Versions Compared

Key

  • This line was added.
  • This line was removed.
  • Formatting was changed.


Friday The 13th 

...

Attendees

Sridhar Rao

Beth Cohen

Al Morton

Rohit Singh Rathaur

Girish

Kanak Raj

Steve Casey




Sl. No.TopicPresenterNotes
1Project Status
  1. Repo is created :
    1. https://github.com/opnfv/thoth
    2. https://gitlab.com/anuket/thoth
    3. https://gerrit.opnfv.org/gerrit/admin/repos/thoth
  2. https://gerrit.opnfv.org/gerrit/settings/#Agreements

TODO: Create LF-Ticket to setup Jira.

2Data Status

3-Approaches:

  1. Use Publicly available.
  2. Request 
  3. Create.

Tool to emulate failures in K8S cloud. 

Demo the Cloudstress to stress the K8S cluster, and see if we can create failures of the pods. 

TODO: Run this tool in Intel-Pod18, and experiment.

Data-Source to Failure-Type listing is completed - Failure Prediction using AI/ML in NFV Environments 

3Model Status

Implementation Status.

  1. Decision Tree and XGboost
  2. NN - LSTM with Attention – 90% (visualization errors) 
  3. GraphNN — To complete in 1

Data is labelled based on the conditions.  This label is used.

Before the failure time-steps (60) - 10s for each step. Trying to increase to 300.

Casey: Better to do it as the failure-time, rather than before.

How many previous time-steps to consider ? Depends on the usecase (Ex: type of the Equipment). 8hrs to 24.

4AlgoSelection Graph Status

Unsupervised: Working on Gaussian Mixture models.

Kanak: Unable to come up with a clear "separation-parameter" at the last level.

Jahanavi: