Re: [midcom] SNMPv3 as MIDCOM protocol: Opinions?

Wes Hardaker writes:
 > FYI, there is also a kerberos SNMP draft (2 actually) and an
 > implementation of it in the Net-SNMP toolkit.  The IESG are not
 > allowing these IDs (or any other security models) to be considered for
 > the standards track at this time, however.  There have been many other
 > discussions surrounding other session-based security models, but no
 > IDs have been written for them yet.

I resemeble one of those drafts :) Frankly, I
think the IESG ought to lay down a policy which
says that anything which defines application layer
security associations MUST specify a key
management protocol to key those security
associations. I don't think this is entirely
different from what Steve Bellovin wrote in his
-use-ipsec-00 draft (one of its cruxes was "if you
can use a previously defined key management
protocol, you are strongly recommended to do so."

The proof is in the pudding: the lack of SNMPv3
keying has lead to not one, but *two* different
key distribution mechanisms by Cablelabs alone,
not to mention Ken's Kerberos mechanism too.
$DEITY knows how many other -- probably poorly
thought out -- other ones lurk out there. As I
said, this is a major weakness of SNMPv3 when
being used as a general purpose frob as is being
proposed by folks in MIDCOM. I'd frankly recommend
that they not use v3 USM features at all and rely
on IPsec instead.

	 Mike