Spark 101

1© Copyright 2013 Pivotal. All rights reserved. 1© Copyright 2013 Pivotal. All rights reserved.

Apache Spark 101

what is Spark all about

Shahaf AzrielySr. Field Engineer Southern EMEA

2© Copyright 2013 Pivotal. All rights reserved.

Agenda

What is Spark

Spark Programming Model

– RDDs, log mining, word count …

Related Projects

– Shark, Spark SQL, Spark streaming, Graphx, Mllib and more …

So what next


What is Spark?

Pivotal Confidential–Internal Use Only

The Spark Challenge

• Data size is growing MapReduce greatly simplified big data analysis

• But as soon as it got popular, users wanted more:

- More complex, multi-stage applications (graph algorithms, machine learning)

- More interactive ad-hoc queries

- More real-time online processing

• All of these apps require fast data sharing across parallel jobs


Data Sharing in MapReduce

iter. 1 iter. 2 . . .

Input

HDFSread

HDFSwrite

HDFSread

HDFSwrite

Input

query 1

query 2

query 3

result 1

result 2

result 3

. . .

HDFSread

Slow due to replication, serialization, and disk IO


iter. 1 iter. 2 . . .

Input

Data Sharing in Spark

Distributedmemory

Input

query 1

query 2

query 3

. . .

one-timeprocessing

10-100× faster than network and disk


Spark is

Fast MapReduce-like engine.

– In memory storage for fast iterative computation.

– Design for low latency ~100ms jobs

Competitive with Hadoop storage APIs

– Read/write to any Hadoop supported systems including Pivotal HD.

Designed to work with data in memory

Programmatic or Interactive

Written in Scala but have bindings for Python/Java /Scala.

Make life easy and productive for Data Scientists

Spark is one of the most actively developed open source projects. It has over 465

contributors in 2014, making it the most active project in the Apache Software

Foundation and among Big Data open source projects.


Short History

Spark was initially started by Matei Zaharia at UC Berkeley AMPLab in 2009.

2010 Open Sourced

June 21 2013 the project was donated to the Apache Software Foundation and it’s founders created Databricks out of AmpLab.

Feb 27 2014 Spark becomes top level ASF project.

In November 2014, the engineering team at Databricks used Spark and set am amazing record in the Daytona GraySort sorting 100TB (1 trillion records) in 23 Min 4.27 TB/min.

http://databricks.com/blog/2014/11/05/spark-officially-sets-a-new-record-in-large-scale-sorting.html

http://databricks.com/blog/2014/11/05/spark-officially-sets-a-new-record-in-large-scale-sorting.html


Spark Programming

Model

RDDs in Detail


Programming Model

• Key idea: resilient distributed datasets (RDDs)

- Resilient – if data in memory is lost, it can be recreated.

- Distributed – stored in memory across the cluster.

- Dataset – initial data can be created from a file or programmatically.

• Parallel operations on RDDs

- Reduce, collect, foreach, …

• Interface

- Clean language-integrated API in Scala, Python, Java

- Can be used interactively


RDD Fault Tolerance

RDDs maintain lineage information that can be used to reconstruct lost partitions

cachedMsgs = textFile(...).map(_.split(‘\t’)(2)).filter(_.contains(“error”)).cache()

HdfsRDDpath: hdfs://…

FilteredRDDfunc: contains(...)

MappedRDDfunc: split(…)

CachedRDD


Demo: Intro & Log Mining

Create basic RDD in Scala:1 2Log Mining - Load error messages from a log into memory, then

interactively search for various patterns

Base RDD

Transformed RDD

Action


Transformation and Actions

Transformations

– Map

– filter

– flatMap

– sample

– groupByKey

– reduceByKey

– union

– join

– sort

Actions

– count

– collect

– reduce

– lookup

– save

Look at

http://spark.apache.org/docs/latest/progra

mming-guide.html#basics

http://spark.apache.org/docs/latest/programming-guide.html


More Demo: Word count & Joins

Word count in Scala and python shells 3 4 Join two RDDs


Example of Related

Projects


Related Projects

Shark is dead long live Spark SQL

Spark Streaming

GraphX

MLbase

Others


Shark is dead but what it was

Hive on Spark

– HiveQL, UDFs, etc.

Turn SQL into RDD

– Part of the lineage

Based on Hive, but takes advantage of Spark for

– Fast Scheduling

– Queries are DAGs of jobs, not chained M/R

– Fast broadcast variables

© Apache Software Foundation


Spark SQL

Lib in Spark Core to treat RDDs as relations SchemaRDD

RDDs are columnar memory store.

Dynamic query optimization

Lighter weight version of Shark

– No code from Hive

Import/Export in different Storage formats

– Parquet, learn schema from existing Hive warehouse


Spark SQL Code


Spark Streaming

• Framework for large scale stream processing

• Scales to 100s of nodes

• Can achieve second scale latencies

• Integrates with Spark’s batch and interactive processing

• Provides a simple batch-like API for implementing complex algorithm

• Can absorb live data streams from Kafka, Flume, ZeroMQ, etc.


• Traditional streaming systems have a event-driven record-at-a-time processing model

– Each node has mutable state

– For each record, update state & send new records

• State is lost if node dies!

• Making stateful stream processing be fault-tolerant is challenging

Traditional streaming


Discretized Stream Processing

Run a streaming computation as a series of very small, deterministic batch jobs

22

Spark

SparkStreaming

batches of X seconds

live data stream

processed results

• Chop up the live stream into batches of X seconds

• Spark treats each batch of data as RDDs and processes them using RDD operations

• Finally, the processed results of the RDD operations are returned in batches


Discretized Stream Processing

Run a streaming computation as a series of very small, deterministic batch jobs

23

Spark

SparkStreaming

batches of X seconds

live data stream

processed results

• Batch sizes as low as ½ second, latency ~ 1 second

• Potential for combining batch processing and streaming processing in the same system


How Fast Can It Go?

Can process 4 GB/s (42M records/s) of data on 100 nodes at sub-second latency

Recovers from failures within 1 sec


Streaming how does it works


MLlib

MLlib is a Spark subproject providing machine learning primitives.

It ships with Spark as a standard component.

Many different Algorithms

– Classification, Regression, Collaborative Filtering:

o Regression: generalized linear regression (GLM)

o Collaborative filtering: alternating least squares (ALS)

o Clustering: k-means

o Decomposition: singular value decomposition (SVD), principal

o Component analysis (PCA)


Why MLlib

It is built on Apache Spark, a fast and general engine for large-scale data processing.

Run programs up to 100x faster than Hadoop MapReduce in memory, or 10x faster on disk.

Write applications quickly in Java, Scala, or Python.

You can use any Hadoop data source (e.g. HDFS, HBase, or local files), making it easy to plug into Hadoop workflows.


Spark SQL + MLlib


GraphX

What are Graphs? They are inherently recursive data-structures as properties of vertices depend on properties of their neighbors which in turn depend on properties of their neighbors.

GraphX unifies ETL, exploratory analysis, and iterative graph computation within a single system.

We can view the same data as both graphs and collections, transform and join graphs with RDDs.

For example Predicting things about people (eg: political bias)

– Look at posts, apply classifier, try to predict attribute

– Look at context of social network to improve prediction


GraphX Demo


Others

Mesos

– Enable multiple frameworks to share same cluster resources

– Twitter is largest user: Over 6,000 servers

Tachyon

– In-memory, fault tolerant file system that exposes HDFS.

– Use as the FS for Spark.

Catalyst

– SQL Query Optimizer


So We is Spark

important for Pivotal


So How Real is Spark?

Leveraging modern MapReduce engine and techs from database, Spark support both SQL and complex analytics efficiently.

There are many indicators that Spark is heading to success

– Solid technology

– Good buzz

– Community is getting bigger https://cwiki.apache.org/confluence/display/SPARK/Committers


Pivotal’s Positioning of Spark

• PHD is a highly differentiated and the only platform that brings the benefits of closed loop analytics to enable business data lake

• With Spark we extend that differentiation by allowing up to 100x faster batch processing

HAWQMap-Reduce Gemfire XDSpark?

Batch Processing Near-Real Time Real TimeBetter/Faster

Batch Processing


Spark 1.0.0 on PHD https://support.pivotal.io/hc/en-us/articles/203271897-Spark-on-Pivotal-Hadoop-2-0-Quick-Start-Guide

Databrick’s announcing Pivotal certification https://databricks.com/blog/2014/05/23/pivotal-hadoop-integrates-the-full-apache-spark-stack.html

We attend Spark meetup’s

Join the SocialCast group!

https://support.pivotal.io/hc/en-us/articles/203271897-Spark-on-Pivotal-Hadoop-2-0-Quick-Start-Guide

https://databricks.com/blog/2014/05/23/pivotal-hadoop-integrates-the-full-apache-spark-stack.html


Thank you

Q&A

Spark 101

Technology

pivotal hd

apache spark

data scientists spark

spark sql

spark streaming

fast data sharing

big data analysisbut

big data open source