Re: Invariants draft

[Martin Thomson <[e-mail]>]

On Fri, Dec 1, 2017 at 5:26 PM, Mikkel Fahnøe Jørgensen
<[e-mail]> wrote:
> Thanks for writing up.
> Some input:
> Appendix A, incorrect assumptions:
> - A 5-tuple is unique to a single connection.
> - A client uses ephemeral ports.
>
> More generally: I find it unfortunate that UDP is made an invariant, anf
> possibly even IP.

That's intentional.  For instance, QUIC over DCCP would be something
else entirely and not covered by this promise.  The same applies to
layering QUIC directly on top of IP, or any imaginable substrate (like
your embedded device bus example).  Nothing we do here would prevent
those things.  Though each - as an entirely new thing - would have to
overcome its own challenges.