On Wed Nov 20, 2024 at 12:18 AM UTC, C. M. Heard wrote:
> For the record, Fred, I do not agree with that. IP Parcels should have a
> new protocol number (once again, see pont #3 here
> <https://mailarchive.ietf.org/arch/msg/tsvwg/QPVVjD0sGhMz9Xw86Xb_Z3EL6Yc/>).
> One is needed for safety and correctness, as there is NEVER any guarantee
> that despite all efforts something intended to be confined to a limited
> domain won't escape, despite your arguments to the contrary. A new protocol
> number assures that parcel-unaware endpoints will simply drop an IP packet
> containing parcels. Yes, I know, that complicates the parcels spec
> somewhat, but the onus for correctness in all domains is on those who may
> use IP Parcels, not the Internet in general. We've seen a variant of this
> movie before with Segment Routing, I think.
SCTP is an example of a well defined protocol which is only used on well
understood paths, because last-mile consumer grade gateways ("ossified
middleboxes") have a bad habit of dropping anything they don't
recognize. At one time the Apple WiFi access point did this. If a new
protocol number could work from many ends to many other ends, we would
have replaced TCP and UDP a long time ago, and QUIC wouldn't be using UDP.
There's not much difference between "you'll need a new IP protocol for
that" and "you'll have to re-engineer the Internet and the P50 tail is
at least 25 years but measurements are incomplete."
_______________________________________________
Int-area mailing list -- int-area@ietf.org
To unsubscribe send an email to int-area-le...@ietf.org
