[IPv6]Re: Working Group Last Call for <draft-ietf-6man-pio-pflag>

[Lorenzo Colitti <[e-mail]>]

On Fri, May 31, 2024 at 12:11 PM Brian E Carpenter <
[e-mail]> wrote:

> Neither of these arguments is very sound, in my view. However, I'd feel
> more at ease with pd-per-device if it *also* proposed to extend subnet
> prefixes to, say, /80 (with all the changes to IP-over-foo documents that
> would imply).


Honest question: can you explain why you think that would be better? Why is
it OK for your phone to have a /64 when it's connected to a mobile network,
but *not* OK for your phone to have a /64 when it's connected to your
enterprise wifi? Why are you so concerned about that, even though
presumably you think /64 per device in mobile networks was a good thing?

I get that /64 makes you uncomfortable (presumably due to concerns about
address wastage?) but is there actually a rational concern here about IPv6
address space usage? If there is, should we start thinking about changing
cellular networks to use prefixes longer than /64? And if not, then why
should other types of networks be more constrained?

More in general, this was debated at length when pd-per-device was being
discussed. I don't think anyone claimed (or at least, convincingly claimed)
that /64 per device would use too much space. IIRC the strongest arguments
against /64 per device were:

   1. Some RIRs at the time didn't have policies that allowed sites to get
   enough space.
   2. There were networks that had already written addressing plans and
   gotten space from RIRs and would need to renumber.

For the policy issue, I believe that only RIR was RIPE and it was fixed by
a policy change that was already in flight (to address another need) while
the debate was hapening. As for networks that had already gotten space and
would have needed to renumber, the rough consensus ended up being that
those networks could simply not adopt the model.

Also note, switching from /64 to /80 will *not* obviate the need for a P
bit. As soon as /80 is widely supported it will become the new
lowest-common-denominator. Each leaf network will only have /80, and we'll
still need a way to determine whether the network can provide each device
with its own prefix or not.

In general I don't object to trying to change the IPv6 subnet prefix length
from /64 to /80, but it's definitely not an easy change, either in terms of
backwards compatibility, or in terms of getting consensus on, or in terms
of avoiding a race to the bottom.

Cheers,
Lorenzo