At 12:26 AM +0200 9/7/05, Harald Tveit Alvestrand wrote:
I believe that the ISMS WG's proposal is about ADDING the
possibility of SNMP over TCP, not about CHANGING SNMP to use TCP.
UDP will still work.
That is correct. UDP and the current SNMPv3 USM security mechanisms
will still work. They will also remain mandatory parts of SNMPv3.
And I believe Eliot's concern is about letting the TCP session that
carries the SNMP PDUs be opened from the agent to the manager,
rather than from the manager to the agent (yes I know - this is
SNMPv1 terminology, but I've forgotten the SNMPv3 terminology); that
is another feature that comes in addition to what the group is
apparently currently working on.
And just BTW: I find "call home" reasonable to specify too, once
you've done TCP. It's obvious enough that I think it will be added
to implementations whether or not we specify it, so we should have
very strong reasons not to do so.
I don't even believe you need to "turn" the session, since SNMPv3
doesn't recognize the concept of a "direction" for a session....
just let the PDUs flow....
Unless I am seriously misunderstanding something, this is a bit more
complicated than you and Eliot seem to think that it is... The
command responder (agent) is a stateless piece of software that
simply responds to queries as they are received. It has no way to
anticipate when queries will be received, and no concept of what
other systems it would like to receive queries from. So, where would
it get the information necessary to open a connection to the manager?
How would it know what to do if the manager could not be reached?
How would it know when he connection should be taken down?
Now, I am not saying that I couldn't come up with an answer to these
questions -- it seems likely that we'd have to grow another SNMP MIB
that would control how/when/if the command responder would attempt to
establish a communication connection with the command generator, etc.
I agree that SNMP over TCP might be an important element of this
solution, but it is already defined in RFC 3430. So, if there really
is sufficient interest in adding call home capability to SNMP, I
don't see why the IETF couldn't do this work.
But, why does anyone think that we should do this work in the
Security area in a WG that is tasked with integrating the SNMP
security model with SSH and RADIUS?
Margaret
_______________________________________________
Ietf@xxxxxxxx
https://www1.ietf.org/mailman/listinfo/ietf