Re:Internet-Draft: Hypertext Transfer Protocol Version 3 (HT...
by skissane on Wednesday December 19, 2018 @05:41PM
(#18720566)
Have any of the people wanting SRV record support in HTTP/3 gone to the working group and proposed it? I searched the working group mailing list and couldn't find any discussion of this topic.
That’s the final answer for HTTP2. They note as well that, were they to have proceeded, that the SRV record type would be inappropriate regardless and some new type would be needed.
Therefore, proposing “SRV” to the HTTP working group for HTTP3 will not succeed. Proposing a new DNS record type might succeed, but it would no longer be called SRV.
Based on their reply, I warn you to set your expectations low: proposing a new DNS record type to address their stated concerns would likely take enough time itself that it cannot ship a final RFC in time to be considered for HTTP/3.
Re:Internet-Draft: Hypertext Transfer Protocol Version 3 (HT...
by floatingatoll on Wednesday December 19, 2018 @06:54PM
(#18721193)
TLS 1.3 (2006) was held up for months-to-years by the "unfortunate limitations of some TLS gateways/proxies", and PCI compliance only requires TLS 1.1 as of this year (2018 — 12 years later).
What solution is available that avoids high-latency additional round-trip requests and would be compatible with today's "unfortunate" gateways and proxies?
DNS-over-HTTPS failed for me constantly while traveling, due to captive wifi portals that try to force you into your browser so they can get your Facebook cookie.
DNS-over-HTTPS does not necessarily address the round-trip concerns, as it is simply a transport layer for DNS requests and does not guarantee the browser's ability to coalesce plural DNS requests prior to sending them up for resolution.
So in light of TLS slowness and DoH unfamiliarity to the marketplace, I personally would expect Mark's described limitations to remain applicable for at least another ten years.
Re:Internet-Draft: Hypertext Transfer Protocol Version 3 (HT...
by wahern on Wednesday December 19, 2018 @08:30PM
(#18721766)
Odd that (AFAICT) nobody mentions the most obvious and useful feature of SCTP compared to QUIC: SCTP preserves message boundaries. That's immensely useful in almost every scenario other than existing protocols (like HTTP) that reinvent chunking.
Building on that, SCTP also supports optional unordered delivery--so reliable but unordered message delivery. I didn't know about this before confirming message boundary preservation with the RFC.
Re:Internet-Draft: Hypertext Transfer Protocol Version 3 (HT...
by rqs on Wednesday December 19, 2018 @10:12PM
(#18722205)
Wow, just reading this RFC makes my more comfortable compare to reading RFC7540. Possibly because they hided a huge portion of technical details under the keyword QUIC.
I'm not sure I should continue to implement my HTTP/2 server now, as HTTP/3 will overlay almost (as far as I can tell) all the benefit that provided by HTTP/2.
One thing makes me confused though, since:
> The QUIC version being used MUST use TLS version 1.3 or greater as its handshake protocol.
And most TLS 1.3 implementation supports RFC6066, then why:
> .... This may be done using the Server Name Indication (SNI) [RFC6066] extension to TLS <bold>or using some other mechanism</bold>.
? I mean, why include any other mechanism while SNI can do the job?
Re:Internet-Draft: Hypertext Transfer Protocol Version 3 (HT...
by vasilia on Thursday December 20, 2018 @04:26PM
(#18729278)
About 30% of all sites have HTTP/2 support. I don't remember where I saw this. I'm curious why the number is so small because nginx has HTTP/2 out of the box. But if your primary users are using Microsoft IE/Edge I have bad news, this browser has worst HTTP/2 implementation. When we do research on streams prioritization Edge showed that it works like ordinary HTTP/1.1. Safari implementation is based on SPDY, so It has priorities but they never been used in our experiments.
https://news.ycombinator.com/item?id=18427795
https://news.ycombinator.com/item?id=18517047
https://news.ycombinator.com/item?id=15906664#15908696
https://news.ycombinator.com/item?id=8549348#8550133
That’s the final answer for HTTP2. They note as well that, were they to have proceeded, that the SRV record type would be inappropriate regardless and some new type would be needed.
Therefore, proposing “SRV” to the HTTP working group for HTTP3 will not succeed. Proposing a new DNS record type might succeed, but it would no longer be called SRV.
Based on their reply, I warn you to set your expectations low: proposing a new DNS record type to address their stated concerns would likely take enough time itself that it cannot ship a final RFC in time to be considered for HTTP/3.
> the unfortunate limitations of some DNS gateways/proxies
With growth in public DNS services like 8.8.8.8 and 1.1.1.1, plus the advent of DNS-over-HTTPS, I wonder for how much longer that will be a problem.
What solution is available that avoids high-latency additional round-trip requests and would be compatible with today's "unfortunate" gateways and proxies?
DNS-over-HTTPS failed for me constantly while traveling, due to captive wifi portals that try to force you into your browser so they can get your Facebook cookie.
DNS-over-HTTPS does not necessarily address the round-trip concerns, as it is simply a transport layer for DNS requests and does not guarantee the browser's ability to coalesce plural DNS requests prior to sending them up for resolution.
So in light of TLS slowness and DoH unfamiliarity to the marketplace, I personally would expect Mark's described limitations to remain applicable for at least another ten years.
[1]: https://datatracker.ietf.org/doc/draft-ietf-tls-tls13/00/
0. https://hn.algolia.com/?query=%22quic%22%20%22sctp%22&type=a...
Building on that, SCTP also supports optional unordered delivery--so reliable but unordered message delivery. I didn't know about this before confirming message boundary preservation with the RFC.
I'm not sure I should continue to implement my HTTP/2 server now, as HTTP/3 will overlay almost (as far as I can tell) all the benefit that provided by HTTP/2.
One thing makes me confused though, since:
> The QUIC version being used MUST use TLS version 1.3 or greater as its handshake protocol.
And most TLS 1.3 implementation supports RFC6066, then why:
> .... This may be done using the Server Name Indication (SNI) [RFC6066] extension to TLS <bold>or using some other mechanism</bold>.
? I mean, why include any other mechanism while SNI can do the job?
The thing I really don't like to see is that they might eventually keep using the "Host" header and thus potentially creating ambiguity.