AI Driven Orchestration, Challenges & Opportunities. Openstack Summit 2018 Sana Tariq (Ph.D.) TELUS Communication

Transcription

1 AI Driven Orchestration, Challenges & Opportunities Openstack Summit 2018 Sana Tariq (Ph.D.) TELUS Communication

2 Agenda Service Orchestration Journey Service Orchestration Operational Challenges Closed Loop Orchestration and Dynamic Policy AI/ML Driven Orchestration

3 NFV and Orchestration Journey 5G IoT customers defined services through customized user portals Automated Assurance AI driven/managed operations, capacity and applications AI driven/managed services (advanced) Applications Onboarding PNFs Building NFV Cloud Q IoT and OTT Services/ User-defined Services/International customers Increased Maturity OSS/BSS interlock, evolution of customer Portals, inventory compliance Increased Maturity of Catalogs/Templates/Blueprints to deliver software defined services

4 NFV Telco Cloud - is different Reliability Self serve APIs Automation NFV Telco Cloud Fast and agile Scalability High Throughput Low Latency 24x7 availability Cost efficient Secure Standards based

5 NFV and Orchestration Journey Building robust cloud infrastructure Virtualizing Applications Provisioning workflows Assurance Automated scaling Traffic steered through SDN network Monitoring through holistic service assurance OSS/BSS EMS VNFs Ticketing Alarms Billing Analytics EMS 1 EMS 2 EMS n VNF 1 VNF 2 VNF 3 VNF n VNFMs End to End Orchestrator SDN Controllers DC WAN Service Assurance Integration with OSS/BSS E2E Service Orchestration plays are major orchestration role NFVI VIM WAN

6 Orchestration: Commercial or Open Source Commercial Opensource Vendor lock-in/proprietary plugins Vendor agnostic/shared community plugins Higher licensing cost Significantly lower costs Trusted support model Self/Community support R&D driven roadmaps Community driver roadmaps Depends on Company size Depends on community participation Opensource SI Vendors

7 Orchestration Functional Elements

8 Cloud Robust Cloud supporting automation features, APIs, elasticity etc. SDN Network Complete softwarization of DC and WAN for on-demand creation of services SERVICE ORCHESTRATION SUCCESS Analytics Robust analytics cross-functional domains for scaling, healing & optimization for cloud and services DevOps Effective DevOps culture and support for short time-to-market and efficiency targets Data Models Homogeneity across stacks for consistent data models for seamless integration, reusability and abstraction

9 Chaos of Multi-Vendor Multi-Domain Customer facing services NaaS/OTT Voice OTT 5G IoT VoD vepc S-GW P-GW MME... E2E Service Orchestration SBC S-CSCF vims... SD-WAN TAS I-CSCF Enterprise NFV Cloud

10 Software Defined Service Operational Challenges 1 Service Design Debug 2 Orchestration (testing in a sandbox) 6 Troubleshooting 3 Package and distribution Launch 4 5 Orchestration (runtime) Service failure Assurance Offline in lab Services Operation in real-time

11 Closed Loop Orchestration Static E2E Service Orchestration Service Orchestration VNFM Policy Engine BigData/Hadoop Analytics LB vfw vfw VAS Collector

12 Closed Loop Orchestration AI/ML Predictive/proactive Inputs BigData/ Hadoop Service Orchestrator Analytics Service Definition Workflows Workloads Onboarding Testing & Validation Life Cycle Management Inventory Resource Orchestrator APIs Adaptors SDK Policy Reactive Inputs Mapped Actions Conditional Rules Cloud Services Security Network

13 What? WHY? HOW? AI/ML Orchestration Data Surge Static Policies Dynamic Policy Analytics Model 5G Services/ IoT increase in traffic volume and patterns Reactive decisions False Spikes Too many policies & conflicts Strategy to develop dynamic policies and testing Robust Analytics framework for feeding AI/ML systems Customer Experience Cloud Optimization Network Optimization Security Closer to the edge Service Healing Service optimization Differentiated QoS Energy optimization Capacity optimization Faults prediction & healing Fast troubleshooting Traffic optimization SDN controller QoS based routing Proactive/predictive threat identification Closed loop decisions to attacks mitigation

14 Building an AI Application STEP 1 Problem definition STEP 3 Prepping the data: cleansing, normalization, feature engineering STEP 5 Testing phase: Experiment with Models, Pick best Models and create feedback Loop RESULT? START PROCESS STEP 2 Gather real, sizeable training data STEP 4 Training phase: robust training environment STEP 6 Evaluation, predictor improvement and re-training

15 AI is becoming easier Add your own description here. Add instructions, or additional information about this slide here. You can delete this item or any item on the slide. Raw AI/ML ML Libraries AI Projects Acumos Data Scientists needed, higher complexity Python libraries lower complexity ONAP leverages SPARK, Mlib, MALLET, WEKA etc. H2O, TensorFlow Predictors & Models developed by shared community effort + data Accessible Data Computing Power Better Analytics Engines, Better data organization Large volumes of managed Data Low Cost High Power GPU, NPU availability

16 ONAP Intelligent Closed Loop Architecture Design Time Run-time CLAMP (Closed Loop Automation and Management Platform) GUI that generates TOSCA SDC (Service Design Creation) AI/ML (Python Library) Hadoop DCAE Policy Execution Synthetic Data (Extension) Ceilometer collector Policy1 DMAAP POLICY EXECUTION ENGINE Service Orchestration SDN-Controller APP-C (GVNFM) Condition Cloud Region1 CPU Utilization >80% Action Shift VNF/ workload from Cloud Region 1 to Cloud Region2 TOSCA Policy1 Ceilometer VNF1 VNF2 Ceilometer Cloud Region 1 Cloud Region 2

17 ONAP Policy Execution Flow 5 2 CLAMP Analytics AI MS DCAE Policy MS 12 POLICY EXECUTION ENGINE 1 3 SDC 4 8 Hadoop Cluster 7 Ceilometer Collector DMAAP Service Orchestration SDN- Controller APP-C (GVNFM) 15 6 Cloud Region 1 Ceilometer Cloud Region 2

18 AI/ML Orchestration Industry Verticals Study of VNFs performance requirements (MME, S/P-GW, vsbcs etc.) Customer with IoT Traffic (Dependency: Closer to Edge Location) Customer with real-time Voice Traffic Dependency: Performance/ TTR Customer with 5G Services: real time, high bandwidth Dependency: Performance/ TTR Customer with real-on-demand viewing Dependency: Performance/ TTR/ throughput Customer with offline downloading Dependency: Throughput only NSD and VNFD Package description to associate color tags for VNF types to support these use-cases DCAE: Dynamic Policy and Cloud optimization use-cases Customer/Resource Facing Portal API Mediation Service Orchestrator Service Definition Workflows Inventory Workloads Onboarding Resource Orchestrator Testing & Validation APIs Adaptors SDK Inventory Life Cycle Management Policy BigData/ Hadoop/ Analytics Policy Rules Initial Placement Cloud Optimization related actions Customer QoS related actions Develop ML Models for cloud optimization use-cases NFV Cloud Enhance ceilometer & support cloud optimization configurations Edge POD: High Bandwidth, Closer to customer Zone1 PODs Optimized for Efficiency/ throughput Zone2 PODs Optimized for Low Latency/HTTR (realtime traffic) Zone3 PODs Optimized for Low Latency/HTTR (realtime traffic)

19 Sana Tariq (Ph. D.)