Minutes IETF105: nmrg
||Minutes IETF105: nmrg
NMRG 54th meeting
IETF 105, Montreal
Audio recording: https://www.ietf.org/audio/ietf105//
Video recording: https://www.youtube.com/user/ietf/playlists
70 on-site participants + 14 participants on Meetecho
* Session 1
Thursday, 25 July 2019, Afternoon Session III 1740-1910
NMRG will hold an interim meeting co-located with IEEE NoF conference,
October 3-4 2019, Roma, Italy. The meeting will focus on
practical/implementation aspects of IBN/IBS.
2. Work from Arthur et al. presented an interactive and iterative intent
expression interface based on Nile (a restricted natural langugage)., and
represent an original attempt to capture and refine intent expressions and
their translation to NFV and SDN environments. The implementation is
3a. The work on intent classification is an important activity to provide
clarity in the domain of IBN. Several on and off line comments were
provided, to continue refining the content and validate the criteria for
3b. The work on intent concepts is also important and should be continued.
Several parts of the drafts will be further refined and complemented.
3c. The presentations on architecture considerations received multiple
comments, notably on "what" the intent-based system will interface with
(sout-bound). Additional options have been proposed by the participants and
further description/level of details is required.
3d. A first proposal on an intent-based managmement architcture has been
presented. the work is still preliminary and due to time constraints,
questions could not be addressed during the session and will be taken to
the mailing list
4. A quick summary of the RG activity on AI for/in NM has been provided by
Pedro (NMRG co-secretary).
1. Introduction + RG information, Chairs
17:40, 5 min.
Remarks: Call for demo for October interim meeting, monthly virtual
meeting doodle (https://doodle.com/poll/vd62wbwdw6zcmu66).
2. Refining Network Intents for Self-Driving Networks, Arthur Selle Jacobs
17:50, 10 min. + 5 min. Q&A
Nile adds an abstraction layer to allow intents to be specified in natural
language, using RNN and assessing the output by asking the assistante
before being deployed.
Diego Lopez: Q: SONATA emulator or full framework? A: Emulator. Q: Using
different backends? A: Just one.
3a. Update on Intent Classification, Will Liu
18:05, ~7 min.
Request to all: Please review/discuss this draft on the mailing list.
Laurent: this is an important work, should continue. Important to provide
clarity in this space and could be re-used by others (industry, research
and standards). Need to continue refining the content and validate the
criteria for classification.
3b. Update on IBN Concepts, lifecycle and functionality, Alex Clemm
18:15, ~7 min.
Will: Q: Some principles are not included in Lifecycle Figure 1. What are
we validating? Suggest to merge figures into one. A: It is a rough
collection of concepts, validation relates to rough aspects of the
architecture. Q: Position of some box in figures mismatch. Laurent:
Validation of Figure 1 (validating the correcteness of the intent
expression) is different from validation in Figure 2 (validating the
fullfillment of the intents). In
https://datatracker.ietf.org/doc/draft-sun-nmrg-intent-framework/ there is
also other validation function(s).
Benoit: main focus seen so far is on configuration. What about the
assurance part, consider operational data. not sufficient to have proper
configuration but also make sure when it's in operational state, it is the
expected one (fullfiling "properly the intent(s))
Expected that outcomes are from intent users.
Different level of intent validation. Move discussion to the mailing list.
Configuration vs provision data?
Domain knowledge and expertise is required.
3c. Considerations for intent-based management architecture(s), Laurent
18:25, 10 min.
Dean on EMS...
Management system must be defined first.
What kind of management system will IBN talk to?
Assumptions must be described. They are in the draft.
Consumption interface is complex.
Interface to devices or management system can be distinct for different
Brian: at least another option: to interface with an agent (cf. ANIMA)
Laurent: More intermediate layers are being defined (ETSI ZSM).
3d. An intent-driven management framework, Sun Qiong
18:35, 10 min.
Will to share his comments on the mailing list.
18:45, 15 min. Q&A
4. Summary of AI for Network Management activities, Pedro Martinez-Julia
incl. presentation on Intelligent Reasoning on External Events for Network
19:00, 7 min. + 3 min. Q&A
Diego: It is important to also ask for meta-data together with the datasets.
Why reasoning? Traceability is not so much relevant.