Hi,

On Tue, May 22, 2012 at 09:09:39PM +0200, Antonio Jara wrote:

> Our approach is mainly focused on the optimizations of the replies included
> in the packet, e.g. only include at the beginning the PTR record instead of
> the PTR + SRV + AAAA.

I had a look at the PDF you enclosed.

First, I can't tell whether you're actually using DNS and mDNS or
not.  Part of V of the paper suggests that you're using a DNS-like
packet format over CoAP, in which case you're not actually using
either of them; instead, you're using the same payload format but a
different protocol, since DNS uses UDP and TCP.

Reading your paper, however, it seems to me you have deep
misunderstandings of the DNS, the way it works, and its relationship
to mDNS.  To be clear: mDNS to date is almost exactly the same thing
as the DNS, only responding to multicast addresses in a
multicast-friendly way and on a different port.  Apart from that, all
the RRTYPEs and everything else about the DNS are the same.  Your
suggestion in (V) that mDNS or DNS-SD add resource record types to DNS
is mistaken.  (There is a problem in that there's no guarantee that
things will stay this way.  That's a problem for another community,
however.)

If you are actually using DNS, then TXT in the way you are using it is
a serious mistake.  There are _no_ semantics for TXT records, and there
is no way you can be sure that someone else won't put a TXT record of
the same name in the DNS.  Importantly, TXT records _are not_
"designed to be associated with the SRV entry", there is no mechanism
for making that association except your client's interpretation, and
there is no way whatever even to be sure that you'll see a consistent
view of the target zone, since there might be caches in the way.

What seems to be a more serious problem is your discussion in (VI)
about the size of responses.  In general, there is no way to control
this from the client side, and if you are actually using DNS you are
going to have to be able to cope with these additional parts of the
message.  In some cases, anyway, you won't be able to do anything
without them, for the Additional section is where glue records
(necessary to bootstrap the entire DNS chain) go. 

I get the impression that you _want_ to use DNS because it gives you
an already-deployed Internet-wide infrastructure, but mDNS doesn't
work outside the LAN, and DNS requires some mechanism of updating a
central database (the mechanism for which is not clear in your plan).
I am also having a hard time understanding the extent to which DNS is
actually a good match for the problem you're trying to solve, since
most of your goal seems to be to provide information about a node
(hence all the TXT records).  I don't see how the DNS is a good place
to store this, and it is poorly adapted for transporting the data.
And, of course, if you're using CoAP you're not using DNS at all, in
which case I don't see why you'd want to pay the cost of the poor
adaptation of DNS RRTYPEs for the problems you seem to have.  Why not
just put a name and an address in the DNS, as you suggest in (V), and
ask that node what its state is?  Why cart all that stuff into the
DNS, and then have to hack around the way the DNS makes your life
hard?
 
> It is also considering the options of include some special query to ask
> specify TXT tuples, instead of general TXT entries, since the idea is to
> describe the devices in a set of TXT records etc...

I don't think I understand this sentence, but if what you're hoping to
do is to cram additional sub-RR queries into the DNS query format,
then there's a working way to do this: get some new RRTYPEs.  TXT is
not what you want if you want structured data.  If you really must,
you could presumably do it with an underscore convention, although I
don't see a reliable way to make most of this tractable given that you
don't know what kind of device someone might be trying to address.

This is the second time in this WG I've seen attempts to use the DNS
to achieve something, and I still don't know what that something is.
Apart from getting the address of a named object -- a task at which
the DNS excels -- I'm not sure I understand how the DNS is adapted to
the task.

Best regards,

A

-- 
Andrew Sullivan
[email protected]
_______________________________________________
Lwip mailing list
[email protected]
https://www.ietf.org/mailman/listinfo/lwip

Reply via email to