- From: Salvatore Loreto <salvatore.loreto@ericsson.com>
- Date: Thu, 4 Oct 2012 10:32:39 +0300
- To: ietf-http-wg@w3.org
Hi there, I like the Patrick analysis as it tries to cover all the different aspects of the issue. for Case 1 - NPN based, the major issue I see is the lack of progress the proposal is doing within the TLS wg (note I may be wrong as I don't follow so closely the TLS wg) for Case 2 - it is interesting to see a proposal based on DNS SRV. is that a too optimistic one, also due to the fact that browser are not really supporting (is this still true?) SRV at moment? having said that, I am with Greg and Willy about the fact that Upgrade is the way to go, also because that was the main reason why Upgrade was defined at the time. However I am open and eager to investigate optimization. So to summarize I support 3b but open to discuss about 3a /Sal -- Salvatore Loreto, PhD www.sloreto.com On 10/1/12 8:03 PM, Willy Tarreau wrote: > Hi Greg, > > On Mon, Oct 01, 2012 at 09:49:42AM -0700, Greg Wilkins wrote: >> Patrick, >> >> this is a good analysis, but my main quibble is that case 3b should be >> renumbered to be case 0 >> >> Connecting to port 80 and upgrading to the protocol/version that you >> want to use should be the basically defined way that a http semantic >> connection is established, regardless of wire protocol and version >> used. All other mechanisms (NPN on 443, DNS SRV, cached redirection >> to known HTTP/2.0 ports) should all be considered as optimisations of >> the basic case. >> >> Saving round trips is important and I'm all for optimisations for that >> - but I think it is a MUST that HTTP/2.0 will work in an environment >> where there is only port 80 and the ability to make a single >> connection. >> >> So I guess that means I support both 3a and 3b. >> >> cheers >> >> PS. Does upgrade really mean an extra round trip? Can't we pipeline >> HTTP/2.0 request behind the upgrade request if we are confident of >> success? > There are possibilities for this that we discussed in the network-friendly > draft, basically pass a few URIs in a dedicated header field that the server > is free to consider or not. > > Willy > > >
Received on Thursday, 4 October 2012 07:33:06 UTC