Lego-like building blocks of Storm and Spark Streaming Pipelines

15
Lego-Like Building Blocks of Storm and Spark Streaming Pipelines For Rapid IOT and Streaming Analytics App Development Speakers: Anand Venugopal, Punit Shah

Transcript of Lego-like building blocks of Storm and Spark Streaming Pipelines

Page 1: Lego-like building blocks of Storm and Spark Streaming Pipelines

Lego-Like Building Blocks of Storm and Spark Streaming Pipelines

For Rapid IOT and Streaming Analytics App DevelopmentSpeakers: Anand Venugopal, Punit Shah

Page 2: Lego-like building blocks of Storm and Spark Streaming Pipelines

Approach to this presentation

• Sharing our learnings and best practices from various Streaming Implementations

• Fairly simple concept - certainly not rocket science – but we do hope there may be some interesting ideas for you.

• Illustrating using a specific tool but you are free to implement the same concepts anyway you like

Page 3: Lego-like building blocks of Storm and Spark Streaming Pipelines

IOT and Streaming Analytics is HOT

30-50B DevicesUSD 661.74 Billion

Page 4: Lego-like building blocks of Storm and Spark Streaming Pipelines

Use Cases for Streaming Analytics

• Store, Warehouse operations – Retail• Predictive Maintenance – Manufacturing, Oil & Gas • Clinical Care and Patient Management – Healthcare - Clinical• Sensor Analytics – IOT, Manufacturing, Others• Fleet Operations – Transportation, Logistics• Fraud and Anomaly Detection – IT Security, Financial Services• Gaming Analytics – Entertainment, Gaming• Churn Analytics – Telecom, Banking, Retail• Network Traffic Analysis and Optimization – Telco• Internet Advertising – Retail, e-commerce

VERT ICALS

Page 5: Lego-like building blocks of Storm and Spark Streaming Pipelines

Use Cases for Streaming Analytics

HORIZONTALS

• Customer Experience• Clickstream Analytics• Context-sensitive Offers And Recommendations• IT Log Analytics• Security • Business Activity Monitoring

Page 6: Lego-like building blocks of Storm and Spark Streaming Pipelines

Use Cases for Streaming AnalyticsCOMBO

• Internet of Things• Mobile App Analytics• Call Center Monitoring and Analytics

Page 7: Lego-like building blocks of Storm and Spark Streaming Pipelines

Adoption Pattern of IOT and Streaming Analytics

Department 1 Department 2Department 3Department 4

Page 8: Lego-like building blocks of Storm and Spark Streaming Pipelines

Adoption Pattern of IOT and Streaming Analytics

Department 1 Department 2 Department 3Department 4

Page 9: Lego-like building blocks of Storm and Spark Streaming Pipelines

With Scale – we need a centralized efficient approach

Department 1 Department 2 Department 3Department 4

CENTRALIZED APPROACH

• Unified multi-tenant visual platform• Collaborative re-use of components

Page 10: Lego-like building blocks of Storm and Spark Streaming Pipelines

Three levels of re-use

FunctionsE.g. ETL functions (Date/ String/ Object/ Integer manipulations)

OperatorsE.g. Kafka Channel, Write_to_HDFS, Time-based aggregation;

Pipelinesi.e. Highest level of abstraction – lego-like building blocks

Page 11: Lego-like building blocks of Storm and Spark Streaming Pipelines

Re-usable stream processing patterns as pipelinesIngest – Pre-processing, Cleanup, De-duplication, re-sequencing; Filters,

Classification/ Routing - Pass on instantly to different downstream subscribers –

Instant anomaly detection – Security breaches / Fraud/ Costly failure scenarios

Rules based alerting - Customer setup rules- Notifications and triggers

Enrichment – Get key fields from the stream – dip into one or more Master DBs; create aggregate record

Time Window calculations - counters, statistics

Visualization block of raw and derived data

Data storage – a) Batch up and write data into HDFS/ HBASE etc. b) Instantly write data into an indexing store

Specific predictive model blocks

Page 12: Lego-like building blocks of Storm and Spark Streaming Pipelines

Connect pre-built pipelines to build an appIngest/ Filter/ Classify

Anomaly DetectionAlerting

Action Triggers

Index, Visualize

Time Window Statistics

Persist, Visualize

Low Latency Engine

Low Latency Engine

Low Latency Engine

Micro-batch engine Micro-batch engine

Rapid DevelopmentBest Engine for the task

Dynamic Routing

Page 13: Lego-like building blocks of Storm and Spark Streaming Pipelines

A-B Testing, Champion Challenger, Hot SwapIngest/ Filter/ Classify

MODEL 1

MODEL 2UI CONFIGURABLE DYNAMIC ROUTING RULES

Page 14: Lego-like building blocks of Storm and Spark Streaming Pipelines

DEMO

Page 15: Lego-like building blocks of Storm and Spark Streaming Pipelines

Thank you

[email protected]