Skip to end of metadata
Go to start of metadata


  File Modified
Microsoft Powerpoint Presentation Distributed_analytics_v3.pptx Nov 28, 2018 by ramki krishnan

4 Comments

  1. It is not clear how multicloud could help on "Normalize from of cloud agnostic event/alert/fault" and dispatching them from edge to central ONAP (will multicloud has SBI to a "analytic as a service"? Will there be some mandatory API should be imposed to those "analytic as a service" so that multicloud could consume?

    1. As depicted in the architecture slide 6, multi cloud micro service will present a consistent Northbound API to ONAP components. Southbound api to multi cloud in analytics application dependent. The multi cloud micro service can be distributed in the edge or central or both. It is highly desired to be fully distributed to minimize communication across latency/bandwidth sensitive WAN.

      For Dublin our thoughts are,

      Infra objects of interest:

      • host, host aggregate (cluster), physical DC, availability zone, VMs, K8S PoD
        • Note: Distributed cloud modelling to support multiple distributed physical DCs be managed by a single cloud control plane.

      Infra resources of interest:

      • memory, cpu, network bandwidth

      1) Threshold crossing Alert - Leverage VES DRAFT 

      • Basic - single resource/object, examples below
        • cpu threshold exceed for host
        • memory threshold exceed for host aggregate (cluster)
      • Advanced - multiple resources/objects, examples below 
        • cpu threshold exceed AND memory threshold exceed for host AND host aggregate
        • cpu threshold exceed OR memory threshold exceed for host

      2) Supply additional details for Policy to perform scale out action


      • Derive impacted virtual infra objects from impacted physical object - for example, impacted VMs/K8S PoDs based on impacted host
      • Derive impacted VNFs from impacted virtual infra objects
      • Trigger scale out event using current format from DCAE to Policy – See Scale Out Flow Diagrams

      3) Supply additional details for OOF to perform fine grained placement action 

      • Exact format to be defined.


      1. What confuses me is exactly the statement: "Southbound api to multi cloud in analytics application dependent".  Due to this dependency, it implicates that whenever some analytics application exposes new API to multicloud (plugin?), multicloud (plugin?) source code should be updated (hence re-deployed) to consume that new API? If I does not get the proposal wrong, that multicloud (plugin) is not part of analytic applications.

  2. How much resources a typical analytic apllication based on PNDA consumes? The estimation from PDNA seems pretty big, any plan to test it over integration lab?