Re: [v6ops] IPv6-Only Preferred DHCPv4 option

[Mark Smith <[e-mail]>]

On Thu, 5 Dec 2019 at 13:20, Lorenzo Colitti <[e-mail]> wrote:
>
> On Thu, Dec 5, 2019 at 9:53 AM Mark Smith <[e-mail]> wrote:
>>
>> If hosts only acquired IPv4 addresses via DHCPv4 based on when the first IPv4 application started running, then when a host is only running IPv6 only applications, it would never acquire an IPv4 address because it wouldn't need one, meaning never try to use DHCPv4 (so would avoid the link-layer broadcast DHCP_DISCOVERS) and never use ARP (avoid link-layer broadcast ARP Requests).
>
>
> Speaking as an OS implementer: how do I know whether an application is an IPv4 application?
>

So these are reasonable questions, and they're uncovering some
assumptions and a view point I have:

- the hosts in question are commonly single homed (and I should know
better *, however I would't be surprised if that is the unstated
assumption of most people when they imagine the scenario being
discussed)

- the on-demand I'm imagining is inspired by the dial-on-demand that
used to be implemented with ISDN, triggered by IP traffic. ISDN
connection establishment happened quickly enough that it was tolerable
by end-users. If a DHCPv4 transaction occurred within 100 ms, users
will perceive it as instant - "Response Times: The 3 Important Limits"
- https://www.nngroup.com/articles/response-times-3-important-limits/

- IPv6 only applications clearly indicate they're IPv6 only, by only
opening IPv6 sockets.

This latter point is probably a significant difference between what is
being proposed here and what I'm thinking. My thinking is probably
about a passive and longer term mechanism that would result in IPv4
disappearing over time, rather than an active mechanism that purposely
switches IPv4 off on dual stack hosts.


Responding to these points for completeness:

> When it calls getaddrinfo() with AF_UNSPEC? Pretty much all apps do that.

They wouldn't be IPv6 Only applications.


> When it creates an IPv4 socket? But what if it's creating an IPv4 socket for the purpose of connecting to another network on the system (e.g., cellular data) that does have IPv4?

Yes, this would be an issue with what I was describing.

What I'm thinking about would probably more result in "IPv6 Only"
being a host level status on a host, covering all interfaces (1 or
many), rather than an interface status. In other words, if a host is
only currently running IPv6 Only applications, then there is no need
for any IPv4 addresses on any of its interfaces.


> When it does a connect() on an IPv4 socket? But what if that's just a happy eyeballs stack that is using IPv4 because it might be available?

As above, not an IPv6 Only application if trying to use HE.


> What if the app knows it needs IPv4 and will refuse to do any networking if it doesn't find an IPv4 address on the system?

Not an IPv6 Only application.


I see value in both applications being explicit about what Internet
Protocol they want to use (e.g. on demand addressing per above), and
also being agnostic about it.



Regards,
Mark.



(* "The Rapid Rise of the Mobile Multi-Homed Host" -
https://www.ausnog.net/sites/default/files/ausnog-2013/presentations/D01%20P06-the%20rapid%20rise%20of%20the%20mobile%20multihomed%20host.pdf)