Re: WG Review: Recharter of Hypertext Transfer Protocol Bis (httpbis)

--On Friday, February 24, 2012 17:57 +0100 Patrik Fältström
<patrik@frobbit.se> wrote:

> 
> 
> On 24 feb 2012, at 17:43, John C Klensin <john-ietf@jck.com>
> wrote:
> 
>> It is
>> the number of folks who, for lots of reasons, haven't upgraded
>> from operating systems, resolvers, etc., that don't support
>> newer RRTYPES.
> 
> As I said, people disagree... ;-)

> As far as I know, there is nothing in any of the operating
> systems you mention that prohibits an application to send a
> random udp packet, and because of that your application can
> include a resolver library.
> 
> What is a problem are the cases where DNS is not used at all
> at the end node, but instead other name binding/lookup
> protocols combined with a firewall policy that because of this
> can and is blocking udp+tcp/53 in various ways.

I'd suggest that there are two other problems.  One is that
per-application resolver setups pretty much prevent caching of
any flavor (possibly not an issue if one opens applications,
keeps them open for a long time, and uses different target sites
with different applications, but, if that scenario has been
studied wrt frequency, I'm not aware of it).  The other, more
important, issue is that it just about guarantees an
inconsistent user experience wrt the treatment of names.  

Of course those are tradeoffs against locally-improved
functionality and reasonable people can disagree about how
important those issues are wrt the other considerations.


> That said, I still ask when it is, in general, time to just
> move forward. I see for example many other reasons why people
> should not use that old software. IE6 for example. Yes,
> economically constrained situations exists, but that problem
> do not go away by having us not start using SRV or HTTP/1.1 or
> SNI or HTML5.0 or...pick your favourite. And with SPF, that is
> not used by the edge node either.
> 
> I am asking more generally why specifically this DNS issue is
> so stuck, because I think that is unfair. We upgrade other
> protocols...

Where I probably agree with you is that I think that we need to
evaluate costs, benefits, and risks and to do so against an
understanding and hope that we will have _many_ more Internet
users a decade from now than we do today.  Accepting the latter
may justify changes even more painful than transition to a new
RRTYPE if we understand that we are inconveniencing a relatively
small number of people today in order to make things much better
for a far larger future number.  And that is why I have never
believed in arguments for guaranteed absolute forward
compatibility in Internet, and Internet-like, situations.

> But my point is, people disagree. As we see here ;-)

Indeed.  Even though I would hope that we can at least mostly
agree about the facts even if we disagree about the tradeoffs to
which they lead.

    john

Received on Monday, 27 February 2012 19:41:51 UTC