a /56 or /48 - which the
   current PA policy permits just fine.  Now if they could get a PI
   block to number their 5 million DSL users, these customers would end
   up with single-IPv6-assignments and IPv6 NAT...

   (We see this in IPv4 today, access ISPs that run all of their DSL
   network on PI space, as the IPv4 policy has a clause that permits
   doing so [transit networks to the customer are considered "part of 
   the PI", and if all you give the customer is a single IP for their
   router...])


... so, given this, "just open PI space and give everbody who asks their
/48" is not without risks and costs - talking about good stewardship, and
such :-)


> Responses like this over the last weeks seem to indicate that
> RIPE NCC's business practices seem to taking over as a more important
> consideration ("you can't get a /48 if that's all you need and it
> makes sense, but if you pay the RIPE NCC you can get a /32 and some
> scarce IPv4 bundled in to boot").

You don't get the IPv4 space unless you demonstrate need.  It's not like
the IPv4 PA allocation is automatic.


> If I were a regulator not intimately involved in RIPE's doings,
> I sure would be raising an eyebrow, or both.

Well - the PI policies are *exactly* what the community(!) made them...

> Would it be time to revisit the coherency of this whole thing as a whole?

Definitely.  There's a "change the PI policy" proposal coming up, and
Sander and I are trying to come up with good ideas to get the whole
complex better "balanced out".

Gert Doering
        -- NetMaster
-- 
did you enable IPv6 on something today...?

SpaceNet AG                        Vorstand: Sebastian v. Bomhard
Joseph-Dollinger-Bogen 14          Aufsichtsratsvors.: A. Grundner-Culemann
D-80807 Muenchen                   HRB: 136055 (AG Muenchen)
Tel: +49 (89) 32356-444            USt-IdNr.: DE813185279