Hi, Craig,
On Sep 13, 2011, at 5:58 AM, "Everhart, Craig"
<Craig(_dot_)Everhart(_at_)netapp(_dot_)com> wrote:
Hi,
Thanks, Joe, for the thoughtful review.
Extracting some light from yesterday's discussion and RFCs 6335 and
2782, much of the contention seems to be around whether there could be
two assigned names for related services that both happen to be provided
over a given assigned port number.
I'd like to propose an adjustment of the request, in light of the RFC
text, and ask your judgment on how that adjusted request would fit
within the services rubric that's outlined.
In RFC 6335, explicit room is carved out for "assigned service name[s]
without [a] corresponding fixed port number" with explicit reference to
RFC 2782. Such a service name assignment would be completely adequate
for purposes of the "NFS domain root" concept. As I read RFC 6335, such
assignments are RECOMMENDED, and that IANA strives to assign such names
under a first-come first-served policy (with reference to RFC 5226).
In this formulation, the existing "nfs" service could stand unaffected.
The proposed service, "nfsdomainroot", would request name assignment
without a port number. The NFS client would seek SRV records under
names such as "_nfsdomainroot._tcp.example.com". The result processing
would continue as before, with SRV records indicating the hosts
providing NFS service for the root of "example.com"'s file system.
The SRV record then includes the port number on which the target NFS
hosts provide "nfsdomainroot" service, which is given as an application
of the NFS protocol in the subject I-D (including a mount point).
The I-D would also need to propose, in the IANA section, that IANA
assign the service name
SRV nfsdomainroot TCP
to this protocol. UDP service name allocation is unnecessary.
In the judgment of the TSVDIR, would this take the I-D in a more
acceptable direction?
There is one caveat: if you indicate the current NFS port in this SRV, and then
validate it somehow, you should be OK with "nsf" as the response.
I.e., port numbers have exactly one canonical name, and no aliases. This should
be made clear in the doc.
Otherwise, this seems viable AFAICT.
Joe
_______________________________________________
Ietf mailing list
Ietf(_at_)ietf(_dot_)org
https://www.ietf.org/mailman/listinfo/ietf