- From: Eliot Lear <lear@cisco.com>
- Date: Mon, 20 Aug 2012 12:56:27 +0200
- To: Willy Tarreau <w@1wt.eu>
- CC: Yoav Nir <ynir@checkpoint.com>, "ietf-http-wg@w3.org Group" <ietf-http-wg@w3.org>
Yoav, Willy, On 8/20/12 11:02 AM, Willy Tarreau wrote: > >>> What about using bonjour / mdns / etc? >> Like DNS SRV records, it's great when it works, but both are often filtered (or not implemented) Good for the home. And nothing stops this. By the way, I am still not convinced, myself, that SRV records are the correct approach. Or at least not what we call SRV records today. Again, the http URI schema defines a port #. To me, that's the end of SRV as we know it. But, one could imagine a record that would address this concern. How about this: www.example.com IN BLARF (80, HTTP2.0,other capabilities...), (83,HTTP1.1,other capabilities), ... IN AAAA 2001:... And in the default case fall back to 1.1, which is what Phillip pretty much suggested. This gets you to 2.0 before you connect. Then for those cases which are not in the DNS, as willy suggests, you do it in path (NPN, upgrade, or whatever). But here's the question: would such a scheme really provide substantial benefit? That is (a) does it really save an RTT and (b) does that RTT matter to people? > > We must never forget that what made HTTP succeed is also the fact that > it's self-contained, does not require additional side-band protocols, > works end-to-end over a single connection and is 100% NAT-friendly by > not storing any IP-related information anywhere. > Well indeed. I work at a routing vendor, for crying out loud. Our routers probably don't have DNS entries when they're being configured. Eliot
Received on Monday, 20 August 2012 10:56:55 UTC