1

AlgoSelector Update

@Sridhar Rao

Last meeting, Completed: Generic, and Supervised.  

2021-09-17 AI/ML for NFV Meeting Minutes

2021-09-24 AI/ML for NFV Meeting Minutes

Now, Completed: Unsupervised and Reinforcement

Open

Open

Request: Please review and share any feedback.

Code Submission for review: 11-10-21

Steve Casey: Have a suite of models for each of the categories - and may be user an provide the data, a tool will run with all the models and give a suggestion. https://github.com/j-planet/machine-learning-big-loop 

Kanak and Akanksha: Please take a look at this link: Call for Contributions - Potential works for contributors 

2

Failure Emulation

No Breakthrough with stress-ng (process, interrupts, etc.). Not being thorough/systematic enough.

1. create a stress-ng container  2. run as pod in k8s  3. run different loads so to create a pod-failure. 

What didn't result in failure: CPU, Memory and Storage (on node, NOT dedicated Storage nodes) – workloads. 

Girish: stress-ng + network load gen (at the same time) – on VM in Openstack. Seen Failures.

3

FP Model Development

Pending:

1. Reorganization: Move data_preprocessing and static_htmls under failure_prediction.

2. Documentation: Include models.rst file as suggested.

3. Documentation: Documenting important cells (jupyter notebook)

4

Data Extraction Tool Status

Completed the Elasticserach - python APIs.

Testing is pending - dependent on Airship ES.

5

Synthetic Data Generation - GANs

1. Looking at the infrastructure data - collectd ?

2. GANs Implementation.

Work in Progress.

Action: Talk to Barometer project, maybe they can add value with GANs implementation.

6

Exploration: Openstack Log Analysis

Openstack Logs: Starting with Nova only. 

NLP for log-analysis: Logs are Unstructured. Reg-Ex are used to analyze. RegEx can become irrelevant. Hence, NLP will be useful.

In NLP, BERT maybe even more suitable