Analytics and Machine Learning with
Red Hat Infrastructure
Kyle Bader, Senior Architect
Sean Pryor, AI Developer
Sherard Griffin, Senior Manager, Open Data Hub
BOSTON, 2019
● PROBLEM STATEMENT
○ Multi-tenant data analytics and machine learning
○ Shared data context
○ Sensitive data can’t leave the country, data governance restrictions
● DATA STRUCTURES
○ Shared data context with Ceph
○ Preparing your data
■ Structured data with Hive Metastore*
■ Semi-structured data
■ Data processing jobs
■ Spark
○ AI/ML
■ Features/Labels/other important terms
■ Background on AI and how it works
■ TensorFlow
● DATA PLATFORM ARCHITECTURE
○ Open Data Hub (Spark, Ceph, JupyterHub, TensorFlow)
○ Follow-up slides for them to learn more
■ ISVs
■ ODH
■ Frameworks
■ Other talks, etc.
PROBLEM STATEMENT
ANALYTICS AND ML CHALLENGES
EXPLOSIVE GROWTH
in analytics teams and analytic tools
MULTIPLE TEAMS COMPETING
for use of the same big data resources
CONGESTION
in busy analytic clusters causing frustration
and missed SLAs
HADOOP
SPARK
HIVE
PRESTO
IMPALA
KAFKA
NIFI
TENSORFLOW
PYTORCH
OPTIONS TO ADDRESS CHALLENGES
Get a bigger cluster
for many teams to share
Give each team
own dedicated cluster,
each with copies of
PBs of data
#1 #2
Give teams ability to
spin-up/spin-down
clusters which can
share common data store
#3
MULTI-WORKLOAD TENANCY
SHARED DATA CONTEXT
HIT SERVICE-LEVEL AGREEMENTS
Give teams their own compute clusters.
ELIMINATE IDLE RESOURCES
By right-sizing de-coupled compute and storage.
BUY 10’s OF PBS INSTEAD OF 100’s
Share data sets across clusters instead of duplicating them.
INCREASE AGILITY
With spin-up/spin-down clusters.
HYBRID CLOUD ANALYTICS AND ML
OPERATOR FRAMEWORK
Provides a managed service like experience
STATEFUL STORAGE SERVICES
Object, block, and file interfaces
DEVICE PLUGIN
GPU acceleration
LOCAL PVS
High performance scratch storage
DATA STRUCTURES
CLEANING AND CONFORMING
SEMI-STRUCTURED DATA
● Sources
○ Stateless applications
○ Sensors
● Common formats
○ CSV, JSON, XML
○ ORC, Avro, Parquet
DATA PROCESSING
● Variety of sources and formats
● Schema detection
● Distributed streaming and batch ETL
STRUCTURED DATA
● Cataloged into databases and tables
● External locations map to object URIs
● Table and column statistics
Select
Model
Select
Features
Model
Training
Model
Evaluation
Model
Tuning
Trained
Models
Model
Serving
&
Scoring
Keras
Microsoft
Cognitive
Toolkit
Horovod
MODELING AND SERVING
DATA PLATFORM
ARCHITECTURE
ARCHITECTURE
ARCHITECTURE
ARCHITECTURE
ARCHITECTURE
ARCHITECTURE
OPEN DATA HUB
Collaborate on a Data & AI platform for the Hybrid Cloud
● Open source community for AI-as-a-service platform
● Cloud-agnostic - AI for the Hybrid Cloud
● No cloud vendor lock-in
● OpenDataHub.io
Sentiment analysis and entity detection
on customer engagements, support
tickets, marketing surveys and more.
Trained on the specific Red Hat product
terminology.
AWS Microsoft AzureOpenStackDatacenterLaptop
CONTAINERIZER APPS
AT RED HAT’S CORE PROCESSES
Internal Use Cases
AWS Microsoft AzureOpenStackDatacenterLaptop
CONTAINERIZER APPS
AT RED HAT’S CORE PROCESSES
Internal Use Cases
Improve Red Hat’s core Engineering and
Operations processes by applying
analytics, machine learning, and AI.
AWS Microsoft AzureOpenStackDatacenterLaptop
CONTAINERIZER APPS
- rules
- heuristics
- ML
CORE DEPLOYMENT
● Container platform
● Certified Kubernetes
● Hybrid cloud
● Unified, distributed
storage
● RESTful gateway
● S3 and Swift compatible
● Radanalytics.io
community
● Unified analytics
engine
● Large-scale data
● Runs on Kubernetes
● Multi-user Jupyter
● Used for data science
and research
Available Now at OpenDataHub.io
Add-Ons
● Part of Open Data Hub
● Set of deployed
pre-defined AI models
available to use
● Monitoring and alerting
toolkit
● Records numeric time
series data
● Used to diagnose
problems
● Analytics platform for
all metrics
● Query, visualize and
alert on metrics
● Deploying machine
learning models on
Kubernetes
● Expose models via
REST and gRPC
● Full model lifecycle
management
Available Now at OpenDataHub.io
Open Data Hub
AI Library
RUNNING AT RED HAT
PLANNED RELEASES
Highlights
July
2019
Data Engineering Additions
- Cloudera Hue deployment
- Spark SQL Thrift Server deployment
- Argo deployment
- MLFlow deployment
- Kubeflow integration
- Kafka (Strimzi) deployment
- Seldon-core deployment
October
2019
To be determined
January
2019
Version 0.1 - Initial ODH Release
- OCP 3.10 and 3.11 support
- JupyterHub + Spark + Ceph-nano
deployment
April
2019
Operator Support + Monitoring
- OCP 4.0+ support
- Open Data Hub operator
- AI Library
- Rook for Ceph deployment
- TwoSigma BeakerX integration
- JupyterHub with GPU support
- Prometheus deployment with Spark
monitoring
AI AND MACHINE LEARNING
IN THIS LAB
AI IN THIS LAB
WHAT NEXT?
● Try Open Data Hub yourself!
○ https://try.openshift.com
○ https://gitlab.com/opendatahub/opendatahub-operator
● Building the Next Generation of Innovation Together
○ Thursday at 8:30 AM
● Kaleidoscope of Innovation: AI and Machine Learning on
OpenShift
○ Part 1: Thursday at 2:00 PM
○ Part 2: Thursday at 3:15 PM
Red Hat data analytics infrastructure solution
red.ht/videos-RHDAIS
MACHINE LEARNING CYCLE
Ingest Prepare Preprocess Discover Develop Train Test Deploy
MKL-DNN
cuDNN

Red hat infrastructure for analytics

  • 1.
    Analytics and MachineLearning with Red Hat Infrastructure Kyle Bader, Senior Architect Sean Pryor, AI Developer Sherard Griffin, Senior Manager, Open Data Hub BOSTON, 2019
  • 2.
    ● PROBLEM STATEMENT ○Multi-tenant data analytics and machine learning ○ Shared data context ○ Sensitive data can’t leave the country, data governance restrictions ● DATA STRUCTURES ○ Shared data context with Ceph ○ Preparing your data ■ Structured data with Hive Metastore* ■ Semi-structured data ■ Data processing jobs ■ Spark ○ AI/ML ■ Features/Labels/other important terms ■ Background on AI and how it works ■ TensorFlow ● DATA PLATFORM ARCHITECTURE ○ Open Data Hub (Spark, Ceph, JupyterHub, TensorFlow) ○ Follow-up slides for them to learn more ■ ISVs ■ ODH ■ Frameworks ■ Other talks, etc.
  • 3.
  • 4.
    ANALYTICS AND MLCHALLENGES EXPLOSIVE GROWTH in analytics teams and analytic tools MULTIPLE TEAMS COMPETING for use of the same big data resources CONGESTION in busy analytic clusters causing frustration and missed SLAs HADOOP SPARK HIVE PRESTO IMPALA KAFKA NIFI TENSORFLOW PYTORCH
  • 5.
    OPTIONS TO ADDRESSCHALLENGES Get a bigger cluster for many teams to share Give each team own dedicated cluster, each with copies of PBs of data #1 #2 Give teams ability to spin-up/spin-down clusters which can share common data store #3
  • 6.
    MULTI-WORKLOAD TENANCY SHARED DATACONTEXT HIT SERVICE-LEVEL AGREEMENTS Give teams their own compute clusters. ELIMINATE IDLE RESOURCES By right-sizing de-coupled compute and storage. BUY 10’s OF PBS INSTEAD OF 100’s Share data sets across clusters instead of duplicating them. INCREASE AGILITY With spin-up/spin-down clusters.
  • 7.
    HYBRID CLOUD ANALYTICSAND ML OPERATOR FRAMEWORK Provides a managed service like experience STATEFUL STORAGE SERVICES Object, block, and file interfaces DEVICE PLUGIN GPU acceleration LOCAL PVS High performance scratch storage
  • 8.
  • 9.
  • 10.
    SEMI-STRUCTURED DATA ● Sources ○Stateless applications ○ Sensors ● Common formats ○ CSV, JSON, XML ○ ORC, Avro, Parquet
  • 11.
    DATA PROCESSING ● Varietyof sources and formats ● Schema detection ● Distributed streaming and batch ETL
  • 12.
    STRUCTURED DATA ● Catalogedinto databases and tables ● External locations map to object URIs ● Table and column statistics
  • 13.
  • 14.
  • 15.
  • 16.
  • 17.
  • 18.
  • 19.
  • 20.
    OPEN DATA HUB Collaborateon a Data & AI platform for the Hybrid Cloud ● Open source community for AI-as-a-service platform ● Cloud-agnostic - AI for the Hybrid Cloud ● No cloud vendor lock-in ● OpenDataHub.io
  • 21.
    Sentiment analysis andentity detection on customer engagements, support tickets, marketing surveys and more. Trained on the specific Red Hat product terminology. AWS Microsoft AzureOpenStackDatacenterLaptop CONTAINERIZER APPS AT RED HAT’S CORE PROCESSES Internal Use Cases
  • 22.
    AWS Microsoft AzureOpenStackDatacenterLaptop CONTAINERIZERAPPS AT RED HAT’S CORE PROCESSES Internal Use Cases Improve Red Hat’s core Engineering and Operations processes by applying analytics, machine learning, and AI. AWS Microsoft AzureOpenStackDatacenterLaptop CONTAINERIZER APPS - rules - heuristics - ML
  • 23.
    CORE DEPLOYMENT ● Containerplatform ● Certified Kubernetes ● Hybrid cloud ● Unified, distributed storage ● RESTful gateway ● S3 and Swift compatible ● Radanalytics.io community ● Unified analytics engine ● Large-scale data ● Runs on Kubernetes ● Multi-user Jupyter ● Used for data science and research Available Now at OpenDataHub.io
  • 24.
    Add-Ons ● Part ofOpen Data Hub ● Set of deployed pre-defined AI models available to use ● Monitoring and alerting toolkit ● Records numeric time series data ● Used to diagnose problems ● Analytics platform for all metrics ● Query, visualize and alert on metrics ● Deploying machine learning models on Kubernetes ● Expose models via REST and gRPC ● Full model lifecycle management Available Now at OpenDataHub.io Open Data Hub AI Library
  • 25.
  • 26.
    PLANNED RELEASES Highlights July 2019 Data EngineeringAdditions - Cloudera Hue deployment - Spark SQL Thrift Server deployment - Argo deployment - MLFlow deployment - Kubeflow integration - Kafka (Strimzi) deployment - Seldon-core deployment October 2019 To be determined January 2019 Version 0.1 - Initial ODH Release - OCP 3.10 and 3.11 support - JupyterHub + Spark + Ceph-nano deployment April 2019 Operator Support + Monitoring - OCP 4.0+ support - Open Data Hub operator - AI Library - Rook for Ceph deployment - TwoSigma BeakerX integration - JupyterHub with GPU support - Prometheus deployment with Spark monitoring
  • 27.
    AI AND MACHINELEARNING IN THIS LAB
  • 28.
  • 29.
    WHAT NEXT? ● TryOpen Data Hub yourself! ○ https://try.openshift.com ○ https://gitlab.com/opendatahub/opendatahub-operator ● Building the Next Generation of Innovation Together ○ Thursday at 8:30 AM ● Kaleidoscope of Innovation: AI and Machine Learning on OpenShift ○ Part 1: Thursday at 2:00 PM ○ Part 2: Thursday at 3:15 PM Red Hat data analytics infrastructure solution red.ht/videos-RHDAIS
  • 32.
    MACHINE LEARNING CYCLE IngestPrepare Preprocess Discover Develop Train Test Deploy MKL-DNN cuDNN