I would strongly support what ScottJ has proposed, I would even say that
this is very pragmatic way to go forward. These new RRTYPEs make everything
quite clean (towards Bundle Protocol's integration as first class
citizen?). Even if not all DNS immediately support, most important thing is
what we can enable with it, not so much the transition period. This is
quite good baseline functionality without too much hassle on the DNS side.
For the backwards compatibility this is nothing new, should not be a real
concern.

On Thu, Jun 27, 2024 at 9:07 PM <sburleig...@gmail.com> wrote:

> Hi, Rick.  I would not fight to the death over this, but I am skeptical
> that having domain names map to node numbers encoded in CBOR, rather than
> to node numbers expressed as unsigned integers, would be advantageous.
>
> My understanding is that RFC 9171 and later
> https://datatracker.ietf.org/doc/draft-ietf-dtn-ipn-update/ define the
> CBOR encoding for BP ipn-scheme endpoint IDs.  Scott has indicated several
> times that he doesn't propose to return EIDs as resources; all he wants
> returned from a DNS lookup by domain name is the fully qualified node
> number (allocator number followed by node number) to which that name maps.
>
> Section 6.1.1 of the dtn-ipn-update draft explains that the encoding of an
> FQNN is a 64-bit unsigned integer and explains how that integer value is
> constructed.  One could CBOR-encode that value, which would tell the
> recipient "Here's an unsigned integer" and indicate the number of octets
> occupied by that integer value.  But it would seem simpler (both in the
> definition and for implementations) to state in the resource record
> definition that "The value returned is a 64-bit unsigned integer", so that
> the recipient instantly knows how to handle it.
>
> It is true that the CBOR representation could compress the returned value
> in many cases, but I would not expect that to be a significant advantage in
> DNS lookup traffic, which shouldn't be extremely heavy.
>
> Again, not a life-or-death consideration for me, but I would prefer the
> simpler mechanism.
>
> Scott
>
> -----Original Message-----
> From: Rick Taylor <r...@tropicalstormsoftware.com>
> Sent: Thursday, June 27, 2024 1:32 AM
> To: Scott Johnson <sc...@spacelypackets.com>; Mark Andrews <ma...@isc.org>
> Cc: Erik Kline <ek.i...@gmail.com>; dnsop <dnsop@ietf.org>;
> sburleig...@gmail.com; d...@ietf.org
> Subject: RE: [DNSOP] Re: [dtn] Re: Re: IPN and CLA RRTYPEs to support
> Bundle Protocol RFC9171
>
> Hi Both,
>
> Comments inline...
>
> > -----Original Message-----
> > From: Scott Johnson [mailto:sc...@spacelypackets.com]
> > Sent: 27 June 2024 04:36
> > To: Mark Andrews
> > Cc: Rick Taylor; Erik Kline; dnsop; sburleig...@gmail.com;
> > d...@ietf.org
> > Subject: Re: [DNSOP] Re: [dtn] Re: Re: IPN and CLA RRTYPEs to support
> > Bundle Protocol RFC9171
> >
> > Hi Mark,
> >
> > On Thu, 27 Jun 2024, Mark Andrews wrote:
> >
> > >> I broached the possibility of CBOR in discussion on DNSOP before
> > >> DTN was CCed, making the above point to Scott Burleigh.  Our
> > >> conclusion there, along with Mark Andrews, was that the current
> > >> verbiage is the current best course of action.  I have no personal
> > >> objection to wire format for the IPN RRTYPE being CBOR, if ScottB
> > >> and Mark agree that there is gain to be had over using 64-bit
> unsigned int.
> > >>
> > >> That said, it is unclear if appropriate CBOR functions/libraries
> > >> already exist/are used inside nameserver implementations. If not,
> > >> that could substantially delay deployment, and/or add burden to
> > >> implementers.  There is an active draft which specifically treats
> > >> CBOR encoding of RRs (
> > >> https://datatracker.ietf.org/doc/draft-lenders-dns-cbor section
> > >> 3.2.1), but that document is still an Individual Draft at this point
> as well.
> > >>
> > >> Mark, ScottB, opinions?
> > >
> > > What real benefit would CBOR bring over a raw 8 byte value other
> > > than saying it was entered via X or X.X?
> >
> > This would be my criteria exactly.  If we can save bytes, ok; lets
> > consider it.  If not, then why?
>
> My point wasn't about saving bytes (but of course one doesn't want to go
> overboard and explode the total length) - as Scott B pointed out, this
> isn't a resource constrained use-case.
>
> My point is about familiarity and commonality.  By the time someone has
> got their head around EIDs, NodeIds, Node Numbers, FQNNs, introducing yet
> another binary encoding to save some bytes seems a mistake.
>
> RFC9171 (the source document for all of BP) has a well-defined CBOR
> encoding, which is updated in a forwards-compatible way in
> https://datatracker.ietf.org/doc/draft-ietf-dtn-ipn-update/ (currently
> with IESG), so why not re-use it?  It has been intentionally designed such
> that if in the future NodeIds are expanded again (because 640K is never
> enough) the format is highly unlikely to need changing.  There is also
> following work within the WG adding 'wildcard' support, whilst still
> maintaining a consistent, efficient, self-descriptive format.
>
> If the RRTYPE is encoding a 'data type' defined by a IETF, I would expect
> there to be a good reason to deviate from the IETF standard binary encoding.
>
> Additionally I think it might make the RRTYPE defining document simpler:
> You can just normatively reference ipn-update.
>
> Cheers,
> Rick
>
> _______________________________________________
> dtn mailing list -- d...@ietf.org
> To unsubscribe send an email to dtn-le...@ietf.org
>
_______________________________________________
DNSOP mailing list -- dnsop@ietf.org
To unsubscribe send an email to dnsop-le...@ietf.org

Reply via email to