NMRG 60th meeting IETF 110, Online Monday, March 08, 2021, 16:00-18:00 (UTC), Room 5 RG Chairs: * Laurent Ciavaglia * Jéröme François RG Secretaries * Jéferson Campos Nobre * Pedro Martinez-Julia Useful links: Materials: https://datatracker.ietf.org/meeting/110/session/nmrg Video recording: https://youtu.be/4SyQA_kdlcU Participants: Notes: 1) Introduction, RG Chairs, 5 min. 2) Status on active Internet Drafts (35 min.) - Concepts of Digital Twin Network, Cheng Zhou, 10 min. + Q&A https://datatracker.ietf.org/doc/draft-zhou-nmrg-digitaltwin-network-concepts/ Q&A: --> questions and comments to the mailing list Laurent (Chat): do you intend to describe telco/network specific research questions? or survey technology enablers? What are the goals of the draft in the context of NMRG? Diego: The plan is to consolidate the framework and identify challenges and (some) solutions Philip (chat): Applying the concepts from existing uses in networking, seems a step up in complexity and scale? Any ideas how to overcome those challenges? are there promising/recommended approaches? Is that in scope of the document? Diego (chat): The idea is to apply the same principles used by current DTs in industrial environments: apply AI models to build a synthetic environment that would support management decisions. Approaches are being considered and, in fact, we have a proto-DTN in our lab we call the “Mouseworld” that we are currently using to generate datasets for ML training and verification Philip (chat): Would be interesting to hear about those approaches, how well they work, compromises etc. I think that would be the really interesting part of the document! Laurent (chat): see DTN has having both engineering and research challenges. In the context of NMRG (IRTF), it would be good to lay down these two sets of questions/challenges. Qin (chat): Our suggestion and approach is to move from simulation to emulation, provide emulation in real time which is built on top of real time telemetry technology. With digital twin platform, we can integrate continuous verification, test what if senarios. Laurent (chat): DTN topic: do you want to have a dedicated NMRG interim to digital twin? (or we could aim to organize this as part of a conference, e.g. IM 2021) - Intent Classification (in RG last call), Olga Havel 5 min. + Q&A https://datatracker.ietf.org/doc/draft-irtf-nmrg-ibn-intent-classification/ Q&A: Philip (chat): I didn’t understand what the different colours and symbol (tick and cross) meant, please could you explain Benoît (chat): I will review the draft, for sure. One observation already. The draft would benefit from using the intent definition from the concepts and definitions draft and not its own definition. Note: this is why I reviewed the concepts and definitions draft first. - Network Measurement Intent, Danyang Chen 5 min. + Q&A https://datatracker.ietf.org/doc/draft-yang-nmrg-network-measurement-intent/ 2) Technical talks on Intent-based Networking (50 min.) - Automated Performance Evaluation of Intent-based Virtual Network Systems, Kazuki Tanabe, 15 min. + Q&A http://dl.ifip.org/db/conf/cnsm/cnsm2020/1570659331.pdf - SDN Heading North: Towards a Declarative Intent-based Northbound Interface, Shiyam Alalmaei, 10 min. + Q&A http://dl.ifip.org/db/conf/cnsm/cnsm2020/1570662931.pdf Laurent: lots of state of the art on e.g. ontologies and semantic graphs, etc. to my limited knowledge, there are dynamic approaches (loading /learning concepts/terms from other domains) Shiyam: Our current state of intent/policy mapping, it’s still static and predefined. However, we are aiming to integrate the Non-functional Requirements (NFR) Framework which helps to evaluate different possible mapping alternatives to microservices based on the evaluation criteria (i.e., NFRs). This could help with the automated mapping between intents and their corresponding microservices based on the evaluation criteria. - An Intent Demonstration in an Adaptive Policy Environment, Joseph McNamara, 15 min. + Q&A http://dl.ifip.org/db/conf/cnsm/cnsm2020/1570679866.pdf 3) Technical talks on AI for Network Management and Operation (30 min.) https://docs.google.com/document/d/1dQOzZustI2mkYr_omtiqu3FqUvoqLgaCp7nbRj4ZJyw/edit - Self-Driving Network and Service Coordination Using Deep Reinforcement Learning, Stefan Schneider, 15 min. + Q&A http://dl.ifip.org/db/conf/cnsm/cnsm2020/1570659307.pdf Stuart Card: Causal reasoning? Stefan: not yet investigated Jerome: any good method to define the reward function rather than just test/try Stefan: not really, no cookbook for reward function. Problems and Strategies implementing in-network AI models, Matthews Jose