message to nanog

[Eliot Lear <lear@cisco.com>]

At the plenary, we agreed that I would draft a message to NANOG to ask 
for feedback on features such as notifications & channels.  Here is my 
first cut at such a message.  Let me propose that if we don't get enough 
responses, we actually go to NANOG and do a presentation.

Subject: YOUR opinions needed for Network Configuration Protocols
Reply-to: netconf@ops.ietf.org


Dear NANOG folk,

The NETCONF working group of the IETF is currently developing a 
collection of protocol specifications for the configuration of network 
elements.  This work originated from a roadshow that many of us went on 
to learn what operators of different types want in such a protocol.  Now 
we would like to checkpoint with you on some of the contents of those 
specifications.

The protocol itself is split into two parts: an abstract set of 
functions, and a binding to specific protocols, including SSH, BEEP, and 
SOAP over HTTP(s).  Each protocol has its pluses and minuses.

As envisioned, the base protocol supports an option for notifications. 
The idea is that a manager would be notified of configuration-related 
events, such as a card insertion or removal, and act appropriately to 
configure the element.  The envisioned format of notifications is either 
reliable syslog from RFC 3195 or something similar.  Because 
notifications are asynchronous, one writes code that implements a 
dispatch mechanism that discriminates on the type of event. 
Notifications would be an option that not all managers would have to 
implement.

The working group is attempting to determine whether notifications 
should remain as part of the base specification.  Here are the choices 
facing the group:

Option A.  Leave them in as currently specified, and require all 
protocol mappings to support them.
Option B.  Allow them to be asynchronous, but don't use RFC 3195, and 
require all mappings to support them.
Option C.  Remove them entirely from the specification and let vendors 
implement RFC 3195 or other notification mechanisms as they see fit (for 
instance, existing syslog).

Do you have an opinion on which of these options you would like?

Related, the NETCONF base protocol currently makes use of the notion of 
channels.  Channels are a basic concept in the BEEP protocol, and they 
exist in SSH as well.  However, use of multiple channels in SSH is not 
supported in common SSH applications.  They are completely absent in 
HTTP, and so the notion of a session would have to be introduced in the 
mapping.

Channels provide a number of benefits.  First, they divorce the idea of 
who is the session initiator and who is the device manager.  This means 
that either the network element or the manager could initiate a TCP 
connection.  This is useful when NATs and firewalls exist between a 
device and the manager.  Channels are also necessary to implement a way 
to determine progress of a configuration operation.

The working group has three choices:

Option A.  Keep channels in the base document and require each mapping 
to support them.
Option B.  Make channels optional.
Option C.  Remove channels from the base protocol and allow their use
           in the protocol bindings.

Which option do you prefer?

If you would like to read the entire set of documents, you will find 
them by going to the NETCONF working group charter page:
http://www.ietf.org/html.charters/netconf-charter.html

Eliot

--
to unsubscribe send a message to netconf-request@ops.ietf.org with
the word 'unsubscribe' in a single line as the message text body.
archive: <http://ops.ietf.org/lists/netconf/>