Re: Confirmation to advance: draft-ietf-6man-ipv6only-flag-05

[Fernando Gont <[e-mail]>]

On 21/5/19 05:37, Philip Homburg wrote:
>> Most usually, it's not sites that decide to use link-locals, but hosts.
>> -- it's a host-side policy that is *not* selected or configured on a
>> per-network basis.
>>
>> That said, in all cases, I expect the network to "just work". Users
>> should not be required to understand this sort of technical details for
>> them to get a good experience.
> 
> Networks don't 'just work', except in rare cases of completely disconnected
> networks.

In the vast majority of cases, the only thing you need to do is select
network and password. The rest happens automagically.


If I understand correctly, you were arguing that IPv4 SD should be
disabled by default for IPv6 only networks.

For the average user, this means that if they actually needed, now they
are in trouble.



> Today the vast majority of all networks has some kind of router. And in 
> simple cases like CPEs, they get configuration from the network. My proposal
> is specifically in the case a router provides IPv6 connectivity.
> 
> If we now tell hosts to stop doing IPv4 links local in that case (i.e. when
> IPv6 connectivity is available) then that can save us a lot of headaches
> later.

What saves headaches is to not disable functionality. At the end of the
day, I wonder the extent to which a few IPv4 packets affect battery on a
normal network as opposed to multiple 200MB+ apps  that drain the CPU
for even the most trivial stuff, continuous wake up do to online
presence apps, etc.


Note: This is not an argument against or in favor of this mechanism. BUt
rather a question whether disabling functionality that may very well be
useful.



> Of course, can choose to not that, just in case in the future a legacy IPv4
> printer is around on an IPv6-only network. But in that case we will have to
> live with the mess of hosts trying to do service discovery using IPv4 
> link local until forever.

Indeed. Not forever, but certainly for a long period of time.

And no, it's not just the printer. a bunch of the smart devices you can
connect at home (cameras, smart plugs, sensors, etc.) are still IPv4
only. An "IPv6-only network" will just break all that.

Not that I like it. But my personal taste doesn't change reality per se.


> 
> If we say: in the context of IPv6, DHCPv4 signals the presence of native IPv4
> on a link, then we make it easy for hosts to figure out what is going on.
> We also make it easy to get rid of IPv4.
> 
> For users that doesn't matter. The CPE can easily have a DHCPv4 server
> that will be on by default in the near future and will be off by default
> later. And will be dropped completely when there is no IPv4 support anymore.
> 
> In managed networks, it makes clear that if the network admin wants IPv4
> there is a DHCPv4 server/relay. Without DHCPv4 server, there should be no
> IPv4.

If you *don't want IPv4*, you filter at layer 2. Otherwise you're mostly
advising your network on the intent.  But nodes are free to communicate
over ipv4 if they wish.

auto-linklocal is, in a sense, -- and by definition --  a kind of
anarchy. Such "anarchy" has proven valuable, both when the network
fails, or there are issues with the infrastructure (or there is no
infrastructure).

-- 
Fernando Gont
SI6 Networks
e-mail: [e-mail]
PGP Fingerprint: 6666 31C6 D484 63B2 8FB1 E3C4 AE25 0D55 1D4E 7492