Minutes 2017 09 06

Meeting Materials	Attending
WebEx Recording password: 2EnxPFEJ	Igor Bryskin, Tianran Zhou, Alexander Clemm, Xufeng Liu, Eric Voit, Zhengguangying (Walker), Henk Birkholz, Andy Bierman

NETCONF Community interactions

Only introduce work to the NETCONF WG when the work is adequately formulated as a full draft. This removes some visible contention,
- The wider NETCONF audience will participate on first draft release
- WG chairs are informed of path, and may direct authors our way.

1. Multiple stream originators

Tianran put together proposal
Aimed at two Use Cases
- Distributed telemetry
- IoT Border Router
Some questions
- Mechanism must exist to partition and check which parts of subscription are served by which line cards
  - Key question: will we retain the assumption from earlier yang-push work of this being promise theory based. If yes, then internal mechanisms and feedback loops might be needed in the publisher to: (a) be able to discover loss across line cards.
    - Is this easier to handle just for periodic subscriptions? This could be hard (and un-necessary) to implement for on-change.
- Distribution to one or many interfaces on the collector?
  - Can we assume that clustering of the receiver might improve receiver scale without requiring addressing modifications on the publisher? (Hopefully yes.)
    - There are technologies available so that a single address for a receiver could be clustered to reduce that as a blocking point.
Support for other transports beyond UDP * Solution should support other transports like COAP, IPFIX * IPFIX has mapping questions which have been solved
- Perhaps transport issues should be a complimentary draft to the partitioning question (which itself is transport agnostic).
- COAP, COMI and others
  - Reliable COAP is already an IETF draft, look at that for info.
  - Look to existing sources to handle fragmentation & reassembly

2. Stateful / Smart Filters and Improvement in Triggers

Alex put together proposal, which he has emailed to everyone
Overall goal: Reduce the amount of data provided based on non-key object values.
This will be accomplished by allowing for more complex semantics than in YANG Push
- Non-key object value based filters
- Ability to go inside or outside ranges (stateful filters)
  - Clear condition, Onset condition, dampening period for fluxuations
Choice of encoding language past XPATH?
- Extensible expression Format for future capabilities
Service Assurance feels like the first opportunity
Not in scope, and addressed by "#3 Automation Framework"
- aggregation of multiple objects or calculations
- Triggered reports

3. Automation Framework

Igor put together proposal, already shared via email
Goal is to generalize yang-push's target-trigger-notify construct into event-condition-action construct
- Things which might be interesting:
  - combining multiple micro-conditions into a single macro-condition via a number of logical operations
  - combining multiple micro-actions into a single transaction with a possibility of specifying policies with respect to handling errors/exceptions of each of the transaction components
- Goal is to reduce processing overhead necessary to discover an element of interest. I.e., reduce the computational bootstrapping of discovering something of interest.
What is the differentiation from filters/selectors from previous work?
- Should the expression syntax be re-used from other subscription work (New and existing)?
- Or do we need a totally new syntax?
SUPA linkage?
In all cases, a very clear demarcation (technology, scope, other) must be made with the #2 above

4. UDP Binary transport

5. Yang notifications & headers + bundled messages

Eric to introduce an overall scope draft for the full set of drafts we eventually support
Leads from today frame five items above as single context paragraphs which can fit into section 2 of the overall scope draft.