NMRG Virtual Meeting July 2019 Monday 01-07-2019 11:00-13-00 CEST Participants (11): Laurent Ciavaglia (co-chair, Nokia), Jérôme François (co-chair, INRIA), Pedro Martinez-Julia (NMRG secretary, NICT), Abdelkader Lahmadi (INRIA), Will Liuscheng (Huawei), Minsuk Kim (ETRI), Walter Cerroni (University of Bologna), Marinos Charalambides (UCL), Dhruv Dhody (Huawei), Vishnu Ram (Technical Consultant, ITU-T ML5G FG), Mehdi Bezahad (Lancaster University) Useful links: * Agenda: https://datatracker.ietf.org/doc/agenda-interim-2019-nmrg-08-nmrg-01/ * Materials: https://datatracker.ietf.org/meeting/interim-2019-nmrg-08/session/nmrg * Webex: https://ietf.webex.com/ietf/j.php?MTID=mf13f05555e4d7f0502d187f3804c9148 * Etherpad: https://etherpad.tools.ietf.org/p/nmrg-virtual-20190701 Agenda: - 10 min Status of WG and next meetings, Laurent & Jérôme - 20 min Intent Based Networking, Laurent & Jérôme + Status progress on Intent Based Networking + Feedback from San Jose meeting ++ Minutes will be sent soon ++ Potential use cases: Data-Center Networks and interconnect, augmented with Enteprise networks, 5G verticals (different 5G use cases on the same infra, 5G slicing), more "common" digital/communication services provider use case ++ WIP: Propose a basic use case template (how to describe and exploit a use case) ++ continue work on the terminology Marinos: architectural aspects discussed ? Laurent: not really. People were not deeply involved in IBN, so we stay at a higher level. Marino: how to move this work forward? Laurent: yes, refinement of IBN workplan, open call for contributions to follow (no existing document) + Intent lifecycle and operating principles (outcome of the discussion of Laurent and Alex Clemm) ++ different from an architecture ++ objective: agree on a lifecycle on IBN (less controversial than an architecture), faster confergence time ++ Initial lefecycle diagram: will be updated in next days Abdelkader: what is the level of automation? seems to be very manual, what is the benefit of using intent in that case? Laurent: once the intent is provided, then everything should be automated or intent can be extracted from the user-related data. Iterative refinement of the intent with the user (work from Lisandro and Arthur). In the middle of this diagram, there is also the operator to define how the system has to catch the intents. Automation will also depend on the capabilities of the system. On the left side, "user" inputs are manual, everything else on the right side is/could be automatic. Abdelkader: so the user is not necessary a human, can be an application? Laurent: here the user is more a human user. So we may extend this diagram. user may also be an interface or system a human user is using to interact with the IB system (e.g. a vertical industry and 5G network). Abdelkader: regarding the work of Arthur, what could be good is to would have a set of labels to recognize elements in the intent expressions. Laurent: an interesting area to explore. The diagram is incomplete, need to identify inputs/ouputs and also supporting elements such as info and data models. In IBN, there is usually a restricted language to express intents but in NMRG how can we be more flexible with dynamic environment, dynamic mapping of ontologies and system capabilities. Walter: need some sort of standardized interface between user space and system space, and even at the lower level when applying intents (preparation and operation phase). Do you think there could be also standardized interfaces between functions ? Laurent: no immediate answer, part of our investigation to define what level of standardization we would need but the less the better to keep flexibility. If we document external intertfaces, it will be a good achivement of the group. The goal is to minimize the standardization needed. Jérôme: more backward paths to the intermediate functions Laurent: agree, need to improve this, not only through report. Avoid to be too much complete/detailed in first shot, in order to avoid entering in architectural design, keep as high level bastraction as possible for the lifecycle but still being useful in guiding work. Marinos: operator should express policies (at the right) and need also to be decomposed. there could be operators intents. Laurent: agree. consider them as orthogonal dimensions: service intent and operation intent. not coming from the same actor, nor targeting the same goals. Mehdi: is this presentation recorded? Is the user space only humans? Laurent: I don't think there is a video recording but materials is available on the nmrg webpage. For second question, this is WIP and diagram incomplete. The user is a human user (in the diagram). The user space is where intent are expressed. Mehdi: I'm involved in a UK project with BT. Who can ask the intents: provider, users.. ? Laurent: question fully valid, work regarding use case is in progress (actors, roles)... Medhi: who is "we" Laurent: Laurent from Nokia, acting as NMRG co-chair and contributor , what I am presetning here is as a contributor. + Montreal meeting call for contributions: call for presentations for the regular session (topic: IBN) and technical meeting (topic also IBN, more deep discussion expected). - Network AI + 15 min: challenge, Jérôme François Laurent: should keep/develop link with IETF (but not only): expertise, knowledge about protocol issues (operational), presentations on measurement and analysis. WG/RG. bidirectional (bring AI/knowledge/novelty)(and beefit from feedback, expertise) Will: take the use cases from (?) Jérôme: identify existing use cses, document new ones. define the criteria. + 20 min: Intelligent Management for Edge-computing, Minsuk kim Laurent: we did not give you enough time (sorry), we'll see what we can do as next steps for these work in NMRG. Need to discuss with Jérôme. + 20 min: Intelligent Reasoning on External Events for Network Management: Positioning, Challenges, and Framework, Pedro Martinez-Julia Will : + 20 min: Self-driving networks, Abdelkader Lahmadi Jérôme: Self-driving networks always suppose programmability? Abdelkader: yes but of course depends on the level of automation Mehdi: if we tweak legacy networks, can we still need programmability Abdelkader: hard task to tweak Mehdi: do we mean a clean-slate approach or do you think that current network with this enabling technology is sufficent Abdelkader: prepare for self-driving networks (clean-slate approach), for current approach we can have less automation xxmin. - Any other business?