Re: the race to the bottom problem

[Alexandre PETRESCU <[e-mail]>]

--
Alexandre Petrescu
[e-mail], tél [phone]

Le 07/11/2020 à 08:31, Lorenzo Colitti a écrit :
> On Sat, 7 Nov 2020, 08:17 Mark Smith, <[e-mail] 
> <mailto:[e-mail]>> wrote:
> 
> The current bottom is a single /64.
> 
> Any provider who is only giving out a single /64 to customers who 
> might or will need more than one /64 has already raced to the
> current bottom.
> 
> A "deeper bottom" or longer default allowed prefix length will just 
> facilitate further racing to the deeper bottom. It won't encourage 
> giving out more than one /64 for those who might need them; it'll do
>  the opposite.
> 
> 
> Exactly.
> 
> The fixed 64 bit boundary prevents this by defining the bottom in
> the *only* way that actually matters: by ensuring that if a network
> goes below the bottom, many hosts just won't work.

I think that might be a speculation explaining the past.

> If it were possible to hand out less than a single /64 and have 
> things work acceptably for their customers, then someone, somewhere, 
> would be doing it already.
> 
> If we change the standards to make this technically possible but ask 
> operators not to do it, some will just do it anyway. After all - if 
> it works for them, why would they do otherwise?

A Variable SLAAC would not request operators to do something else than
they do already do today - allocate a /64 to end users.

This Variable SLAAC would allow _some_ operator and non-operator people
to not use IPv6 NAT, or 1subnet-restricted 64share or IPv6-in-IPv4 VPN
single points of failure and audio quality reducers.

> Because an RFC they maybe haven't even read says they shouldn't?

I think operators should all read RFC7934 which is a BCP and whose title
is "Host Address Availability Recommendations".

That BCP recommends "that networks provide general-purpose end hosts
with multiple global IPv6 addresses".

Do you think a single /64 represents multiple IPv6 addresses?

> The logical consequence is that we end up at /128. This is 
> particularly risky because any operator with lots of experience with 
> IPv4 will find "one /128 per host" obvious, expected and natural.
> One device has one IP address, right? And once get to that point,
> we'll need to bring back NAT in order to extend further. That would
> pretty much make the entire transition to IPv6 pointless. We might as
> well have stayed with IPv4.

I tend to agree, although not entirely.

> 
> Personally I think that once there's more industry experience and 
> IPv6 is widely deployed in all network types (say, once we reach 80% 
> of the internet, and many networks are IPv6-only), we might be able 
> to safely move the bottom, once. But it's much too early for that.

Might be too early - care to explain why?

Or maybe a bit late you meant?

Late, because IPv6 hotspots and similar moving IPv6 networks, all
plugged onto 4G, are already there.  Smartphone tethering uses 64share
some mobile WiFi access points use IPv6-in-IPv4 VPNs or, worse, IPv6
NATs.  I agree it is not 80% of IP deployments, but it goes in that
direction.

Alex