2021-10-22 AI/ML for NFV Meeting Minutes

Anuket Project

2021-10-22 AI/ML for NFV Meeting Minutes



Attendees

@Beth Cohen

@Girish

@Sridhar Rao

@Ildiko

@Al Morton



Sl. No.

Topic

Presenter

Notes

Sl. No.

Topic

Presenter

Notes

1

Model-Selector Update

@Sridhar Rao

https://github.com/opnfv/thoth/tree/master/tools/modelselector

Patch is submitted.

Source of the flowchart is also included. .drawio file. Can be modified with https://app.diagrams.net/, or its desktop version

ToDo: Create documentation, more testing.

2

FP Model Dev. Update



Patch is merged.

https://github.com/opnfv/thoth/tree/master/models/failure_prediction/jnotebooks 

Orange Data: https://www.kaggle.com/imenbenyahia/clearwatervnf-virtual-ip-multimedia-ip-system

ToDo: Documentation.

Note: There will be no progress on this work going ahead until we get a fresh data (maybe a different failure).  Desperately need a common data model.  @Beth Cohen will talk to the EUAG about actually working on this project of creating the data model.  Also talk to @smcasey about anonymizing some VZ data to share.  Not sure what the status is of the state of the data.

Question what type of data are we looking for?

  • Look at Orange's anonymized port stats, CPU memory, storage usage.  No traffic data is available currently.

  • Can we look at  creating traffic flow data?  Need to standardize definitions 

3

Kubernetes Failure Emulation



@Sridhar Rao: going ahead this will be the focus 

4

Whitepaper

@Beth Cohen

@Beth Cohen: Put-together CMCC, Telecom Italia and Verizon.

Key Takeaways: maturity is low, more research and development needed, most of the work in operational and maintenance domain, low-hanging fruit - service assurance issues, shared understanding of intelligent n/w needed, reliable data access is difficult, community can play important role in creating a standardized shared data set.  Paper will be published Nov 2021, with a webinar scheduled for Nov.  @Sridhar Raois invited to participate in Webinar.

5

Tool: Data Extraction



No Progress.

6

Explore: GANs for Synthetic Data Generation



Work in progress. Next week (29th Oct), demo of initial metrics-data generation.

@Sridhar Rao  and @Girish to attend Barometer project and share the work for feedback and getting more reference data.

7

Explore: Openstack Log Analysis with Google BERT



Nova-Logs (Nova-compute, nova-scheduler, and nova-api). 

BERT for anomaly detection - work in progress.