On Fri, Feb 27, 2015 at 09:45:10PM +0100, Patrik Fältström wrote: > > On 27 feb 2015, at 21:37, Nico Williams <nico@xxxxxxxxxxxxxxxx> wrote: > > On Fri, Feb 27, 2015 at 08:07:25AM +0100, Patrik Fältström wrote: > >> My feedback to Andrew when he presented this to me was that: > >> > >> - In general I am nervous of moving HTTP header attributes into the > >> DNS, as it might create inconsistencies when for example the data in > >> DNS do not match what is in the HTTP header, and we already have a > >> content-negotiation mechanism in HTTP > > > > If anything, it may not provide the optimization that's desired. (Any > > numbers?) > > Ok, to go back in history, this is why I originally did believe more > in Gopher than Web... :-) But I wanted current numbers, as in performance numbers :) > I though a proper negotiation would be to know already in the source > of a referral what kind of data the target was. This was how Gopher > worked, but Web was different. The link was (is) neutral and the > negotiation happens at the target. Very large references would have been unwieldy in HTML back then. Content-addressed storage is still unwieldy for cross-domain referencing, and always will be. > That story, early 1990's, gave me the lesson that the "correct" > solutions do not always win. The "best" solution wins. That's hardly the only case, and I'm not sure that your approach would have been more correct. The more metadata [about a target resource] you put into the referring entity's content, the better the chances of the reference going stale. Content-addressing is an extreme example of this, though for your approach I guess one would have used some fuzzy matching (which is not applicable to content-addressing). > And this is the reason I am nervous over "gopher like features" in > DNS. Even though I think it is good...I think it will loose... I don't think that's a good enough reason here. There may be other reasons to tread carefully here. Nico --