Re: Hidden connection spawning

[Mikkel Fahnøe Jørgensen <[e-mail]>]

When you say you prefer option 2 (0-RTT), do you mean to not hide the
handshake, or do you mean the until now unlisted option 4:

4. carry out a full 0-RTT handshake over a separate secure channel (QUIC or
otherwise)

Option 4 may skip the QUIC encryption layer if the outer layer is equally
strong (this is a can of worms).


Correction: in my earlier mail I wrote “not complex” in relation to option
1, the “not” should not have been there.

On 27 July 2018 at 02.31.03, Kazuho Oku ([e-mail]) wrote:

2018-07-26 15:31 GMT+09:00 Mikkel Fahnøe Jørgensen <[e-mail]>:
> I came to think about a use case where you want to spawn a new connection
> from an existing connection because the new connection runs a different
> protocol. For example a http connection where you want to run a separate
> live conference call, or to control tunnels from http control connection.
>
> You have roughly three options:
>
> 1. extend the current protocol with the new feature set in extension
frames
> or similar
> 2. develop or use a purposes specific QUIC protol using 0-RTT for
spawning
> the connection
> 3. negotiate the new connection handshake within the current handshake.
>
> The first solution is not complex, limited, and not modular.
> The second solution can be used to get information about the connection,
and
> for example block connections that appear to be doing that.
> The third solution allows for an invisible connection spawn, but requires
> addition handshake logic which is already complex as it is.

I prefer the second approach, since it allows spawning connections in
0-RTT across different transports.

For example, I think that it would be nice to have a way to spawn a
0-RTT HQ connection from a HTTP/2 connection (via alt-svc).

Note also that there is an individual draft submitted to TLSWG that
discusses of adding a way for a TLS server to issue session tickets
that could be context-specific: see
https://datatracker.ietf.org/doc/draft-wood-tls-ticketrequests/.

>
> I suggesting solution 3 as possible option, although I am not convinced
that
> this is worthwhile, at least in V1.
>
>
> This was in part inspired by recent discussion on tunneling, and in part
by
> the following issue where partial reliability is being discussed and
> possibly shoe-horned into an extension. I am not following that issue
> closely though - it is just an example.
> https://github.com/quicwg/base-drafts/issues/1606#issuecomment-407951495
>
>
> Mikkel



-- 
Kazuho Oku