Re: [tae] New draft: announcing the supported transports via DNS

[Caitlin Bestler <[e-mail]>]

On Sep 18, 2009, at 8:11 AM, Michael Welzl wrote:

> Hi all,
>
> This discussion reminds me of something else:
> someone (I think Jana?) mentioned the possibility of negotiating
> more than just the transport protocol, e.g. even the usage of IPv6,
> with a negotiation protocol.
>
> I recently talked about this with someone who knows more about
> IPv6 than me (actually not hard to find such a person!), and that
> someone said that a standard is already in place for determining
> whether IPv6 can be used **via DNS**.
>
> So, whether or not the approach discussed here is adopted, for
> the choice of IPv6 at least, DNS seems to be the way to go...
> right? Opinions?
>
> Cheers,
> Michael
>
>

I agree that we definitely should be making an attempt to be  
consistent with prior IETF decisions.


Choosing IPv6 vs IPV4 via DNS is a more natural fit than choosing  
between SCTP and TCP
because it is indeed a *host* that supports IPV6. Since most  
applications do not explicitly
bind their local address, or even examine the address of the remote  
peer, they will just use
IPV6 automatically if the host stack provides it. It is not a per- 
service decision.

The other precedent to examine is the determination of RPC protocol  
for NFS and other RPC
services. This was not done via DNS. I always found the resulting  
procedures awkward, but
that is something that there should be a group consensus on. If the  
final solution does not
follow the RPC strategy, it should explain the rationale for not doing  
so, and cite the experience
of people who have been working with RPC far more intensely than I have.





--------
Caitlin Bestler
[e-mail]
http://www.asomi.com/CaitlinBestlerResume.html