Re: [v6ops] I-D Action: draft-colitti-v6ops-host-addr-availability-01.txt

[Andrew 👽 Yourtchenko <[e-mail]>]

On 7/28/15, Gert Doering <[e-mail]> wrote:
> Hi,
>
> On Tue, Jul 28, 2015 at 03:58:04PM +0200, Andrew ????  Yourtchenko wrote:
>> I did not look at it as a problem given that every mobile phone on
>> IPv6 will already get a /64 per host, and the number of mobile phones
>> is dramatically bigger than the number of fixed installations.
>
> Bigger than then number of L2 *segments*, undoubtly.
>
> But no way bigger than "the number of devices that could attach to
> a L2 link" - as all these mobile phone also has wifi, so you have way
> more devices that attach to "shared links".

I agree on the principle, though in reality it's even more complicated
since one device can attach to more than one link, etc. -  so I think
it's actually even worse.

>
> Network structure is also way different - in mobile, all devices attach
> to some sort of aggregation router (via 3G PDP tunnels etc) - while
> in "classic" networking, you have a multitude of independent segments
> that do not normally have aggregation infrastructure or provisioning
> available.  Prefix mobility in a typical enterprise networks (where
> you'd have enough devices in a L2 segment to start think about scaling)
> isn't really there.
>

Agreed.


>> But I pulled that assumption more or less out of my thumb, based on
>> observed anecdata, so would be happy to be proven wrong.
>>
>> If we say we want to absolutely avoid NAT, then something has to give,
>> and I don't know which tradeoff is a better one, both can be argued
>> for and against. I think we might need both.
>
> If you want my opinion, I think DHCPv6-PD to single hosts (= not something
> that does tethering) is not a reasonable approach.

My line of logic was that DHCPv6-PD with a /128 prefix is not much
different from IA_NA.

>From there it is trivial to say "no, thou shalt not do /128, use
something shorter", and have the luxury to not define what is the
value of "shorter" is because it's just a prefix length, and we can
change per-situation.

Going from one IA_NA to 10 IA_NA to 30 IA_NA is much harder, so we
will be forced to freeze on the "right" value.

>
> If something does tethering, you need to decide what you're talking about,
> "enterprise-ish", "mobile" or "homenet".  In mobile, DHCPv6-PD, or just
> sharing the PDP /64.  In homenet, HNCP or DHCPv6-PD.  In the enterprise?
> No idea what can be implemented with the typical constraints on
> trackability, security, etc.  (like: if the device attaches to *this*
> network, it's permitted to go *there* by IP ACLs - whether or not this
> is a reasonable approach in itself anymore stands to be debated, but it
> will be with us for a long time).

I think you pointed the crux of the issue here: the enterprise network.

An enterprise does not want one uncontrolled device behind another
uncontrolled device, so realistically it will run IA_NA. And because
it will want the cheapest possible hardware to provide the function,
it will try to minimize the number of IA_NAs per host.

And unless the hosts will crash and burn if they can not get X IA_NAs
upfront (they won't), this number will go down to a number providing
the most important 90% of functionality.
The remaining 10% will have to adapt.

--a


>
> Gert Doering
>         -- NetMaster
> --
> have you enabled 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: [phone]           USt-IdNr.: DE813185279
>