Re: Policy of IPv6 DNS Discovery

[Rob Austein <sra+dnsop@hactrn.net>]

At Sat, 02 Aug 2003 12:00:13 -0700, Bob Hinden wrote:
> 
> I think the IPv6 w.g. did answer this question several times.  It concluded 
> that it did want a solution in addition to DCHPv6.  For example from the 
> minutes of the Salt Lake City IETF in December 2001:
> 
>     Deering took poll of room: How many people think w.g. should continue
>     work on stateless DNS discovery? Consensus to continue work.

And there's at least part of the problem.  "Stateless DNS discovery"
is not the same thing as "a solution in addition to DHCPv6" unless
"stateless" is just a euphemism for "anything but DHCP".  DHCPv6-lite
is also "stateless" for every definition of "stateless" that applies
to, eg, the RA-based proposals.  This confusion persisted through the
entire discussion in the IPv6 WG, and at least some of the arguments
in favor of RA-based or ND-based solutions on this list still appear
to originate in this confusion.

> and again in the minutes of the Yokohama Japan Thursday, July 2002:

Same problem.

> I would also note that at the Vienna meeting, a "hum" was taken in the 
> DNSOP group and there was not a consensus that there should only be a 
> DHCPv6 solution to this problem.

Correct.  Please see the message to which I was responding, which
called for an immediate decision in favor of a non-DHCP-based
solution.  I do not think that the discussion is over yet, which is
why I do not think an immediate decision is appropriate.

> I conclude from this that there is a significant number of people
> who think that a solution is needed in addition to DHCPv6.

And a significant number of people are still waiting to hear reasons
more substantial than "I don't like DHCP" (or, even worse, "I don't
understand DHCP but I know that I don't like it").

If there are serious -technical- reasons for why DHCP-lite is a bad
solution, fine, put 'em on the table.

Real issues I've heard so far:

1) Pekka points out that the DHCPv6 spec is a pretty big chunk of text
   for an implementor to have to read just to do a little discovery
   frob.  The DHC WG is already aware of this problem and is already
   working on a doc to help: see draft-ietf-dhc-dhcpv6-stateless-00.
   So I think this issue is under control.

2) Alain points out that having every node in a building poll at once
   when the building's power comes on can ruin one's afternoon.   I'm
   not as concerned about this as Alain is, but Alain has already come
   up with a proposal for an optional extension that will address his
   concern.  So I think that this issue is also under control.

3) Ohta-san has resurfaced the well-known-anycast hack.  Issues with
   this are already on record in the IPv6 WG list archives, so I won't
   restate them here, but I will note that sites which chose to do
   this with unicast addresses out of their own address space (as
   opposed to well-known unicast or anycast addresses) are already
   free to do so, and, indeed, I've been told that many ISPs already
   do this for v4 -- pretty much by definition, nobody but the ops
   staff at those ISPs would know.  At a high level, though, the key
   point here is that this is not new technology, and sites which
   chose to use it can do so already without further protocol work.

Other than the above, what I've heard has been the same tired refrain
of "stateful address assignment is optional in IPv6, therefore we need
a DNS discovery solution which doesn't use DHCP", which just tells me
that the people singing this song still don't understand how the DHCP
Information-Request mechanism works.  See (1), above.

In summary, while it's possible that there's a credible case to be
made for why further protocol work is needed, I haven't seen it yet.
#----------------------------------------------------------------------
# To unsubscribe, send a message to <dnsop-request@cafax.se>.