Advanced data science algorithms applied to scalable stream processing by David Piris and Ignacio García

Advanced data science algorithms applied to scalable stream processing

David Piris Valenzuela

Nacho García Fernández

[email protected]

@0xNacho

[email protected]

@davidpiris

mailto:[email protected]


3

About Treelogic R&D intensive company with the mission of adapting technological knowledge to

improve quality standards in our daily life

8 ongoing H2020 projects (coordinating 3 of them)

8 ongoing FP7 projects (coordinating 5 of them)

Focused on providing Big Data Analytics in all the world

Internal organizationResearch lines

Big Data

Computer vision

Data science

Social Media Analysis

Security

ICT solutions

Security & Safety

Justice

Health

Transport

Financial Services

ICT tailored solutions

CONTENTS

1. WHY WE NEED BIG DATA2. BIG DATA: SOLUTIONS3. BIG DATA: REAL-TIME PROCESSING4. INCREMENTAL ALGORITHMS5. WHAT WE WANT6. WHAT WE NEED

1. A stream processing engine

2. Online incremental algorithms

3. A distributed data storage system

4. A use case

5. A visualization layer

CONTENTS





4. A use case


6

Why we need Big Data

7

Why we need Big Data Public and private sector companies store a huge mount of data

Countries with huge databases store data from Population

Medical records

Taxes

Online transactions

Mobile transactions

Social Networks

In a single day, tweets generates 12 TB!!

8

Why we need Big Data

2.5 Exabytes are produced every day!!!

530.000.000 million songs

150.000.000 iPhones

5 million laptops

90 years of HD Video

9

Why we need Big DataHow can we manage all data?

CONTENTS





4. A use case


11

Big Data: SolutionsFirst we can manage all historical repository, and retrieve some value fromdata stored

Batch architecture

MapReduce

Hadoop Ecosystem

12

Big Data: Solutions

13

Big Data: SolutionsBatch processing with Hadoop takes a lot of time and the need to processingested data and display results in a shortest way possible brings newarchitecture and tools

Lambda architecture

Spark (memory vs disk)

14

Big Data: Solutions

CONTENTS





4. A use case


16

Big data: real-time processing Faster results

Accurate results

Less expense

Please consumers

17

Big data: real-time processingAs previously said, we need to extract and visualize information in near realtime…

18

Big data: real-time processing Flink as engine process

Stream processing

Windowing with events time semantics

Streaming and batch processing

19

Big data: real-time processingKappa architecture

Batch layer removed

Only one set of code needs to be maintained

20

Big data: real-time processing No need to use batch layer

Avoid use disk in engine process (latency)

CONTENTS





4. A use case


22

Big data: available tools

23

Incremental algorithms BI & BA people always want to made some common operations to retrieve

value and visualize data We have operational tools in a relational or batch environment How we can obtain average for a data stream that is changing every

second, minutes or even milliseconds…? Common average operation is indicated for historical repository, data input

without any changes in the moment we start the process to obtain it. Do we have tools to make it possible in a real time deployment?

24

Incremental algorithms

Answer is NO!

25

Incremental algorithmsFlink gives us the chance to operate with a new window processing concept.We can decide and configure "small time pieces", and make someoperations or manipulate data in that time space.

26

Incremental algorithmsWith Flink and windowing…

27

Incremental algorithms These algorithms consume streams of data and are able to update their

results in a parallel manner without the need of saving the processed data

Using checkpoints in windowing, allows us to store result from previouswindow process

28

Incremental algorithmsOur analytics & visualization solution implemented in a real time architecture

29

Incremental algorithmsIf you are a BI or BA professional...we care about you!

30

Incremental algorithms Currently, we have implemented:

Average

Mode

Variance

Correlation

Covariance

Min

Max

31

Incremental algorithms Currently we are working on:

Median

32

Incremental algorithms In roadmap…

Standard deviation

Order by

Discretization

Contains

Split

Validate range values

Set default value to specific output

CONTENTS

1. WHY WE NEED BIG DATA2. BIG DATA: SOLUTIONS3. BIG DATA: REAL-TIME PROCESSING4. INCREMENTAL ALGORITHMS5. WHAT WE NEED

1. A stream processing engine2. Online incremental algorithms


4. A use case


34

Apache Flink vs Apache Spark

Pure streams for all workloads

Optimizer

Low latency, high throughput

Global, session, time and count based

window criteria

Provides automatic memory management

Micro-batches for all workloads

No job optimizer

High latency as compared to Flink

Time-based window criteria

Configurable memory management. Spark

1.6+ has move towards automating

memory management

35

CONTENTS



2. Online incremental algorithms3. A distributed data storage system

4. A use case


37

Incremental algorithms in Flink

38

Incremental algorithms in Flink Default behavior in Apache Flink:

With incremental algorithms:

39

Incremental algorithms in Flink

CONTENTS




3. A distributed data storage system4. A use case


41

Apache Kudu Provides a combination of fast inserts / updates and efficient columnar

scans to enable real-time analytic workloads

It is a new complements to HDFS and HBase

Designed for use cases that require fast analytics on fast data

Low query latency

V1.0.1 was released on October 11, 2016

CONTENTS





4. A use case5. A visualization layer

43

PROTEUS: a steel making scenario Steel industry is a key sector for the European community.

PROTEUS was introduced last year at Big Data Spain by Treelogic *

Hot Strip mills (sometimes) produces steel with defects

Predict coil parameters (thickness, width, flatness) using real-time and historical data

Detecting defective coils in an early stage saves money. The production process can bemodified / stopped.

Proposed architecture is being validated in this project

7870 variables with a frequency of 500ms: data-in-motion

700.000 registers for each variables. 500GB time series and flatness map: data-at-rest

* https://www.youtube.com/watch?v=EIH7HLyqhfE

44

PROTEUS: a steel-making scenario Steel industry is a key sector for the European community.

PROTEUS was introduced last year at Big Data Spain by Treelogic *

Hot Strip mills (sometimes) produces steel with defects

Predict coil parameters (thickness, width, flatness) using real-time and historical data

Detecting defective coils in an early stage saves money. The production process can bemodified / stopped.

Proposed architecture is being validated in this project

7870 variables with a frequency of 500ms: data-in-motion

700.000 registers for each variables. 500GB time series and flatness map: data-at-rest

* https://www.youtube.com/watch?v=EIH7HLyqhfE

CONTENTS





4. A use case


46

Websockets Websocket is a computer communication protocol providing full-duplex

communication channels over a single TCP connection.

Extremely faster than HTTP

Its API is standardized by the W3C

47

Apache Flink & Websockets Data sinks consume DataSets and are used to store or return them.

Flink comes with a variety of built-in output formats that are encapsulated behindoperations on the DataSet: writeAsText()

writeAsFormattedText()

writeAsCsv()

print()

write()

We’ve developed a WebsocketSink enabling Flink to send outputs to a givenwebsocket endpoint. Based on the javax-websocket-client-api 1.1 spec.

48

Incremental architecture: our approach

49

50

ProteicJS

https://github.com/proteus-h2020/proteic/

https://github.com/proteus-h2020/proteic/tree/development

51

ProteicJS: Visualizations

52

ProteicJS: Researching on visualization Currently researching on new ways of visualizing data and ML models

53

ProteicJS & Apache Flink

54

How to get it all

https://github.com/proteus-h2020/proteus-docker

https://github.com/proteus-h2020/proteus-docker

Advanced data science algorithms applied to scalable stream processing

David Piris Valenzuela

Nacho García Fernández

[email protected]

@0xNacho

[email protected]

@davidpiris



Advanced data science algorithms applied to scalable stream processing by David Piris and Ignacio García

Technology