ARIN IPv6 Policy Proposed

Brian E Carpenter brian at hursley.ibm.com
Tue Mar 13 10:22:00 EST 2001


Antonio,

Let's do a slight reset here, because I don't want to seem to
dispute operational experience.

The most important suggestion in the draft is: if you allocate subnet space,
allocate a /48. Otherwise, allocate a /64.

We seem to be arguing about how to make that judgement. I'm not even sure it's
worth arguing about or trying to write down global criteria for making the
judgement. My gut feeling is to be more liberal about allocating /48s than
you want to, because we have a *lot* of them. But as long as we stick to
allocating either /48 or /64, to keep things clean, it doesn't ultimately
matter if different operators make different judgement calls.

  Brian

Antonio Querubin wrote:
> 
> On Mon, 12 Mar 2001, Brian E Carpenter wrote:
> 
> > Antonio Querubin wrote:
> 
> > > Just some random thoughts on several of the above recommendations:
> > >
> > > With regard to #1 I'm curious as to what constitutes a 'home network
> > > subscriber' in this draft?  Our experience with what we generally consider
> > > home network subscribers to be is that none subnet at all.  That being the
> > > case, why assign a /48 when a /64 is quite adequate while still providing
> > > the home user with full capability?
> >
> > Because what we expect to see, technologically, is rapid growth in in-home
> > (or in-vehicle) networks. So it is risky to assume that current practice
> > represents the future.
> 
> Rapid growth in home networks has already been occurring but we don't see
> subnetting happening in that environment.  Is this draft to be reviewed
> and revised based on operational experience or on something that to me
> seems to be somewhat unrealistic?
> 
> > > With regard to #2 most small organizations do not bother to subnet - they
> > > generally use switches to divide up traffic.  Subnetting requires routers
> > > of which the majority aren't IPv6-aware anyway.  Another alternative for
> > > #2 would be to combine it with #3:  small enterprises be assigned one or
> > > more /64s while very large enterprises receive a /48.  It seems to me that
> > > we should use a SLA for it's intended purpose - that it be specific to a
> > > 'site'.  But if sites are being assigned /48 where does that leave the
> > > NLA
> >
> > We're asserting that due to the growth, we expect even small organizations
> > will become sites in the sense of needing subnets.
> 
> No problem with that argument but why begin assigning them at the /48
> prefixlength?
> 
> > > It seems that #1 and #2 above could/should be qualified with having a
> > > requirement to subnet or be geographically dispersed.
> >
> > Do you want to be in the business of making that judgement on your users?
> 
> We do that now and why would that change just because we have lots of IPv6
> address space to give out?  The customer still makes the determination of
> how much address space they need and we accomodate what their stated
> requirements are.  But if someone were to come to me and say 'I need 200
> or so /48s' do you really think I'd say sure without asking them just a
> few questions?  The questions I'd ask of a large organization are still
> the same questions I'd ask of a small organization.
> 
> Consider ARIN's existing IPv4 allocation guidelines.  In order for an ISP
> to obtain additional address space ARIN expects the upstream to enforce
> guidelines similar to its own.  Eg. you need to demonstrate 80%
> utilization of existing space before you can acquire more.  What applies
> to the ISP applies to the end user.
> 
> With IPv6 I imagine ARIN will continue to impose some kind of criteria
> based on this draft.  While one might assume the IPv6 allocation criteria
> might be noticeably more liberal than the IPv4 criteria is there any
> reason for ARIN not to expect the ISP to impose similar IPv6 criteria on
> its downstream customers?  That being the case the ISP still has to make
> judgement calls.



More information about the V6wg mailing list