Re: QUICker connection establishment with out-of-band validation tokens

[Ted Hardie <[e-mail]>]

I would like to suggest that the Chairs rule this out of scope for the QUIC
working group to consider during the completion of v1.  Given where we are
in the process, taking on the security and privacy analyses required will
not be a short or simple exercise.

To give a flavor of the analysis required, consider this from the proposal:

To save a round-trip time via the proposed out-of-band tokens, the client
> needs to receive the token before sending the connection request to the corresponding
> QUIC server. Furthermore, clients usually do a domain name query to look
> up the source address before they send their connection request. Thus,
> DNS seems to be a suitable place to distribute out-of-band tokens as the
> connection request often directly follows the corresponding DNS lookup
>

First, this proposal contains major assumptions about the DNS interactions
that would appear to run counter to DNS caching semantics and, more
generally, its distribution by recursive resolvers.  While it might be
possible to overcome the implied limitations by the use of short TTLs and
very aggressive use of EDNS client-subnet, the performance impact of doing
so would be unknown and the privacy implications appear to be poor.
Second, it requires a level of trust between the resolver operator and the
entity generating the tokens that is near total, acknowledged in the draft
here:

Note, that to construct valid out-of-band tokens, the resolver needs to be
> trusted by the server hosting the specific domain name. Thus, the
> respective server operator shared in advance the instructions and the
> secret keys required to generate valid tokens for this domain.
>

There are many additional questions raised by the proposal, including when
to consider including the novel RR as additional data, the interaction with
DNSSEC, and the potential need to use a STUN-like service to get a view of
the IP as it will be seen by the QUIC-speaking server.

I appreciate the work of the authors to consider out-of-bind distribution,
but I do not believe this proposal should be considered by the working
group at this time.

regards,

Ted Hardie
(wearing no hats)


On Fri, Apr 12, 2019 at 7:18 AM Erik Sy <[e-mail]>
wrote:

> Hi folks,
>
> I suggest introducing out-of-band validation tokens to save a round-trip
> time during QUIC's connection establishment.
> These tokens can be distributed via DNS resolvers or other QUIC servers
> and can avoid up to 100% of stateless retries during connection
> establishments.
> Below is a detailed description and evaluation on performance, security,
> privacy and scalability of this proposal:
>
> https://svs.informatik.uni-hamburg.de/publications/2019/2019-04-12-Sy-preprint-QUICker_connection_establishment_with_out-of-band_validation_tokens.pdf
>
> Best,
> Erik
>
>