**NMRG 72nd meeting**
**IETF 118, Prague + Online**
* Friday 2023-11-10 13:00-15:00 CET
* https://datatracker.ietf.org/meeting/118/session/31741.ics
* * *
Contacts:
* RG Chairs
* Laurent Ciavaglia
[laurent.ciavaglia@nokia.com](mailto:laurent.ciavaglia@nokia.com)
* Jérôme François
[jerome.francois@uni.lu](mailto:jerome.francois@uni.lu)
* RG Secretaries
* Jéferson Campos Nobre
[jcnobre@inf.ufrgs.br](mailto:jcnobre@inf.ufrgs.br)
* Pedro Martinez-Julia
[pedromj@gmail.com](mailto:pedromj@gmail.com)
* * *
**Useful links:**
* Materials: https://datatracker.ietf.org/meeting/118/session/nmrg
* Meetecho: https://meetings.conf.meetecho.com/ietf118/?group=nmrg
* Notes: https://notes.ietf.org/notes-ietf-118-nmrg
* ## Video recording: https://www.youtube.com/user/ietf/playlists (available post-meeting) {#video-recording-httpswwwyoutubecomuserietfplaylists-available-post-meeting}
**Agenda**:
* 13:00 Introduction, RG Chairs, 05 min
***NMRG Research Agenda topics***
* 13:05 **Knowledge Graphs for Network Management, Ignacio
Dominguez-Martinez**, 15 min
Abstract: The advent of AI opens the door for new opportunities that
represent a step closer to the promised land of autonomous networks.
However, AI applications are expected to consume combinations of
data from multiple heterogenous sources, scattered throughout the
network. In this sense, Knowledge graphs have appeared as promising
technology due to their data integration and semantic
interoperability capabilities. This presentation highlights some of
the key aspects and challenges when building knowledge graphs within
the scope of network management.
* 13:20 **Semantic Metadata Annotation for Network Anomaly Detection,
Thomas Graf**, 15 min
https://datatracker.ietf.org/doc/html/draft-netana-opsawg-nmrg-network-anomaly-semantics
https://wiki.ietf.org/en/meeting/118/hackathon, example
implementation (IETF Hackathon)
* Alfons: Is this architecture with all the level of controllers
suitable for ML applications?
* Thomas: the semantic for control plane is well defined, main
challenge is for the management plane
* Roland: annotated data will be super helfpul for academia
* Laurent: have you consider approaches where you inject faults or
stressing events? If so, how to make sure that the process is
helpful for operational purposes?
* Thomas: We have been strating with a kind of lab environment but
there is a difference between lab and real network
* Qin: Work related: ECA draft (how to respond to network changes)
* 13:35 **Artificial Intelligence Framework for Network Management,
Pedro Martinez-Julia**, 15 min
Abstract: The adoption of artificial intelligence (AI) in network
management (NM) solutions is the way to resolve many of the complex
management problems arising from the adoption of NFV and SDN
technologies. The AINEMA framework, as discussed in this document,
includes the functions, capabilities, and components that MUST be
provided by AI modules and models to be successfully applied to NM.
This is enhanced by the consideration of seamless integration of
different services, including the ability of having multiple AI
models working in parallel, as well as the ability of complex
reasoning and event processing. In addition, disparate sources of
information are put together without increasing complexity, through
the definition of a control and management service bus. It allows,
for instance, to involve external events in NM operations. Using all
available sources of information --- namely, intelligence sources
--- allows NM solutions to apply proper intelligence processes that
provide explainable results instead of simple AI-based guesses. Such
processes are highly based in reasoning and formal and target-based
intelligence analysis and decision --- providing evidence-based
conclusions and proofs for the decisions --- in the form of AI
method outputs with explanations. This will allow computer and
network system infrastructures --- and user demands --- to grow in
complexity while keeping dependability.
https://datatracker.ietf.org/doc/html/draft-pedro-nmrg-ai-framework
* Alfons: the control and management service bus looks very similar to
SMO architecture in ORAN.
* Pedro: we did not look at ORAN but we are looking more at a highler
level view. Use kafka.
*
* 13:50 **AI-Based Distributed Processing Automation in Digital Twin
Network ;
Considerations of deploying AI services in a distributed method,
Yong-Geun Hong**, 15min
https://datatracker.ietf.org/doc/draft-oh-nmrg-ai-adp/
https://datatracker.ietf.org/doc/html/draft-hong-nmrg-ai-deploy
* 14:05 **Performance-Oriented Digital Twins for Packet and Optical
Networks, Albert Cabellos**, 15 min
https://datatracker.ietf.org/doc/draft-paillisse-nmrg-performance-digital-twin/
* Roland: your definition is more generic
* Roland: events simulaotr can be used on different
abstractions/layers and it can works better than discrete event
simulator
* Albert: simulations are very slows and there is case where is no
simultion, for example for optical networks
* Roland: SIGCOMM papers on data-driven model of networks
* Marco Liebsch
14:20
To Albert: Agree that NTD should answer the question ‘how changes
the behaviour of my network if THAT happens?” One point about THAT:
Do you think the enforced change or event (THAT) is more on the load
on a path, or more in the configuration of a node in a NDT, e.g. its
queue size, scheduler, etc. The latter may require creating a new
model, while the impact of adding load to one interface/path is what
an existing well trained model can help to answer. Do you agree?
* 14:20 **Intent-Based Network Management Automation in 5G Networks,
Jaehoon Paul Jeong**, 15 min
Abstract: This document describes Network Management Automation
(NMA) of cellular network services in 5G networks. For NMA, it
proposes a framework empowered with Intent-Based Networking (IBN).
The NMA in this document deals with a closed-loop network control,
network policy translation, and network management audit. To support
these three features in NMA, it specifies an architectural framework
with system components and interfaces. Also, this framework can
support the use cases of NMA in 5G networks such as the data
aggregation of Internet of Things (IoT) devices, network slicing,
and the Quality of Service (QoS) in Vehicle-to-Everything (V2X).
https://datatracker.ietf.org/doc/html/draft-jeong-nmrg-ibn-network-management-automation-02
https://wiki.ietf.org/en/meeting/118/hackathon, IETF-118 Hackathon
Project related to this draft
* Laurent: we have sevral IBN use cases. Discuss with other authors of
other IBN use cases, commonolaties, connections (for example
measurment intent)
* Alex: interesting to have this as a use case and also an experience
report.
* 14:35 **Challenges and Opportunities in Green Networking, Alex
Clemm**, 15 min
***Open forum***
* 14:50 **KIRA: Scalable Zero-Touch Routing for Autonomous Control
Planes, Roland Bless**, 15 min
Abstract: KIRA is a scalable zero-touch distributed routing solution
that is tailored to control planes, i.e., in contrast to commonly
used routing protocols like OSPF, ISIS, BGP etc., it prioritizes
resilient connectivity over route efficiency. It scales to 100,000s
of nodes in a single network, it uses ID-based addresses, is
zero-touch (i.e., it requires no configuration for and after
deployment) and is able to work well in various network topologies.
Moreover, it offers a flexible memory/stretch trade-off per node,
shows fast recovery from link or node failures, and is loop-free,
even during convergence. Additionally, it includes a built-in
Distributed Hash Table (DHT) that can be used for simple name
service registration and resolution, thereby helping to realize
autonomic network management and control and zero-touch deployments.
https://datatracker.ietf.org/doc/draft-bless-rtgwg-kira/
https://s.kit.edu/KIRA, more background information
* Laurent: different domains: have you investigated the cases where
different areas are disconnected?
* Roland: we looked at network partitioning/split with possible
rejoining
* Paul: what is the difference of your aprpoach with more common SDN
architectire?
* Roland: this is a fully distributed solution to provide connectivity
from the controller to all switches. It is only for control plane,
data plane routing will be a completely different thing
*
# Minutes {#minutes}
* 13:00 Introduction, RG Chairs, 05 min
***NMRG Research Agenda topics***
* 13:05 **Knowledge Graphs for Network Management, Ignacio
Dominguez-Martinez**, 15 min
* Jeffrey: Annotations will be good to add to Yang to define a
module for ontologies to specifiy elemnts like interfaces +
augment modules
* 13:20 **Semantic Metadata Annotation for Network Anomaly Detection,
Thomas Graf**, 15 min
* Alfons: Is this architecture with all the level of controllers
suitable for ML applications?
* Thomas: the semantic for control plane is well defined, main
challenge is for the management plane
* Roland: annotated data will be super helfpul for academia
* Laurent: have you consider approaches where you inject faults or
stressing events? If so, how to make sure that the process is
helpful for operational purposes?
* Thomas: We have been strating with a kind of lab environment but
there is a difference between lab and real network
* Qin: Work related: ECA draft (how to respond to network changes)
* 13:35 **Artificial Intelligence Framework for Network Management,
Pedro Martinez-Julia**, 15 min
* Alfons: communication between all elements as a control and
management service bus looks very similar to SMO architecture in
ORAN.
* Pedro: we did not look at ORAN but we are looking more at a
highler level view. Use kafka in OSM. We do not target to have a
single bus but integrate all of them
* Alfons: SMOS also uses Kafka with REST
* 13:50 **AI-Based Distributed Processing Automation in Digital Twin
Network ;
Considerations of deploying AI services in a distributed method,
Yong-Geun Hong**, 15min
* Laurent: there is a quite stable and mature work in the draft
with the extension to an application/user-case. We can run a
last call for adoption.
* 14:05 **Performance-Oriented Digital Twins for Packet and Optical
Networks, Albert Cabellos**, 15 min
* Roland: your model is more generic. It can be used for
predictions on ho thinks works
* Alber: we use to predict what happens
* Roland: events simulaotr can be used on different
abstractions/layers and it can works well defpening of what you
want to achieve (for example flow model)
* Albert: simulations are very slows and there is case where is no
simultion, for example for optical networks there is no
simulator but we can ask a model
* Roland: SIGCOMM paper on data-driven model of networks
* Laurent: dedicatd session of Digital Twins for 6G in recent 6G
Net conference with an interesting presentation from Roberto
Minerva.
* Laurent: question about the definition of a DT. A signle answer?
Views are incompatible? It is not a definitive definition.
Depending on what you want to do with a DT, there are different
expected properties which are not exclusive. Do we need to spent
time to the most representative definiton ?
* Albert: the confusion comes that we do not have models to run
networks (no equation)
* Cheng: DT is not only a model. It includes also control,
decision. We control the model with a control loop
* Olga: is there a third option? with intent integration ?what if
scenarios
* Albert: the control view of a DT is a controller
* Marco Liebsch: Agree that NTD should answer the question ‘how
changes the behaviour of my network if THAT happens?” One point
about THAT: Do you think the enforced change or event (THAT) is more
on the load on a path, or more in the configuration of a node in a
NDT, e.g. its queue size, scheduler, etc. The latter may require
creating a new model, while the impact of adding load to one
interface/path is what an existing well trained model can help to
answer. Do you agree?
* 14:20 **Intent-Based Network Management Automation in 5G Networks,
Jaehoon Paul Jeong**, 15 min
* Laurent: we have sevral IBN use cases. In additon to a use case,
you propose an analysis. A suggestion is to discuss with other
authors of other IBN use cases: commonolaties, connections (for
example measurement intent) as you want to provide connectivity
* Alex: interesting to have this as a use case and also an
experience report.
* 14:35 **Challenges and Opportunities in Green Networking, Alex
Clemm**, 15 min
* Jérôme: regarding this draft like other stable drafts, assess
the maturity and go for a kind of "last-call": a couple of
emails to be sent. Also, as green networking might be part of
NMRG research agenda, this drafts gives already some direction
* Vetsna: express support to this draft and the idead of a joint
workshop with e-impact
***Open forum***
* 14:50 **KIRA: Scalable Zero-Touch Routing for Autonomous Control
Planes, Roland Bless**, 15 min
* Laurent: different domains: have you investigated the cases
where different areas are disconnected?
* Roland: we looked at network partitioning/split with possible
rejoining
* Paul: what is the difference of your approach with more common
SDN architectire?
* Roland: this is a fully distributed solution to provide
connectivity from the controller to all switches. It is only for
control plane, data plane routing will be a completely different
thing