Re: [AVTCORE] QUIC version 2 and RFC 7983bis

[Lucas Pardue <[e-mail]>]

Hey

On Thu, 12 May 2022, 00:54 Bernard Aboba, <[e-mail]> wrote:

> Lucas said:
>
> "Thanks for sharing this analysis. It might also be worth mention
> draft-ietf-quic-bit-grease [1], which allows endpoints that don't need to
> worry about demux to change the "second-to-most significant bit of the
> first byte in every QUIC packet". Not sure if you need a reference
> (informally?) to the greasing draft draft."
>
> [BA] How easy is it for an application to indicate that it "wants to worry
> about demux"?  As an example, the WebTransport API doesn't provide a
> mechanism to influence QUIC transport parameter negotiation.
>

There's probably two answers, depending what you're looking for.

As you indicated, endpoints tell their peer they are willing to receive
greased QUIC bits via the grease_quic_bit transport.

1. QUIC Implementations provide a range of configuration approaches, since
we aren't defining common QUIC transport APIs. I might expect that
implementations supporting QUIC bit grease would offer configuration
options for whether to send the Transport Parameter, and whether to grease
the bit on sent packets if the peer sent the parameter.

2. Other extensions to QUIC are permitted to augment the behaviour of
endpoints that receive grease_quic_bit

To give an example of answer 1. If I'm running an HTTP/3 server on UDP 443,
I wont expect other types of UDP-based traffic and I don't benefit from the
client sending the QUIC bit to 1. In this deployment, my grease bit
configuration options are clear. Deployments already have to make such
choices about several socket oriented issues.

For WebTransport on the client side I'm not as familiar with your
requirements. Is there really some expectation that clients would want to
demux UDP traffic on the same listening port? If not, I don't see much
issue with letting browsers apply common approaches to the QUIC bit grease
(Firefox already sets it IIRC). If there is a potential issue, maybe that
needs to be discussed in a suitable WebTransport venue.


> draft-ietf-quic-bit-grease should probably reference RFC 7983bis rather
> than RFC 7983 since the latter doesn't cover QUIC multiplexing.
>

I suspect Martin Thomson is lurking here. But I went and created an issue
against the draft to capture this
https://github.com/quicwg/quic-bit-grease/issues/23

Cheers
Lucas