Re: Call for Adoption: Invariants

[joel jaeggli <[e-mail]>]

On 2/6/18 6:54 AM, Spencer Dawkins at IETF wrote:
> Asking with no hat, but ...
>
> On Mon, Feb 5, 2018 at 1:00 PM, Mike Bishop <[e-mail]
> <mailto:[e-mail]>> wrote:
>
>     I support adoption.  The way to change the invariants will be to
>     mint a new protocol, and not claim that your new protocol is a
>     version of QUIC.  If it happens to be startlingly similar, all
>     well and good.
>
>
> If you roll a new protocol, that's not a version of QUIC, is it
> obvious to everyone but me whether you can run both protocols on UDP/443? 
>
> I thought I remembered that UDP/443 was chosen because firewalls often
> pass port 443, so that aids deployment. Hence, my question.
In my experience as a vendor of and consumer of commercial firewalls.
UDP and TCP port numbers receive separate treatment.

So a decision to pass TCP 80/443  or UDP 443 would be undertaken
independently.

Outgoing UDP through a stateful NAT/PAT tends to create state for a
binding which will be used to map incoming packets with obverse 5-tuple
to the original sender. unsolicited incoming UDP (for which no binding
or rule exists) will be dropped.

There is the complexity with UDP of determining when you can remove the
state since connection oriented signaling may well be (is) opaque to the
stateful device. if you have to fall back to a timer or some kind of LRU
policy for state entries you will tend break long lived flows that don't
use keep-alives.
> Thanks,
>
> Spencer