Re: QUIC re-chartering: include DNS-over-QUIC?

[Paul Vixie <[e-mail]>]

On Thursday, 6 February 2020 21:27:14 UTC Erik Nygren wrote:
> On Thu, Feb 6, 2020 at 4:16 PM Paul Vixie <[e-mail]> wrote:
> > i see DoQ as nec'y for QUIC's success but not relevant to DNS beyond the
> > fact that DNS is a convenient second example of how QUIC can be used. if
> > there are use cases or failure modes not handled well by DoT, i'd like
> > to learn more about those.
> 
> Some possible reasons (not comprehensive):

forgive the long delay in responding, please.

> * Reduce the impact of head-of-line blocking during loss.  If packets are
> lost, only the request(s) associated with the loss are impacted.

you're assuming that the loss isn't self-induced. sometimes slowing down is 
exactly what's needed. for example when a busy full resolver has a queue of 
hundreds or thousands of requests to be sent to a zone authority, rate 
limiting of UDP is necessary to prevent a meltdown. bandwidth is finite. we 
sometimes load-balance toward several of a zone's authority servers even 
including those known to not have low-ish RTT, just to avoid flooding one of 
them, or to avoid flooding our nearby links. moving to QUIC won't help with 
any of that.

> * Faster resumption and 0RTT behaviors (although you get some of these with
> TFO + TLS 1.3 0RTT).  But this substantially reduces the performance impact
> when new connections need to be established.

in DNS, new connections rarely need to be established. the reason RFC 6013 
permitted connections to go into a quiet (resumable) state and keep the 
learned cwin in compressed state so that the cost of that state would be no 
worse than what was needed for syn flood protection. since you mentioned TFO i 
want to say two things, first it was insecure and can't be used, exactly as we 
predicted (william simpson and i), and second TCPCT allowed for payloads as 
part of the initial or resumptive segments to get exactly the 0RTT benefit. 
but i don't want to drone on like a sore loser. the point is, in DNS, novel 
connections are rare, and careful statekeeping to support resumption (such as 
DNS cookies as defined in RFC 7873) is already present in pre-QUIC DNS.

> * The previous also means servers can be more aggressive about timing out
> connections to reduce the state they need to keep, but while minimizing
> perf impact, which can be critical at-scale.

this reduces to a noop because the statekeeping is subject to botnet attacks. 
just as all TCP listeners have to implement syn flood protection, so it is 
that all stateful endpoints (whether DNS, TCPCT, TFO, or QUIC) have to be 
aggressive about timing out connections they need not keep fully open. QUIC 
has no special capabilities or requirements in this regard.

> * Better ability for load balancers at-scale to do creative things,
> including being able to recover better when Anycast moves routes between
> data centers.  (The server-supplied Connection ID means that options become
> available to better recover without needing to reset re-establish
> connections and without needing to share state between clusters.)

while it's theoretically possible to share DNS cookie state (RFC 7873) across 
many instances of DNS Anycast, the cost:benefit of that complexity is less 
compelling than the cost:benefit of having to renegotiate during route changes 
either from mobile-ip or network-in-the-middle attacks. i don't think QUIC has 
different game theory than DNS cookies have in this regard. route changes are 
orders of magnitude less frequent than same-pair reuse, and always must be.

> * QUIC provides a framework for future extensions that may be useful in the
> DNS space.  For example, FEC might actually be useful for DNS in some cases.
> 
> I'm sure there are plenty more reasons...

i appreciate the time you put into listing these initial motivations. however:

> On Thu, Feb 6, 2020 at 4:16 PM Paul Vixie <[e-mail]> wrote:
> > i see DoQ as nec'y for QUIC's success but not relevant to DNS beyond the
> > fact that DNS is a convenient second example of how QUIC can be used. if
> > there are use cases or failure modes not handled well by DoT, i'd like
> > to learn more about those.

...my question seems to stand. if the potential motivations you've listed so 
far were actual, then we'd have had DNS over SCTP decades ago, or something 
like DTLS that actually worked at scale rather than just in a lab, or we would 
have been able to get TCPCT the code point we needed to experiment globally, 
or we would have had UDP options by now.

i consider DoQ inevitable, but for QUIC's purposes ("proof of broader 
applicability"), not DNS's purposes ("working better"). DNS is old and warty 
but it does move and it has evolved.

-- 
Paul