AcumosONAP DCAE Integration Project Overview Demo Presentation 114

Acumos-ONAP DCAE Integration Project Overview & Demo Presentation 11/4 to 11/13 2019 1

Goals Demonstrate AT&T can ingest marketplace AI/ML solutions to address operations challenges Enable Acumos solutions to be deployed on ONAP/ECOMP Automate (simplify where possible) delivery of ONAP DCAE microservices Integrate with Policy & Orchestration/Controller APIs to support Control Loop Automation Apply capabilities to AT&T’s 5 G network Intelligent Solutions Network Management Marketplace B Marketplace C Marketplace D Accomplished AT&T Networks Mobility Access Acumo s Adaptation • • • Service Assurance • Simplify, Automate • Policy, Control Loop m. S & Policy Onboarding SA/CL Flow Design Wireline Access Core m. S & Policy Config. Deployment Management Customer 2

Acumos-ONAP Integration Project Evolutionary Improvements 1. Initial scope was to build an adaptor to design and deploy Acumos ML models via DCAE 2. We encountered many SDC m. S Onboarding/Design deficiencies and lack of resources to fix them 3. The “DCAE MOD” project was established to simplify the DCAE onboarding and design process Acumos. DCAE Adaptor DCAE m. S Deployment ONAP SDC X BPG ONAP SDC Manual Onboarding via CLI X DCAE m. S Deployment No flow design on DCAE-DS X DCAE MOD Project Acumos. DCAE Adapter Onboarding APIs Acumos. DCAE Adaptor POC Use newly developed design process with onboarding APIs and Designer for the Self Service Control Loop Feature DCAE Designer for SA Flow Design & m. S Configurations Runtime APIs BPG DCAE m. S Deployment MOD = Microservices Onboarding & Design 4. 1 new feature to Acumos Clio, 1 feature + 1 POC to ONAP Frankfurt. DCAE MOD also to be used in ECOMP Add new ML Onboarding input data type Acumos Clio ONAP Frankfurt AT&T ECOMP DCAE 20. 02 3

Acumos Clio New Features to Support ONAP/DCAE integration Problem with using Acumos models directly in ONAP/DCAE • Acumos uses Google’s Protocol Buffers (Protobuf) as both a data description language and for data serialization/transport • But some important data structures for ONAP (such as VES records) are not easy to implement directly using Protobuf • …And Acumos’s data types are largely anonymous, whereas DCAE maintains a dictionary of types which it needs for inter-microservice communication Solution in Acumos Clio Release • Implement a new pass-through (“raw”) mode where Acumos doesn’t try to serialize/deserialize model inputs and outputs—it’s up to the model methods to do their own packing/unpacking • Can support the type of typical ONAP/DCAE model inputs/outputs as just raw byte strings containing JSON-encoded records • Additionally, enrich model metadata with names of DCAE datatypes associated with model method inputs/outputs, so during microservice generation for ONAP, these type names are available 4

Frankfurt Release Features Self Service Control Loop Feature CLAMP DCAE Acumos DCAE Inventory SDC Service Design XDCAE-DS Acumos Policy Existing API to send policy model BPG DCAE m. S CLAMP Retrieve Flow Definition SA Artifact Distribution ML DCAE Designer DCAE m. S Runtime BPG DCAE Designer DCAE m. S API for m. S Onboarding ML DCAE MOD Project ML-m. S Adaptor DCAE m. S Retire Runtime DCAE MOD Project Policy Model ML-m. S Adaptor Acumos-ONAP DCAE Integration POC X DMaa. P x. NF Artifact Distribution X API for m. S & Policy Onboarding DMaa. P Policy m. S Onboarding DCAE-DS Artifact Package & Distribution Service Design Service CSAR BPG DCAE m. S SDC Done Self Service Control Loop Feature Runtime Policy m. S Onboarding Policy Model To do DCAE Inventory Acumos-DCAE Integration POC 5 5

Project Status Highlights 1) 4 internal/external project reviews & demos 11/4 - 11/12 2) One feature to Acumos Clio release 3) 1 POC & 1 feature approved for Frankfurt ONAP release • AT&T ONAP TSC & ONAP Community (PTLs, TSC & Arch) 4) DCAE-MOD on track for ECOMP 20. 02 & ONAP Frankfurt • Minimized cost/risk; avoided SDC impacts 5) Delivering value • Simplified & automated onboarding (ML/AI/Analytics/µServices) • Improved development and SA design user experiences • Consolidated design & configuration tools • Automated and eliminated manual configuration steps • E 2 E tracking of service assurances flows and µServices Phase 1 Project Timeline April May June July 2019 Aug. Phase 2 Demo TBD Pending JDA Phase 1 Review & Demo Sept. Oct. Nov. 2020 Dec. Jan. Feb. Additional Iterations for DCAE-MOD Functions Deliver DCAEPhase 1 Dev. Demo & Read. Deliver DCAE-MOD in Complete out November 4 to ONAP Frankfurt ECOMP 20. 02 October 14 12 2/13 (code freeze) Rel New ML data type for raw data Phase 1 E 2 E Test Phase 2 ML stream – Delivered to Acumos Complete October Training after JDA CLIO 25 (TBD) Design & Development Iterations Project Concept April Initial Arch. & Requirements May Requirements Complete June 24 6

AT&T & Ericsson Partnership ATT & Ericsson collaboration to solve Service Assurance problems with Acumos AI/ML model executed in ONAP DCAE • Explore existing or develop new AI/ML solutions in Acumos • Train & Refine ML models with runtime specific data (e. g. , v. MME FM/PM data) – Phase 2 Use Case Solution • Design and provide a reference implementation of the microservice interactions to support ML functionalities in ONAP DCAE • Platform development to adapt ML models into an ONAP DCAE compliant microservice • Platform development to onboard, design, and deploy the ML as a microservice in ONAP DCAE • Contribute platform enhancements back to Acumos & ONAP Open Source Communities Platform Solution • Expansion & integration with Policy to support Control Loop automation • Focus on platform enhancements applicable to Mobility applications (O-RAN-SC, RIC, SEBA, etc. ) AT&T participating organizations • Jack Murray – Acumos & ONAP platform requirements & development • George Sultana – DCAE MOD development • Glenn Shull – ATS production data (for ML training in Phase 2) • Sal Talamo – Self service requirements • Gilbert/Lu/Abella – ONAP Frankfurt release agreement & submission (SDC/DCAE/Policy/CLAMP) Ericsson participating organizations • Nimish Radia - Ericsson Global AI Accelerator (GAIA), NA GAIA Site & Ecosystem • Shishir Kant Mishra – Ericsson GCU AT&T 7

Acumos-ONAP Integration Development Team Sub-team Task Major Contributors Company/AVP Program Management Sponsorship Dean Bragg, Shishir Mishra ATT & Ericsson Schedule & Status Tracking Kendra Woodson Ericsson E 2 E Technical Lead Platform Solutions Ting Lu ATT Jack Murray ML Use Case Solutions Wenting Sun Ericsson m. S flow design and implementation David Stone, Anirudh Mittal, Prasanna Chalapathy, Zhaoji Huang Acumos Federation Support Kunal Rajan Deshmukh ML model development Lu Le, Nicolas Ferland I-Lab Support Todd Larchuk RDL Support Dominic Lunanuova ATS v. MME Production Data Fredrik Andersson Acumos Development & Support Manoop Talasila, Guy Jacobson, Ken Kristiansen Team 2 ONAP DCAE µS Adapter Requirements Ting Lu Implementations Andrew Gauld, Tommy Carpenter Team 3 ONAP DCAE µS Onboarding & Design Requirements Team 1 ML Development, ML Training (Phase 2), Acumos Development Jack Murray ATT Glenn Shull ATT Jack Murray Carol Nelson, Phil Blackwood Ting Lu Implementations Michael Hwang Lisa Revel, Dhrumin Desai, Tom Bednarski, Renu Salver, Rohith Doraiswamy Konoor ONAP Platform & Lab Ericsson T-Lab Configurations & Support Rich Wright, Spondon Dey, Nicolas Hu Carol Nelson Venkatesh Mannepalli, Satish Male, Zhaoji Huang ATT George Sultana ATT Jack Murray Ericsson 8

Animation Mode Acumos-ONAP DCAE Integration Use Case Demo Pre-demo setup Phase 1 demo (11/12) Phase 2 demo (*TBD) ONAP Lab µS Developer * Pending JDA 5 Acumos-R Federation Ericsson Catalog 2 B 1 ML Model AT&T Catalog µS Devel. Catalog (CLI-DB) 4 E 5 Develop v. MME Data Analytic ML Use Case 1 ML Developer Train ML with Refine production data Data Lake Production raw FM/PM Data DCAE Designer GUI DCAE µS Onboarding APIs DCAE Design Catalog B 1 µS ONAP DCAE µS Adapter ML Onboarding 3 6 Service Assurance Flow Design VES Collector µS v. MME ETL µS B 1 µS ML Report µS µS Design Time Configuration B 1 ML Model Service Designer ETL µS Report µS Create DCAE Model Runner Wrapper Generate Spec. json DCAE µS Data. Format. Metadata json Generate µS Image DCAE µS (K 8 s) Docker Image µS Deployment Configuration Submit µS for Blueprint Generation 7 Blue. Print Generator Operations Cloudify BP 8 10 Analytics Report Docker Central (Nexus) Report µS Traffic Generator v. MME VES Collector Pre-instantiated ETL µS DCAE Runtime B 1 µS (ML) 9 OOM Dashboard µS Deployment DCAE Inventory Deploy µS 9