Table of Contents maxLevel 2
...
Analytics as a Service Closer to Edges
Problem Statement:
The goal of Analytics as a Service closer to edges is address edge Scalability, Constrained Environment and Service Assurance Requirements.
- Avoid sending large amount of data to ONAP-Central for training, by letting training happen near the data source (Cloud-regions).
- ONAP scale-out performance, by distributing some functions out of ONAP-Central such as Analytics
- Letting inferencing happen closer to the edges/cloud-regions for future closed loop operations, thereby reducing the latency for closed loop.
- Reference: Reference: https://wiki.onap.org/download/attachments/8225716/ONAP-edge-automation-update-arch-use-case-10-23-2018.pdf?api=v2
5G use case relevance
5G/performance Analysis and Optimization: High Volume and RT Data Collection/Analytics/Closed Loop of performance metrics at the Edge Cloud
- Reference: ONAP-edge-automation-update-arch-10-29-2018-followup-11-07-2018.pptx
- Reference: 5G_UseCase_for_Dublin_v4.pptx (reference slide adapted from this deck)
Dublin Requirements Summary: Distributed Analytics as a Service (Dublin Summary) - Edge Automation
Architecture Scope:
- Instantiation of edge and connectivity to ONAP central (out of scope for ONAP)
- Edge Cloud Registration [Ref. Arch. Impact Details (1)]
- Automation of registration when scale (>100s)
- ONAP edge functions or 3rd party edge functions deployed at edge (e.g. Analytics, Closed Loop Control) [Ref. Arch. Impact Details (21 , 22)]
- Registration of the edge functions to ONAP central (Intent, capabilities, capacity)
- Intent Example: “Infrastructure Analytics as service for Alerts at Cluster Level and Host Level”
- Registration of the edge functions to ONAP central (Intent, capabilities, capacity)
- Deploy Network Services in an optimal way to the edges using edge/central functions [Ref. Arch. Impact Details (3)]
- Includes multiple VNFs on multiple edges/core which make a service
- Cloud region (means one control plane) choice
- Connect the service to the functions
Networking of ONAP Central and edge functions functions [Ref. Arch. Impact Details (5) ]
Reference: References: https://wiki.onap.org/download/attachments/41422290/ONAP-edge-automation-F2F.pptx?version=1&modificationDate=1540934276000&api=v2; https://wiki.onap.org/download/attachments/28379482/ONAP-edge-automation-update-arch-10-29-2018-followup-11-07-2018.pptx?api=v2
Jira | ||||||
---|---|---|---|---|---|---|
|
Jira | ||||||
---|---|---|---|---|---|---|
|
ONAP-based Analytics as a Service Details:
...
(see Distributed_analytics_v3.pptx in Edge Automation through ONAP)
- What does ONAP-based Analytics Service encompass?
- Support analytics-as-a-service in the cloud-regions that have K8S site orchestrator.
- Use same analytics framework to have analytics even in ONAP-Central.
- Two packages - Standard package and inferencing package.
- Use existing analytics applications - TCA to prove this framework.
- As a stretch - Showcase one ML based applications
- Training application
- Inferencing application
- How to Develop?
- Use PNDA as a base
- Create/adapt Helm charts
- Ensure that no HEAT based deployment is necessary.
- Use components that are needed for normal analytics as well ML based analytics (Apache Spark latest stable release, HDFS, OpenTSDB, Kafka, Avro schema etc..)
- Use some PNDA specific packages - Deployment manager as one example.
- Develop new software components
- that allow distribution of analytics applications to various analytics instances
- that allow onboarding new analytics applications and models.
- that integrates with CLAMP framework (if needed)
- Impacted ONAP Projects
- DCAE, CLAMP, A&AI (TBD), Multi-VIM/Cloud (TBD)
- How to Test?
- Deploy using Helm charts (PNDA+ : to address large number of Cloud regions, Machine learning workloads) at ONAP-Central as well as in K8S based data centers.
- Use existing analytics frameworks that have already instantiated.
- Deploy analytics applications from ONAP
How to Test:TCA (Changes - Convert this as a spark application)
New Machine learning models for KPI (packet loss) prediction (New use case)
3rd Party Analytics Application - Dublin Scope
- Register VMware VIO (OpenStack-based) edge cloud region(s) in ONAP Central (Note: These Edge Cloud Regions support K8S)
- Deploy VMware vROps 3rd party infra analytics framework/application in target edge cloud region to monitor multiple edge cloud regions
- References:
- https://docs.vmware.com/en/VMware-vCloud-NFV-OpenStack-Edition/3.0/rn/vCloud-NFV-OpenStack-Edition-30-Release-Notes.html
- https://docs.vmware.com/en/vRealize-Operations-Manager/6.7/vrealize-operations-manager-67-reference-architecture-guide.pdf
- In vrops reference architecture this is known as multiple data centers with remote collectors.
- ONAP project impact: None
- Note: 3rd party infra analytics application configuration is out of scope for Dublin
- References:
- Deploy microservice in target edge cloud region for cloud infra Event/Alert/Alarm/Fault Normalization & Dispatching to ONAP Central
- Note: This microservice uses the Multi-VIM/Cloud project repository for development
- ONAP project impact:
- Multi Cloud project impact
- Cloud infra Event/Alert/Alarm/Fault Normalization & Dispatching microservice development
- Integrate DMaaP (Kafka) client for communication to ONAP Central
- Receive Event/Alert/Alarm/Fault from 3rd party infra analytics application
- Normalize from cloud specific Event/Alert/Alarm/Fault format to cloud agnostic (ONAP internal) Event/Alert/Alarm/Fault format
- ONAP internal format references
- Control Loop Design
- Questions: Does each control loop need a separate policy component?
Alert examples (Note: Host CPU Threshold & Memory Contention Threshold in a cloud region are defined separately)
Exemplary Alert Definition:
Standalone host has memory contention due to overpopulation of virtual machines.
- Symptoms include the following:
- Host memory workload at warning/immediate/critical level
- Host memory contention at warning/immediate/critical level
- > 50% of child virtual machines have Virtual machine memory workload at warning /immediate/critical level
Exemplary Alert Action:
- Scale out VNF
...
- ONAP internal format references
- Dispatch Event/Alert/Alarm/Fault to ONAP central using DMaaP (Kafka) client
- Cloud infra Event/Alert/Alarm/Fault Normalization & Dispatching microservice deployment
- Develop K8S Helm chart
- Note:
- This microservice is focussed on operational workflow & and independent of the current deployment focused micoservices in multi vim/cloud.
- Cloud infra Event/Alert/Alarm/Fault Normalization & Dispatching microservice development
- Multi Cloud project impact
- Register 3rd party infra analytics application in ONAP Central (Stretch Goal)
- ONAP project impact:
- Multi Cloud impact - Below
Populate Intent in A&AI
Infra Analytics as service exemplary intent -- “Infrastructure Analytics as service for Alerts at Cluster Level and Host Level for a Cloud Region”
- Capabilities (not exhaustive) corresponding to intent in A&AI (Note: Cluster CPU Threshold & Memory Threshold are defined separately)
Cluster has memory contention caused by more than half of the virtual machines
Cluster has memory contention caused by less than half of the virtual machines
Cluster has unexpected high CPU workload
- ...
- Capabilities (not exhaustive) corresponding to intent in A&AI (Note: Cluster CPU Threshold & Memory Threshold are defined separately)
- Populate Cloud Region List in A&AI corresponding to Intent
- A&AI: Leverage existing HPA/Intent key-value pair schema
- Multi Cloud impact - Below
- ONAP project impact: