Edge Work Update ONAP Arch. Sub Committee Montreal Meeting 10/29/2018 ONAP Edge WG Wiki: https://wiki.onap.org/display/DW/Edge+Automation+through+ONAP Meeting Date/Time: Wednesday, 4:00pm UTC - 5:00pm UTC Leads: Ramki Krishnan ([email protected]) , Raghu Ranganathan ([email protected]) Presenters: Ramki Krishnan, Raghu Ranganathan, Srini Addepalli, Margaret Chiosi ONAP Edge WG Participation 2 Agenda Quick Recap on Edge Functional Requirements Edge Architecture Scope 3 s Quick Recap on Edge Functional Requirements
Akraino etc. & ONAP Positioning Study usage of MobileEdgeX/TIP vs Enhance ONAP Edge Functions (Near-term) ONAP Central ONAP or Other (Note 1) Adapted from Akraino Blueprint: Link: https://wiki.akraino.org/display/AK/Akraino+Network+Cloud+Blueprint+-+Reference+Architecture?preview=/1147251/1147516/image2018-8-9_16-32-22.png Note 1: The Edge Functions for NFV Orchestration could be provided by ONAP or other cloud solutions 5 Akraino etc. & ONAP Positioning (Mapping Requirements) Edge App provisioning Security Service
Assurance & CL MEC API Study usage of MobileEdgeX vs Enhance ONAP Edge Functions (Near-term) ONAP Central ONAP or Other (Note 1) Constrained Environment (CNF support, SRIOV-NIC, FPGA, GPU etc..) Zero touch provisioning (ML Model Inference at edges) Performance Aware
Workload Placement & Mobility Scalability (By offloading some functions to edges/closer to edges) Adapted from Akraino Blueprint: Link: https://wiki.akraino.org/display/AK/Akraino+Network+Cloud+Blueprint+-+Reference+Architecture?preview=/1147251/1147516/image2018-8-9_16-32-22.png Note 1: The Edge Functions for NFV Orchestration could be provided by ONAP or other edge cloud solutions 6 Suggested Dublin focus Areas * Under Discussion * (need to clean this up to match scope) Network Analytics closer to edges (Scalability, Constrained Environment, Service Assurance Reqs.) Lead: Intel
Bring up of Apache Spark + HDFS + Kafka framework from ONAP in multiple edge clouds. Bring Network analytics applications from ONAP-Central on various analytics instances. Machine learning model creation; Inferencing package at the edges API from Central ONAP to Edge (Note 1) Workload Placement & Mobility closer to Edges (Scalability, Performance/Isolation-aware Workload Placement/Mobility, Service Assurance Reqs.) [Ref. 1] Lead: VMware (in finalization) Bring up several ONAP components (OOF etc.) edge clouds. API from Central ONAP to Edge (Note 1) Study of Edge Application Orchestration and various open source projects. Leverage MEC APIs from ONAP workloads for Edge MEC Applications. Note 1: API from Central ONAP to Edge provides the flexibility of using Edge Cloud Provider Solutions as an alternative to ONAP components in Edge Ref. 1: http://networks.cs.ucdavis.edu/presentation2017/WeiWang-02-10-2017.pdf; http://www.ict-coherent.eu/coherent/wp-content/uploads/2016/08/Katsalis_MEC_CLOUD_2016.pdf; Important work item in https://5g-ppp.eu/ Suggested Beyond Dublin Additional focus Areas * Under Discussion *
Closed Loop (ML Model Inference at edges) (Scalability, Service Assurance Reqs.) API from Central ONAP to Edge (Note 1) Complete sub-service Orchestration in Edge (Massive Scalability Reqs.) API from Central ONAP to Edge (Note 1) Note 1: API from Central ONAP to Edge provides the flexibility of using Edge Cloud Provider Solutions as an alternative to ONAP components in Edge s Edge Architecture Scope Architecture Scope Instantiation of edge and connectivity to ONAP central (out of scope for ONAP) Edge Cloud Registration [Ref. Arch. Impact Details (1) slide 15] - 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 (2) slide 16] - Registration of the edge functions to ONAP central (Intent, capabilities, capacity) Closed Loop Control exemplary intent - X percentile closed loop recovery time is Yms where X could be 99.99 and Y could be 20. This example means your closed loop recovery time is less than 20ms 99.99% of the time under a defined observation interval.
Deploy Network Services in an optimal way to the edges using edge/central functions [Ref. Arch. Impact Details (3) slide 17] - 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 [Ref. Arch. Impact Details (5) slide 19] 10 s Edge Analytics as a Service (Example) Analytics as a Service (An example) Register analytics site instance Onboard ONAPAnaltyicspackages
Register inferencing site instance ONAP-Central Instantiate standard analytics instance Instantiate inference analytics instance Cloud-region Standard Analytics engine Cloud Region Cloud Region Inference engine Cloud-region Cloud Region
Cloud Region Using PNDA & Open source, create packages & helm charts Two packages Standard package consisting of Apache Spark, HDFS, OpenTSDB, Kafka And ML packages including MLLib, sparkscikit-learn, sparktensorflow Inference package consisting of Kafka, Apache Spark Notes: Not shown, but standard package can be instantiated in ONAP-central and edges too 12 Analytics - App Image onboarding and usage Register analyticsapp Onboard Analytics App - (for every new app and model):
Register Model Analytics BP VNF instantiate ONAP-Central Upload app, if does not exist Run application Cloud-region Standard Analytics engine Onboard VNFs : For every VNF (if analytics is needed) Analytics BP Upload app and model Run application Cloud-region
Cloud Region Inference engine Cloud Region Cloud Region Upload blueprint (Enhance CLAMP?) Set of mappings, with each map to have VNFs and metrics to analytics app + model Configuration of analytics app Whether to bring up new APP or use existing App. VNF Instantiation: Cloud Region Register Analytics-App
Register Model. Upload application image and training app if they dont exist in the cloudregion Upload app/model to inferencing site if they dont exist Run the applications VNF termination Stop the application Get model and register with new version (if needed) 13 s Architectural Details Architectural Impact Details (1) Preparatory Steps Edge Cloud Region Registration in ONAP Central
Edge Cloud Region Registration - Leverage current Multi-cloud, ESR, A&AI registration mechanism - Gaps w.r.t to Edge Handling dynamic capacity addition to edge cloud region Add edge cloud region dynamically when ONAP is in operation ONAP Project Impact: TBD 15 Architectural Impact Details (2) Preparatory Steps Edge Function handling in ONAP Central Edge Function Examples - ONAP-based or 3rd Party Microservice(s) for e.g. analytics, closed loop service assurance, workload placement/mobility etc. - ONAP Multi-VIM/Cloud (MC) to provide generic abstraction layer for all Edge Functions Edge Function Packaging/Deployment Example - ONAP-based Cloud Native Analytics Microservice On-boarding/Designing Leverage current DCAE work? SDC to select new model (e.g. ML) Distribute new model (e.g. ML) to Analytics App in DCAE Deployment Readiness
Leverage current DCAE work? Supply model parameters through environment file Create K8S Helm chart - Deploy ONAP-based/3rd Party Cloud Native Analytics Microservice in Edge Cloud w/ K8S support Leverage current K8S cloud region work? Edge Function (e.g. ONAP-based/3rd Party Cloud Native Analytics Microservice) Registration - Gaps w.r.t to Edge The Edge Function is discovered through ONAP Multi-VIM/Cloud (MC) and the capability/capacity details are registered in A&AI w/ appropriate cloud region(s) association ONAP Project Impact: TBD 16 Architectural Impact Details (3) Network Service Deployment - ONAP Central Deploy Network Service (e.g. vcpe) on edge cloud region which supports Edge Function (e.g. ONAP-based Cloud Native Analytics Microservice) - Predominantly Leverage current deployment method (SO, OOF, MC, Policy etc.) - Gaps w.r.t to Edge Cloud Region Selection Edge Function being there is also used in the cloud region selection process by OOF. The following properties of Edge Function are used for this purpose capability/capacity details in A&AI cloud region association in A&AI
network service type association in A&AI intent in A&AI Closed loop service assurance exemplary intent -- X percentile closed loop recovery time is Yms where X could be 99.99 and Y could be 20 The above example means your closed loop recovery time is less than 20ms 99.99% of the time under a defined observation interval ONAP Project Impact: TBD Network Service Association Associate Network Service instance which is deployed in the edge cloud region to the selected Edge Function ONAP Project Impact: TBD 17 Architectural Impact Details (4) Network Service Operation - Edge Edge Function communicates with network service instance and the edge cloud region where the network service instance is deployed to gather various types of information (e.g. infra and app metrics) Gaps w.r.t to Edge - ONAP-based Edge Function ONAP Project Impact: TBD - 3rd Party Edge Function No ONAP impact
18 Architectural Impact Details (5) Network Service Operation Edge to ONAP Central Communication Exemplary communication - Network Analytics Service - Aggregate infra metrics from Edge VIM and app metrics from Edge VNF Gaps w.r.t to Edge - ONAP-based Edge Function Edge Function could directly communicate with ONAP Central (e.g. DCAE) or through MC ONAP Project Impact: TBD - 3rd Party Edge Function Edge Function communicates with ONAP Central only through MC No ONAP impact 19 s Additional Material Edge Network & Application Profile Details Application Network/Application Examples Classification
(RTT-based) Network / Service Deployment Behavior Type Component SP Edge Cloud Third Party Edge Deployment Application Edge Cloud Constraint (RTT- Provider based) 1 Real-time Network Function Data Plane e.g. 5G CU-UP, UPF Network Data (20100ms) Processing Cloud Edge Yes No
Hard NF Vendor, Service Provider 2 Near-realNetwork Function Control Plane e.g. 5G CU-CP time (500ms and above) Network State Processing Cloud Edge or Yes Cloud Central No Soft NF Vendor, Service Provider
3 Real-time (20-100ms) Real-Time Network Open 5G CUState Control CP VNFC Yes No Hard NF Vendor, Service Provider, 3rd Party 4 Near-realSlice monitoring, performance analysis, fault time (500ms analysis, root cause analysis, SON applications, and above) Optimization (SON Drive Test Minimization etc.), ML methodologies for various apps.
Network Analytics & ONAP DCAE Optimization could be leveraged Yes No Soft NF Vendor, Service Provider 5 Near-realIoT Video Analytics/Optimization, Customer time (500ms Geolocation information, Anonymized customer and above) data etc. Workload Analytics, Cloud Edge or Potential Typical Cloud Central Optimization & Context processing
Soft Typically 3rd Party, Emerging Service Provider 6 Real-time (10-20ms) Hard Typically 3rd Party, Emerging Service Provider In service path optimization applications which run in open CU-CP platform (also known as RAN Intelligent Controller, or SD-RAN controller). Third party applications that directly interacts with Workload UE or Cloud the UEs, like AR/VR, factory automation, drone
Automation/AR-VR/ Edge control, etc. Content, etc. Potential Typical 21 Deep Dive SA + Workload Placement/Mobility Edge Application Placement/Mobility/Service Assurance Policy - App to be bounded by a maximum RTT latency from a group of users w/ 99.99% availability This policy should be continuously met during operation besides initial placement Note: Group of users map to specific physical DC in Cloud Edge - App Examples (from Network & Application Profile table) 5G CU-UP, Drone Control, IoT Video Analytics - Initial focus on Network Functions 22 Deep Dive SA + Workload Placement/Mobility for NFs Edge Cloud Topology (single SP) Single Edge site w/ Single Cloud Control Plane
Key Edge Functions w/ ONAP Project Mapping Cloud Inventory (A&AI), Multi-Cloud Support (MC), Closed Loop Controller (APP-C), Closed Loop Policy (Policy), Closed Loop Analytics infrastructure/application logs/events/metrics/faults processing (DCAE, MC, Logging), IPAM (part of SDN-C) (Note: Service Orchestration is deliberately omitted to avoid HA state Complexity) Multiple Distributed Edge Sites (metro distance) w/ single Cloud control plane Multiple Distributed Edge Sites (metro distance) w/ multiple Cloud control planes Above + Initial Placement/Continuous Optimization (OOF) (Note: Service Orchestration is deliberately omitted to avoid HA state Complexity) Above + All deployment and operation components including Service Orchestration (SO) 23 Architectural Impact Details Preparatory Steps Edge Cloud Bring up & Connectivity - Out of Scope for ONAP Bring up for VIM each Edge Cloud Communication between Central Cloud, where ONAP Central runs on
K8S, and Edge Cloud - Different services providers for Central Cloud & Edge Setup a private VPN - Single service provider for Central Cloud & Edge Private connectivity to Edge Cloud may already be available; If not, setup a private VPN 24
Go to www.yourEconPortal.com. With EconPortal, you get interactive content developed specifically to complement your text . and access to the . resources you need to be successful in your . course all in one location. See below for how to...
What are spreadsheets used for? ... What is a Spreadsheet? A . spreadsheet (or worksheet) is an arrangement of cells in columns and rows used to organize, analyze, calculate, and report information, usually in numerical form. ... such as one...
The lysosomal membrane protects the cytosol, and rest of the cell, from the degradative enzymes within the lysosome. The cell is additionally protected from any lysosomal acid hydrolases that drain into the cytosol, as these enzymes are pH-sensitive and do...
Electricity Meters - intro Per Fridh Electricity Meters from ABB Single and three phase meters IEC approved MID approved Communication modules DIN rail mounted Pro M design Ideal for sub metering and revenue metering ABB DIN Rail Electricity Meters Overview...
Write a one page journal entry about your experience.You should include details about where you are, what kind of music you are listening to and how you feel.Also, talk about how this unifies you and the people around you.
Precursor metabolites, energy from ATP, and enzymes are used in anabolic reactions. ... Activation energy. without enzyme. Activation energy. with enzyme. Products. Figure 5.4 The effect of enzymes on chemical reactions. Active sites similar. to substrate's.
Transparencies for Chapter 6 of textbook_x000d_ Database Systems: A Practical Approach to Design, Implementation, and Management ... Add a new column to a table. Drop a column from a table. Set a default for a column. Drop a default for...
Ready to download the document? Go ahead and hit continue!