Re: [dnsoverhttp] Post-Seoul thoughts

[Patrick McManus <[e-mail]>]

hiya

On Tue, Nov 22, 2016 at 9:59 PM, Shane Kerr <[e-mail]>
wrote:

> Patrick,
>
> At 2016-11-22 14:35:53 -0500
> Patrick McManus <[e-mail]> wrote:
>
> > On Tue, Nov 22, 2016 at 1:12 PM, Paul Hoffman <[e-mail]>
> > wrote:
> >
> > > t someone mentioned was that HTTP servers might not want to use
> > > DNS-over-HTTP to push DNS data into clients but instead
> >
> >
> >
> > that doesn't remove the use case. link tags add N round trips, and
> removing
> > that latency literally is the use case. Presumably the server has cached
> > copies of the data it needs to staple.
> >
> > I think we still have all 4 use cases.
>
> Well, the lookups can be done in parallel so N round trips needn't take
> much longer than 1 round trip. Still 0 trips is of course better than 1
> trip, everything else being equal. ;)
>
>
one lookup can take multiple rtts.. one for (each) ns, one for the a/aaaa..
etc. obviously highly dependent on state. N seemed an easier way to put it,
but you're right that the focus is on > 0.


> I'm not sure what "to staple" means in this context. (Actually I only
> know about stapling physical things, like paper in an office or sheets
> of plastic in construction.) Can you explain or give a link? :)
>
>
just to inline into a response content that would have normally been
fetched by the client after receiving the response. OCSP Stapling is a
great example from TLS. https://en.wikipedia.org/wiki/OCSP_stapling




> Cheers,
>
> --
> Shane
>