Versions Compared

Old Version 1

changes.mady.by.user Sridhar Rao

Saved on

compared with

New Version 2

changes.mady.by.user Sridhar Rao

Saved on

Key

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

Attendees

Sridhar Rao

Lei Huang

Girish

Beth Cohen

Karthik Ganesh M 

Atif



Sl. No.TopicPresenterNotes
1

China Mobile - Thoth 

Q&A Contd.


End-Goals of Using GANs.

  1. Anomaly Detection using GANs - Both Normal and 'Abnormal-Data' is provided for discriminator. This work is part of PhD Thesis and not part of the Thoth Project.
  2. Synthetic Data Generation: This work is part of the Thoth Project. Table GANs and Cycle GANs. WIP.  A CM - NLP + Sequence GANs.
    1. Metrics: Infrastructure. Ex: Collectd.
    2. Logs: System, Service (Phase-2)

collectd-data:  Consider (hosts), ranges of Values a metric can take, Variation (step-changes) possible, 

"Intelligent Networking" Platform.

Data Domains:  NFV Clouds (infra + apps) + Backhaul (legacy N/w - wired and wireless) + RANs (Many AI applications in this domain)

NFV Network vs Legacy Networks (Data):  Legacy Networks Infra-metrics doesn't exists (Device stats/flow-stats/), EM-Data Vs Mano-Data, 

Examples:

P-router Vs vRouter

P-RoutervRouter

Statistics (Flows, Interface-stats, etc.

Statistics (Flows, Interface-stats, etc.) +

Infrastructure Metrics.

Controller/Element-manager DataVNFM/ViM Data.


(a) Infrastructure Metrics: Ex Zabbix, Collectd, Node-Exporter, Icinga, sensu. CPU, Memory, Interfaces, Processes, Storage. Node-Level, or Instance (VM) level. Multiple-Nodes or all nodes from the cloud. 4-7 Days. Text-Data: size should not be an issue.

(b) Logs: Depends on the type of the failure: Failure Prediction using AI/ML in NFV Environments Ex: for VM failure:  nova-compute.log, nova-api.log, nova-scheduler.log, libvirt.log, qemu/$vm.log, neutron-server.log

2LFN Developer Forum

3ITU Proposals Status
2 Proposals. Decision on the 'Entity' who'll be proposing is not yet Decided (Thoth/Anuket/LFN).
  1. Packet Loss-Characterization (Data from ViNePerf project). Start with Binary classification - Losses due to (a) Transients (b) Lack of Resources.
    1. Synthetic Data Generation.  Documentation in Progress.
4ViNePerf (pronounced as Wine - Perf) Collaboration for FP

In ViNePerf we are building a tool to emulate Failures, based on Stress-ng. Making stress-ng to behave a VNF (With time-varying and load-varying).  End-Goal is to 'create' failure data. This stress-ng is also used in all 'Chaos-Solutions' (ex: Litmus). Cloud - K8S clouds.