Fast Cars, Big Data - How Streaming Can Help Formula 1

© 2016 MapR Technologies© 2016 MapR TechnologiesMapR Confidential © 2016 MapR Technologies1

Fast Cars, Big Data How Streaming Can Help Formula 1

Tugdual Grall@tgrall

© 2016 MapR Technologies© 2016 MapR Technologies@tgrall

{“about” : “me”}Tugdual “Tug” Grall • MapR

• Technical Evangelist • MongoDB

• Technical Evangelist • Couchbase

• Technical Evangelist • eXo

• CTO • Oracle

• Developer/Product Manager • Mainly Java/SOA

• Developer in consulting firms

• Web • @tgrall • http://tgrall.github.io • tgrall

• NantesJUG co-founder

• Pet Project : • http://www.resultri.com

• [email protected] • [email protected]

mailto:[email protected]


© 2016 MapR Technologies 3

Agenda• What’s the point of data in motorsports? • Live demo • Architecture • What’s next?


How data plays in F1 motorsports



Data in Motorsports

http

://f1

fram

ewor

k.bl

ogsp

ot.d

e/20

13/0

8/sh

ort-g

uide

-to-f1

-tele

met

ry-s

pa-c

ircui

t.htm

l


Difference is due to later and sharper braking


Real Analytics as Well as Visualization• Inputs

• Predictive analysis of consumables and tires • Physical models of car + driver performance

• Tire wear slows lap times, lower fuel weight speeds lap times • Competitors’ options • Weather conditions • Current GP points status

• Outputs • Tactical options, outcome distributions


Data for Marketing as well

http

://fo

rmul

a1.fe

rrar

i.com

/en/

info

raci

ng-h

unga

rian-

gp-2

015/


Some Examples?


• Up to 300 sensors per car

Some Examples?


• Up to 300 sensors per car• Up to 2000 channels

Some Examples?


• Up to 300 sensors per car• Up to 2000 channels• Sensor data are sent to the paddock in 2ms

Some Examples?


• Up to 300 sensors per car• Up to 2000 channels• Sensor data are sent to the paddock in 2ms• 1.5 billions of data points for a race

Some Examples?


• Up to 300 sensors per car• Up to 2000 channels• Sensor data are sent to the paddock in 2ms• 1.5 billions of data points for a race• 5 billions for a full race weekend

Some Examples?


• Up to 300 sensors per car• Up to 2000 channels• Sensor data are sent to the paddock in 2ms• 1.5 billions of data points for a race• 5 billions for a full race weekend• 5/6Gb of compressed data per car for 90mn

Some Examples?


• Up to 300 sensors per car• Up to 2000 channels• Sensor data are sent to the paddock in 2ms• 1.5 billions of data points for a race• 5 billions for a full race weekend• 5/6Gb of compressed data per car for 90mn

US Grand Prix 2014 : 243 Tb (race teams combined)

Some Examples?


So how does that work? Especially for real-time data?


Production System Outline


Simplified Demo System Outline

Archive MapR DB

Jetty /Bootstrap /

d3

Apache Drill (SQL access)

TORCS race simulator

MapR Streams


TORCS for Cars, Physics and Drivers

TORCS is a pseudo-physics based racing simulator with full graphics output and pluggable control modules.

TORCS is commonly used for AI research, but the control model can just as well collect data


Let’s see it work!

© 2016 MapR Technologies© 2016 MapR Technologies@tgrall 16

IoT : Racing Cars

Producers Consumers

sensors data

Real Time

Analytics

https://github.com/mapr-demos/racing-time-series



IoT : Racing Cars

sensors data


Kafka Producer

(Java)

Kafka Consumer +

OJAI (Java)

Kafka Consumer +

WebSocket (Java + JS)

SQL



Big Datastore

Distributed File SystemHDFS/MapR-FS

NoSQL DatabaseHBase/MapR-DB

….


Store data as File or Row?

HDFS / MapR-FS • Data stores as “files” • Fast with Large Scans • Slow random read/writes

NoSQL (HBase/MapR-DB) • Data stores as row/documents • Fast with random read/writes


Kafka & MapR Streams


What is Kafka?

• http://kafka.apache.org/ • Created at LinkedIn, open sourced in 2011 • Implemented in Scala / Java • Distributed messaging system built to scale


Key Concepts

• Feeds of messages are organised in topics • Processes that publish messages are called producers • Processes that subscribed to topic and process messages are

consumers • A Kafka cluster is made of one or more brokers (== node)


Topics and Partitions

• Split topics into partitions for scalability

0 1 2 3 4 5 6 7 8

0 1 2 3 4 5

0 1 2 3 4 5 6 7

Partition 0

Partition 1

Partition 2

Writes


Consumer Groups

• Single consumer abstraction for scalability • Max 1 consumer per partition • Any number of consumer groups


Produce MessagesProducerRecord<String, byte[]> rec = new ProducerRecord<>( “/stream/car_1_topic“, eventName, value.toString().getBytes());

producer.send(rec, (recordMetadata, e) -> { if (e != null) { … });

producer.flush();


Consume Messageslong pollTimeOut = 800; while(true) { ConsumerRecords<String, String> records = consumer.poll(pollTimeOut); if (!records.isEmpty()) { Iterable<ConsumerRecord<String, String>> iterable = records::iterator; StreamSupport.stream(iterable.spliterator(), false).forEach((record) -> { // work with record object

… record.value();…

}); consumer.commitAsync(); } }


Big Picture

Producer

Producer

Producer

Consumer

Consumer

Consumer


More real life Kafka …

Zookeeper

Broker 1

Topic A Topic B

Broker 2

Topic A Topic B

Broker 3

Topic A Topic B

Producer

Producer

Producer

Consumer

Consumer

Consumer


• Distributed messaging system built to scale • Use Apache Kafka API 0.9.0 • No code change • Does not use the same “broker” architecture

• Log stored in MapR Storage(Scalable, Secured, Fast, Multi DC)

• No Zookeeper


Kafka

Zookeeper

Broker 1

Topic A Topic B

Broker 2

Topic A Topic B

Broker 3

Topic A Topic B

Producer

Producer

Producer

Consumer

Consumer

Consumer


MapR Streams

Stream

Topic A Topic B

Stream

Topic A Topic B

Stream

Topic A Topic B

Producer

Producer

Producer

Consumer

Consumer

Consumer


What’s next?


Sensor Data V1• 3 main data points:

• Speed (m/s) • RPM • Distance (m)

• Buffered

{ "_id":"1.458141858E9/0.324", "car" = "car1", "timestamp":1458141858, "racetime”:0.324, "records": [ { "sensors":{ "Speed":3.588583, "Distance":2003.023071, "RPM":1896.575806 }, "racetime":0.324, "timestamp":1458141858 }, { "sensors":{ "Speed":6.755624, "Distance":2004.084717, "RPM":1673.264526 }, "racetime":0.556, "timestamp":1458141858 },


Sensor Data V2• 3 main data points:

• Speed (m/s) • RPM • Distance (m) • Throttle • Gear • …

• Buffered

{ "_id":"1.458141858E9/0.324", "car" = "car1", "timestamp":1458141858, "racetime”:0.324, "records": [ { "sensors":{ "Speed":3.588583, "Distance":2003.023071, "RPM":1896.575806, "gear" : 2 }, "racetime":0.324, "timestamp":1458141858 }, { "sensors":{ "Speed":6.755624, "Distance":2004.084717, “RPM":1673.264526, "gear" : 2 }, "racetime":0.556, "timestamp":1458141858 },


• It works, is available on github, ASL 2

• Data collected is unrealistically limited, lacks – Tire pressure, temperature x 4 – Brake usage, temperature x 8 – Engine monitoring is primitive (RPMs only, no KERS) – Data rate is fixed, real data comes in at highly variable rates – Real data has variable delays due to RF dropout + buffering

• Data collected is in pure JSON – Real data is columnar compressed blobs


Next Steps• Near Real Time Data Processing

• Aggregation • Machine Learning • Alerts


• Cluster Computing Platform • Extends “MapReduce” with

extensions – Streaming – Interactive Analytics

• Run in Memory • http://spark.apache.org/

http://spark.apache.org/


• Streaming Dataflow Engine • Datastream/Dataset APIs • CEP, Graph, ML

• Run in Memory • https://flink.apache.org/

https://flink.apache.org/


IoT : Racing Cars V2.0

sensors data


Alerts



Spark & Streams val topics = “/app/racing/stream:all_cars" val sparkConf = new SparkConf().setAppName(“SensorStream") val ssc = new StreamingContext(sparkConf, Seconds(2))

// Create direct kafka stream with brokers and topics val topicsSet = topics.split(",").toSet val kafkaParams = Map[String, String]( ConsumerConfig.GROUP_ID_CONFIG -> "race1", ConsumerConfig.KEY_DESERIALIZER_CLASS_CONFIG -> "org.apache.kafka.common.serialization.StringDeserializer", ConsumerConfig.VALUE_DESERIALIZER_CLASS_CONFIG -> "org.apache.kafka.common.serialization.StringDeserializer", ConsumerConfig.AUTO_OFFSET_RESET_CONFIG -> "earliest", ConsumerConfig.ENABLE_AUTO_COMMIT_CONFIG -> "false", "spark.kafka.poll.time" -> "1000" )

val messages = KafkaUtils.createDirectStream[String, String](ssc, kafkaParams, topicsSet) val sensorDStream = messages.map(_._2).map(parseSensor)

sensorDStream.foreachRDD { rdd => // There exists at least one element in RDD if (!rdd.isEmpty) { ….. } }


Streaming Architecture & Formula 1• Stream data in real time

• Big Data Store to deal with the scale • NoSQL Database, Distributed File System

• Decouple the source from the consumer(s) • Dashboard, Analytics, Machine Learning • Add new use case….


Streaming Architecture & Formula 1• Stream data in real time

• Big Data Store to deal with the scale • NoSQL Database, Distributed File System

• Decouple the source from the consumer(s) • Dashboard, Analytics, Machine Learning • Add new use case….

This is not only about Formula 1! (Telco, Finance, Retail, Content, IT)


MapR Converged Data Platform

Open Source Engines & Tools Commercial Engines & Applications

Utility-Grade Platform Services

Dat

aPr

oces

sing

Enterprise StorageMapR-FS MapR-DB MapR Streams

Database Event Streaming

Global Namespace High Availability Data Protection Self-healing Unified Security Real-time Multi-tenancy

Search & Others

Cloud & Managed Services

Custom Apps

Unified M

anagement and M

onitoring


MapR Platform Services: Open API ArchitectureAssures Interoperability, Avoids Lock-in

MapR-FS Enterprise Storage

MapR-DB NoSQL Database

MapR Streams Global Event Streaming

HDFS API

POSIX NFS

SQL, Hbase

APIJSON API

Kafka API



Q & A@mapr | @tgrall maprtech

[email protected]

Engage with us!

MapR

maprtech

mapr-technologies


© 2016 MapR Technologies© 2016 MapR TechnologiesMapR Confidential 47

Fast Cars, Big Data - How Streaming Can Help Formula 1

Technology