TUT1151 Simplifying AI Applications with containers and Kubernetes - PowerPoint PPT Presentation

TUT1151 Simplifying AI Applications with containers and Kubernetes Glyn Bowden, Chief Architect, AI & Data Science Practice SUSECON 2019

The world is replacing programming with training 23 million developers worldwide 4 in 10 companies mention lack of analytical skills Handful of as a key challenge data scientists By 2020, 50% of organizations will lack sufficient AI and data literacy skills to achieve business value 2

What you need to do defines what you have to do How do you prepare and integrate How long do you have data for advanced analytics ? to take action? Edge Cloud Where is the data What are your generated? business goals? What do you have to What governance and do to put that data in What does that security regulations do a form you can use? data consist of? you need to comply with? 3

AI ecosystem A working solution requires every layer of this stack – Advisory / consulting services Expertise – Vertical SMEs – Public data sources – Data labeling providers Data – Proprietary data collections – Custom applications – Machine learning and deep learning frameworks and libraries Software (TensorFlow, Caffe, Scikit-learn, …), data platforms – As-a-service offerings – Systems software and libraries (CUBLAS, MKL, cuDNN, MKL-DNN) – Memory – Network Hardware – Storage – Hardware accelerators 4 4

Challenges to gaining insight Existing AI process Individual – AI solutions tend to be bespoke. Designed and deployed as part of the project Time consuming – AI solution cycle is long due to bespoke nature of design Risk – Bespoke solutions and iterative platform development introduces risk of inconsistent results and even project failure Inefficient – Skilled and costly resources used on tasks which could easily be done by others 5

The Four Pillars of Data Science “simplify AI consumption bringing together the AI ecosystem” Data Sources Data Management Analytics Insight – Multiple formats – Data Platform – Multiple Machine – Business Use – Multiple – Data Lake Learning Libraries Cases Sources – On Premise – Multiple AI trained – Integrated Analysis – Multiple – Hybrid Cloud models – 360 view of data Standards Integration – Multiple AI – CDO Dashboards – Security languages 6

AI Ecosystem Summary 14 49 15 Distinct ML Services ML Model Candidates Distinct ML Services Open Source Caffe2 Modelzoo 20+ 50+ 40+ AI / ML Candidate Partners Language Frameworks Model Categories 7

AI – Vertical Solution Use Case Examples Batch or Real-time Real-time manual data feed execution Finance Manufacturing Healthcare Telco / Media Enterprise Algorithmic Material Genome Customer Supply chain Trading Analysis Sequencing Profiling optimisation Trade Activity & Directed Order Market data Simulation Conclusion Simulation Conclusion Simulation Execution social analysis advertising execution Network Predictive Resource Image / Video Performance Risk Analysis Maintenance Management Recognition Analysis Sensor Sensor Config. Decision Simulation Notification Resource data Proposal Image data Classification analysis Execution Analysis execution Natural Self-Functioning Connected Fraud Detection SmartCity language Devices Health comms. Environ. Task Patient Medical Log analysis Notification Financial data Notification Text analysis Chat comms Analysis execution sensors opinion 8

Layered Architecture Business Users Task Execution Engine Visualisation Engine Models and Algorithm (Content) Data Scientists Developer Frameworks Big Data Frameworks AI Frameworks SUSE CAP Unified Data Warehouse On-Site DWH Container Platform Cloud DWH IT Operators SUSE Enterprise Storage SUSE CaaSP Infrastructure Cloud Cloud Technology Partners Hybrid IT Practice 9 HPE Internal Use Only

Accelerate time-to-value in your AI journey with OneAI Rapid deployment of solution components during PoV leading to, smooth production Conventional steps for a PoV / Production implementation Automated setup with OneAI Provisioning of Infrastructure Automated Provisioning Setup AI s/w stack Containerization process Containerize Messaging and transformation components Easy Monitoring Monitoring tools Elapsed Time: Weeks to Months Elapsed Time: Hours Faster PoV  Getting proof points early and fail fast ! Automation deployment of infrastructure, Monitoring tools and Management: simple

How Project OneAI works – an illustrative view Templates OneAI Automated Provisioning Use cased based output UseCase Parameters of Container based ecosystem • Use Case parameters Use Case • Data sources Orchestrated w/ • Dashboard Kubernetes parameters • Accounts / Roles Solution • Monitoring parameters Architecture Unified monitoring on open frameworks (Prometheus) 11

Addressing the challenges to Insights Existing AI process HPE Pointnext Framework Individual – AI solutions tend to be bespoke. Reuse – Leverage HPE globally industrialized Designed and deployed as part of the project templates and project IP Time consuming – AI Solution cycle is long Rapid – Deploy an AI solution, so it is ready to due to bespoke nature of design start receiving data Learn & Evolve – Solution gallery will grow Risk – Bespoke solutions and iterative platform development introduces risk of inconsistent over time as more AI projects are successfully results and even project failure delivered Efficient – Consultants are driving insight from Inefficient – Skilled and costly resources used the customer data not configuring and on tasks which could easily be done by others experimenting with underlying technologies 12

Project OneAI Functional Architecture v2.0 OneAI Microservices hub.docker.com catalog Keycloack OAUTH2/OIDC Ambassador API Gateway environment Kubernetes Cluster HPE Pointnext Docker Registry Istio Service Mesh {{api_client}} Content datasource oai (cli) Private Docker Registry lifecycle operations NVIDIA NGC Web browser Grommet UI 13

Service Scaling 14

Cluster Scaling 15

Component Hierarchy Design 16

17

18

19

20

Thank You gjb@hpe.com 21