RE: Proposed response to the 3GPP Liaison Statement

["Lubashev, Igor" <[e-mail]>]

RFC 8558 is very relevant.  It says, in Recommendations, that implicit signals should be replaced by explicit signals when originators wish to provide them to network elements on the path.

Since QUIC is removing implicit signals, I read the question from 3GPP to mean “what are you replacing them with [for originators who wish to provide them]?”.  QUIC WG response reads as: “nothing, short of attaching encryption keys to packets”.  Even if QUIC v1 today can offer no replacement for the removed signal, QUIC WG should make an effort to at least suggest a practical way to satisfy RFC 8558’s recommendation of an explicit signal “for those who want it”.  Maybe point to some other specific layer where the WG believes the signal belongs (and can be deployed relatively quickly)?


As for the content of the reply, it includes “The impact upon network operators was discussed extensively as part of that decision.”  Were network operators a part of those discussions, and what was their feedback?


  *   Igor


From: Ted Hardie <[e-mail]>
Sent: Thursday, October 31, 2019 9:58 AM

On Wed, Oct 30, 2019 at 6:14 PM Martin Thomson <[e-mail]<mailto:[e-mail]>> wrote:
https://tools.ietf.org/html/draft-iab-path-signals-03<https://urldefense.proofpoint.com/v2/url?u=https-3A__tools.ietf.org_html_draft-2Diab-2Dpath-2Dsignals-2D03&d=DwMFaQ&c=96ZbZZcaMF4w0F4jpN6LZg&r=Djn3bQ5uNJDPM_2skfL3rW1tzcIxyjUZdn_m55KPmlo&m=bqfwnQ_X0sS2s_-ECuNdlHuwGgGS1ExL0htjWFkdcmA&s=n0t-4YZjV8NbJEQGurqtHwTmD1WjZbvGY_rqh6eYGNE&e=> seems like it might be a helpful supporting document.
If you do choose to cite it, it has been published as RFC 8558, so it would be best to use that.

Ted


On Thu, Oct 31, 2019, at 12:09, Mark Nottingham wrote:
> QUIC WG participants,
>
> Lars and I have discussed a response to this statement and plan on
> sending the message below. If you have comments, please respond to this
> e-mail by 6 November.
>
> Regards,
>
> ~~~
> Thank you for your input to our specifications.
>
> Regarding the encryption of headers in addition to payload, this was a
> conscious design decision by the Working Group. Transport header
> encryption prevents modification by middleboxes, thereby avoiding
> ossification of the protocol, and also makes traffic analysis more
> difficult. The impact upon network operators was discussed extensively
> as part of that decision.
>
> Regarding the spin bit, we believe that your characterisation is correct.
>
> Regarding packet loss measurements, QUIC does support them, but only by
> the endpoints of communication -- not by third parties observing it.
> Keys may be exported by one of the endpoints to allow decoding of
> packet traces (e.g., with Wireshark), but that is not part of the QUIC
> standard.
>
> Regarding performance analysis, it would be helpful if you described
> your use case in more detail. However, it's important to know that the
> Working Group decided early on that QUICv1 would focus on the use case
> of HTTP/3.
>
> In turn, HTTP standardisation focuses primarily on the Web browsing use
> case; while we acknowledge that there are other uses of the protocol,
> and we accommodate them where possible, we cannot change the protocol
> in ways that harms that use. Exposing transport metadata would likely
> do so.
>
> Kind regards,
>
> Mark Nottingham and Lars Eggert, QUIC Working Group chairs
> ~~~