Minutes IETF106: nmrg
||Minutes IETF106: nmrg
NMRG 57th meeting
IETF 106, Singapore
. 2 first IBN drafts to call for RG adoption.
(draft-li-nmrg-intent-classification and draft-clemm-nmrg-dist-intent)
. Revised charter available and under final evaluation, containing
updated research agenda reflecting the IBN, Network Management AI and
Self-driving/-managing frameworks work plan. . Report on IBN practical
activities with demos and tools. . Connections to other proposed
frameworks and activities in IETF OPS and NETMOD on ECA framework for
self-management and Service Assurance Graph related to IBN. . First
steps on Network management Articificial Intelligence with proposition
for a Research Challenges document. . First proposal for Network
Management AI Challenge and stabilization of the Network Management AI
*********** Session 1 ***********
Thursday 2019/11/21 10:00 - 12:00
Remote participants (4): Mouli Chandramouli, Olga Havel, Will Liu, Rafael
1. Introduction + RG information, Chairs
2. Quick report on October interim meeting on practical aspects of IBN, and
follow-ups, Chairs 3. Status and progress of active I-Ds 3.1 Update on
Intent Classification, Olga Havel (remote)
Jérôme (as participant): criteria on "feedback". seems exclusive (feedback
or not). the demanding user may not need feedback on the realization, but
the network administrator could require feedback/information. Olga:
correct. in the table, it is specified if the user is non-technical or
technical (administrator). Alex: not clear in the tables: how going to use
them, very detailed. different methods applied for different cases, or as a
catalog. Olga: advice come from Laurent. intent was to classify the
PoC/demos using the draft and thus ended up with the mapping table
proposal. How many read the draft: 11+, useful: same+ ; ready for adoption:
a few/mixed. Chairs will send call on the mailing list and check based on
3.2 Update on IBN Concepts and Overview, Alex Clemm
Diego: categories still missing in the other draft. something that could be
done there. security considerations: important and long forgotten. desserve
more attention: intent security and how to express security intents. How
many read the draft: 12+, useful: same ; ready for adoption: 10. Chairs
will send call on the mailing list and check based on feedback.
4. Technical presentations and discussion
4.1 IBN for network slicing in 5G-EVE project, Luis M. Contreras,
Laurent: orientation of the draft: type of draft (informational,
experimental), for IETF or IRTF? Luis: work on making the EVE interface
really compliant with GSMA/3GPP NS template.
4.2 A Solution for Goal-oriented Policy Refinement in NFV Management and
Orchestration Systems, Michel Sales Bonfim What is DL = Description Logic.
HTN = Hierarchical Task Decomposition Diego: planning descisoin done once,
or regularly? Michel: short overview. another mechanism. solution can
detect if policies are not working properly. the planner can generate
alarms, when triggered can reconfiguration Diego: NFVO inject NS goals...
Michel: extension to NFVO to have goals in the NS descriptor. Diego: in the
planner, consider interactions between different NS goals? Michel: not
considered yet. Laurent; clarification on the acronm 'HTN'. method used for
task decomposition. Laurent: other proposals exists for task composition,
related work? Michel: paper under submission, we can discuss more the
related work (there are many papers for sure) Laurent: relationships with
the proposed lifecycle, exchange with drafts' authors. encourage to link
with the group for practical aspects (tools/demos), and the
concept/lifecycle work as contribution for the different functionalities
4.3 Intent-based networking for OTT applications: concepts, lifecycle and
challenges, Sabine Randriamasy Dean: "emergence" traffic? -> Sabine:
emergency (in the sense of high priority, critical communication) Diego:
topology is essential in composition. How do you define a slice? Sabine:
slice = cloud + network resources to run the applications, chain of
functions, so basically connectiviy + application. Diego: Change OTT >
Verticals. Dean: multiple service topologies per slice. Sabine: agree. this
presentation skips the intent definition step. here very simple example.
more complex scenarios possible.
4.4 Service Assurance for Intent-based Networking Architecture, Benoît
Laurent: Assurance graph can be used in the IBN lifecycle. Can be used as a
tool for this step in the lifecycle. Where the proposal would fit in the
IBN architecture? Benoît: consder intent as a service definition. From the
specific service type that is described, we can deduce the specific actions
and so the tree. Jeff: L3VPN not an intent. need upper layer to process the
intent, and then map to lower layer choosing e.g. L3VPN to instantiate the
service. Alex: it is not really intent-specific but more widely usable. How
to maintain these assurance graphs? practical issues? Benoît: partly from
the definition itself, from network configuration.
4.5 Framework for Use of ECA (Event Condition Action) in Network
Self-Management, Daniel King
https://tools.ietf.org/html/draft-bwd-netmod-eca-framework-00 Diego: yes to
the 3 questions raised in the presentation, esp. link with intent
framework, e.g. for the low-level application of intent. Laurent: on Q2:
ECA can be at the end of chain of intent realization, so there are
relations to clarify and develop.
Proposal (RG chairs): send drafts from Benoit and Dan and to the NMRG
mailng list ask feedback from the group. Where it would fit in the RG
work (on IBN in particular).
4.6 Next steps on Architecture and functionalities, use cases and
techniques remaining time. 5. AOB
items skipped due to lack of time.
*********** Session 2 ***********
Friday 2019/11/22 12:20 – 13:50
Remote participants: 4
1. NMRG recharter, Chairs
Dean: management architecture concern. not changed in last 20 years.
may be re-thinking the basic mgt arch. (against controller/agent),
other options. gather requirements. lots of new paths. can those paths
be done over existing arch. running into problems is using the curent
arch. to do e.g. autonomics. would be curious to hear from the
community. Jérôme: see that in the charter. please bring it to the
mailing list. Colin: if proposals for arch. could be topic for conclude
IBN resonable plan. suggest the rate of progress, not as energetic as
other groups. intesify. AI for NM. plausible topics of the future.
looking at today's agenda. if Netwokr AI, could be a good sign for the
recharter. if only reports from other groups, then sign of low
interest/issue. Colin: use another term than self-driving networks, too
close to self-driving car. Alex: having AI for network in the charter
as a means to attract community. Colin: should not be a a goal in the
charter to see if there is a Network AI community. there should be
already enough interest before going into the charter. Will: following
other activiites...? (don't got it fully) Marie-Jose: how AI can help
NM work. not only study AI for network.
2. Artificial intelligence for networking and network for AI
2.1 Adversarial Network Benchmarking, Andreas Blenk
Dean: interesting work. draft from Comcast. point to you.
2.2 AI activities in other communities
2.2.a) ETSI ISG Experiential Networked Intelligence (ENI), Will Liu
2.2.b) ETSI ISG Zero-touch network and Service Management (ZSM), Laurent
Ciavaglia Will: ENI and ZSM will have colocated meeting in December in
2.2.c) IEEE ETI for Network Intelligence (NI), Laurent Ciavaglia
Jerome: what are the actions of the ETI NI on "network data as a key
challenge" Laurent: no clear path yet, identifying it as a clear challenge
was our first goal, next steps consist in documenting items detailing types
of network data and how commplex for AI. Dean: we have already today basic
sets of data to be used (and identified). It gives you some starting points
rather than trying to take all and then find what's relevant. Laurent:
agree from operational point of view. Diego: operatorational teams do not
report so much, we may share something in IETF 107 or 108 (Madrid) on Data
Representation Framework. Nalini: enteprise network. encryption everywhere.
how to run AI on encrypted data. packet data and meta-data at the end
points is also very challenging Laurent: no experience on encrypted user
data for use with AI techniques. more on the network management data (logs,
scripts, config files), you can access these data sets (as an operator).
Jérôme may have better feedback on the use of AI with user data. there are
also data that you can access from a netman point of view.
2.2.d) ITU-T FG on Machine Learning for Future Networks including 5G
(ML5G), Vishnu RAM OV
2.3 Network AI challenge
2.3.a) Update on the NMRG activity on the AI challenge, Jérôme François
2.3.b) Challenge proposition, Albert Cabellos
NMRG welcomes the challenge. Let us know how NMRG can help. RG, if
interest, please contact Albert or the chairs.
2.4 Wrap-up of Network AI side meeting, Chairs
Shen: work on labeled data sets that can be re-used.
Jérôme: agree we need to have that. however, data will be diverse so
universal decrpition will not be always suitable. Who's interested in
contributing to the report: 5-6 participants expressed interest to actively
end of meeting.