1

Review 2021

We have met met 22/25 times over last 6 months.

Significant Contributors: @Rohit Singh Rathaur and @Girish  - Our Researchers

Published model for Failure prediction (VM) based on Orange Dataset.

Published a tool Model-Selector.

Published two research works.

Ongoing works: Call for Contributions - Potential works for contributors 

Challenges:

1. Availability of the Dataset.

2. Contributions are only from Students.

@Girish: Does CM plan to share Problem/Dataset?

@Lei Huang: Internal Evaluation is ongoing, and once completed will update.

2

Thoth - ITU - 2022

(a) Problems

@Sridhar Rao

Thoth to Host two problems next year at ITU-AIforGood.

Host: Thoth/Anuket/LFN

Problems:  

1. Packet-Loss Classification in NFV Environments: Start with two categories (a) Lack of resources (b) Transients. Dataset: Packet-loss trends (every 1-5 secs) from traffic-generators. Source: Commercial traffic-generators or opensource. We need to agree on a single data-model. Minimal data: timestamp, pkts-sent, pkts-recvd, pkts-lost, 

2. Synthetic monitoring data generation using GANs: dataset: CollectD metrics of 1-week, openstack-logs for 1 week. Timeseries data.

3

Going Ahead - 2022

1. Continue to work on building models - Mainly start with ITU-dataset.

2. MaaS: Right framework to use. @Karthik Ganesh M and Atif are working on evaluating - Kubeflow and Acumos.  Potential frameworks - https://github.com/opnfv/thoth/blob/master/research-studies/oss-projects-nfv.md

   1. ML Framework for Model As a Service

   2. Model As A Service (MaaS)

3. Continue to build the tools.

   1. 2 and 6 from this link: Call for Contributions - Potential works for contributors 

      1. Log analysis: Pattern Analysis, Anomaly Detection - Using NLP.

4

Data Summit at BIT Mesra

https://datathon.sdsbitmesra.in