changing engines in midstream: a java stream computational model for … · 2016-01-29 · changing...

Changing Engines in Midstream: A Java StreamComputational Model for Big Data Processing

Xueyuan Su, Garret Swart, Brian Goetz,Brian Oliver, Paul Sandoz

Oracle Corporation

VLDB’14September 1st - 5th, 2014

1 Xueyuan Su etc. DistributableStream for Big Data Processing

Xueyuan SuGarret Swart

Brian GoetzPaul Sandoz

Brian Oliver


MotivationDistributableStream

Next Steps

Challenges We Try to AddressClarifications

Motivation



Next Steps


Big Data Space

Many data sources!Many compute engines!

Many tools to learn, use, and maintain!



Next Steps


Usability

Simple computational model.Friendly programming interface.

School kids can process Big Data!

Daddy's Hadoop app will take over

the world!

Cool!I hope to do the same in Java 101.

Alice Bob



Next Steps


Portability

A single API supported over multiple engines.Reuse applications developed for old engines.Leverage the investment in past development.

They also want your Hadoop app on Spark.

Sure... Maybe in 6 months?

Manager Developer

Customers



Next Steps


Query Federation

Various data processing requirements.Varied engine capabilities.

Price, data locality, and resource availability.

BIGDATA



Next Steps


Proposal

A Java streamcomputational model

and interfacefor Big Data processing



Next Steps


Clarifications

Q: Why Java?

A: User friendliness, big user base, broad adoption in Hadoopecosystem (with other JVM-based languages), ...

Q: Why not SQL?

A: We certainly love SQL – but not all Java programmers useSQL, less natural to implement certain applications in adeclarative language, one can build a SQL compiler on top, ...

Q: Yet another data-parallel MPP system?

A: No. A clean computational model and API for federatingdifferent MPP systems both between and within a query.



Next Steps

ModelDesign & ImplementationPerformance

DistributableStream



Next Steps


Take-Home Message

DistributableStreamis an abstraction that supports

generic, distributed and federatedqueries on top of an extensible

set of compute engines.



Next Steps


Concise Yet Expressive

WordCount

public static Map<String, Integer> wordCount(

DistributableStream<String> stream) {

return stream

.flatMap(s -> Stream.of(s.split("\\s+")))

.collect(DistributableCollectors

.toMap(s -> s, s -> 1, Integer::sum)); }



Next Steps


Generic Programming on Distributed Engines

ThreadPool

Hadoop MapReduce

Apache Spark

Oracle Coherence



Next Steps


Stream Stages On Respective Optimized Engines

ThreadPool

Hadoop MapReduce

Apache Spark

Oracle Coherence

Initial Parsing & Filtering

IteratingUpdating Summary &

Evaluating Termination Condition



Next Steps


Model



Next Steps


The Java 8 Stream Model

A stream represents a sequence of elements thatsupport sequential and parallel aggregate operations.

A stream pipeline consists of a source, zero or moreintermediate operations, and a terminal operation.

Data itemIntermediateoperation

ResultIntermediateoperation

Terminaloperation

...



Next Steps


Stream Transforms: Intermediate Operations

An intermediate operation returns a new stream from astream and is processed lazily.

Commonly used intermediate operations include filter,map, flatMap, distinct, and so on.

Stream Streamfilter,Map,

flatMap,distinct,…



Next Steps


Terminal Operations

A terminal operation triggers the traversal of dataitems and consumes the stream.

Two commonly used terminal operations are reduceand collect.

StreamResult

reduce,collect,...



Next Steps


Collectors

Collect method usually works with a Collector.

A Collector is defined by a Supplier, an Accumulator, aCombiner, and an optional Finisher.

Data item

Container

Accumulator

Supplier Container

AccumulatorData item

Supplier

Combiner

Container

ContainerContainer

ContainerCombiner

Container

Finisher

Result



Next Steps


We extend the Streammodel to allow the use of

distributed enginesfor processing

distributed data sets.



Next Steps


Design & Implementation



Next Steps


DistributableStream

Function shipping via Java serialization.

Assemble local streams from local data partitions.

Engine specific immutable distributed collections.

Compute Engine Client Node

Distributedjob

optimizations

Serializedpipeline

DistributableStream

Worker Node

Computational Stage

DeserializedPipeline

Stream

Data Storage

Runtime JVMoptimizations

Localcollection

Enginespecific

distributedcollection

Datapartitions



Next Steps


Mapping Streams into Job Plans

Break stream computations into stages at the pointswhere shuffle is required.

PageflatMap collectToStream

flatMap collectToStream

flatMap collectPage

...PageRank:



Next Steps


Engine Interface

Engine interface for separating low-level details fromthe computational model and negotiating data/statemovement between engines.

Each compute engine needs to implement the Engineinterface.

Engine parameters are configured in the engineconfiguration object.



Next Steps


Configuring Engine Parameters

MapReduceEngine



Next Steps


Changing Engines

DistributableStream.withEngine(Engine) switches theunderlying engine and returns an instance of theDistributableStream associated with the new engine.

Hadoop MapReduce

Oracle Coherence

dstream.map(...).filter(...).withEngine(engine)



Next Steps


Data Movement When Changing Engines

Push vs. Pull

UpstreamEngine

Push

Task

Data Storage

DownstreamEngine

Task

Data Storage

Read

UpstreamEngine

Task

Data Storage

DownstreamEngine

Task

Data Storage

Write Pull



Next Steps


Push

Push is the default option.

Upstream engine writes to downstream storage.

UpstreamEngine

Push

Task

Data Storage

DownstreamEngine

Task

Data Storage

Read



Next Steps


Pull

Pull is usually used when upstream engine cannot writeto downstream storage.

When upstream engine is in memory, pulling from itsaves disk access costs.

UpstreamEngine

Task

Data Storage

DownstreamEngine

Task

Data Storage

Write Pull



Next Steps


Special Pull: Short-Circuiting

Short-circuiting enables downstream engine to pull fromupstream storage without running tasks there.

Use case: Hadoop InputFormat → Coherencein-memory cache, similar to HadoopRDD in Spark.

UpstreamEngine

Task

Data Storage

DownstreamEngine

Task

Data Storage

Pull



Next Steps


Example Applications

WordCount

Distributed Reservoir Sampling

PageRank

K-Means Clustering

Refer to the paper for actual code



Next Steps


Performance



Next Steps


Environment

Oracle Big Data Appliance (BDA):- Each node has:

- 2 × eight-core Intel Xeon processors.- 64GB memory.- 12 × 4TB 7200RPM disks.- InfiniBand interconnections.

- Cloudera Hadoop CDH 5.0.2.- Oracle Coherence 12.1.2.- Java SE 8u5.- Recompile Hadoop and Coherence source code

with JDK8, install JRE8.



Next Steps


Overhead?



Next Steps


Stream vs. Native Implementations

WordCount implemented with DistributableStreamand native Hadoop.

Input: 45GB Wikipedia dumps.

Writable and Java immutable types for comparison.



Next Steps


Stream vs. Native Implementations (cont.)

0 1 2 3 4 5 6 7 8 9

10

Writable Immutable

Norm

alized

Tim

e

Object Type

MRStream

Hadoop

1

4.63

1.17

8.54

Native Hadoop implementation is slower, 1.17× forWritables, and 1.84× for immutables.

Partial in-memory merging before MapOutputBuffer.



Next Steps


Why Federation?



Next Steps


Hadoop MR vs. Coherence as Engine for Iterations

K-means implemented with DistributableStream.

Input: 45GB raw data representing one billion vertices.

Hadoop MR for parsing raw input, {Hadoop MR,Coherence} for iterations, Local SMP for updatingcentroids and evaluating termination condition.

Most disk IOs are avoided during the job execution bycaching the input data and all intermediate results inthe OS cache.



Next Steps


Hadoop vs. Coherence as Engine for Iterations (cont.)

0 2 4 6 8

10 12 14 16 18

1 5 10 20 30

Norm

alized

Tim

e

Number of Iterations

Hadoop + Coherence + Local

Hadoop + Hadoop + Local

11.86

2.88

4.87

6.98

3.1

4.79

6.92

11.67

16.13

Iterations w/ Hadoop MR is slower, 2.3× to 3.1×.

Accessing OS cache is slower than Java heap.

In-memory Java objects avoid deserialization cost.



Next Steps

Next?



Next Steps

Short Term

More Data Sources,such as Databases.

More Compute Engines,such as Apache Spark and Tez.



Next Steps

Long Term

Job Planner and Optimizer,for automatical engine assignment.

Job Progress Monitor,for fault tolerance across engines.

Java JIT Optimization,for low-level JVM tuning.

API Extension,for supporting DAGs.



Next Steps

Development Community

We need your help!

A JSR (Java Specification Request).

An OpenJDK project.



Next Steps

Take-Home Message

DistributableStreamis an abstraction that supports

generic, distributed and federatedqueries on top of an extensible

set of compute engines.

You are welcome to make contributions!



Next Steps

Thank You!

Questions?



Next Steps

Backup Slides



Next Steps

LocalStream

A wrapper DistributableStream implementation thatwraps a Java 8 Stream inside.

Operations are delegated to the wrapped Stream.



Next Steps

MapReduceStream

Pipelining several map phases inside a single mappersimilar to the ChainMapper but w/o the need forconfiguring parameters for each phase separately.

Partial in-memory merging before MapOutputBuffer.

Hadoop Cluster

Reducer

collector

Combiner,Merger

HDFS

Localcollection

Shuffling

Container,key/value

Mapper

filter(predicate)flatMap(mapper)

collector

Stream

HDFS

Runtime JVM optimizations

Container,key/valueInputSplit



Next Steps

CoherenceStream

Portable object format (POF).

Invocation services.

Use of filter and partition affinity for performance.

Coherence Grid

ReducerService

collector

Combiner,Merger

NamedCache

Localcollection

Partitionaffinity

Container,key/value

MapperService

filter(predicate)flatMap(mapper)

collector

Stream

NamedCache

Runtime JVMoptimizations

Container,key/value

Localpartitions



Next Steps

Creation of DistributableStreams

By Engine instance methods over a persistent enginespecific data set, e.g.,MapReduceEngine.valueStream(conf)

From an Engine specific distributed collection, e.g.,coherenceDistMap.entryStream()

Use the result ofDistributableStream.collectToStream(collector)


changing engines in midstream: a java stream computational model for … · 2016-01-29 · changing...

Documents

Midstream Environmental Developments

bp midstream partners focused on safetybp midstream partners 2Q 2020 results. 3. bp midstream. Cautionary statement partners. FORWARD-LOOKING STATEMENTS. This presentationincludes

Midstream/Energy Fund

DCP Midstream Partners - RBN Energy Midstream Partners... · DCP Midstream Partners is responding to today’s difficult market environment through a number of initiatives. In its

MIDSTREAM LIQUID PIPELINE

How to Use Midstream Incentives to Promote ENERGY STAR ... · PDF fileExpansion on Consumer Electronics Midstream Program Designs ... Alliance (NEEA) evaluation of their midstream

Changing Regional Midstream andChanging Regional Midstream ...€¦ · Changing Regional Midstream andChanging Regional Midstream and Transportation Developments Terry McGill Energy

Gas Midstream Brochure

EERC - North DakotaOasis Midstream Partners, ONEOK, Andeavor, and Goodnight Midstream. By the end of 2019, three additional members (DCP Midstream, Whiting Petroleum, and Enbridge)

Sponge: Portable Stream Programming on Graphics Engines

Midstream f

TPH Midstream Primer

The Aurora and Borealis Stream Processing Engines · The Aurora and Borealis Stream Processing Engines 339 Fig. 1 Aurora graphical user interface 2 The Aurora Centralized Stream Processing

The Aurora and Borealis Stream Processing Enginesmagda/borealis-book.pdf · The Aurora and Borealis Stream Processing Engines U˘gur C¸etintemel1, Daniel Abadi2, Yanif Ahmad1, Hari

EnLink Midstream / Tall Oak Midstream Acquisition Investor Call

Scalability Benchmarking of Stream Processing Engines with

Flow Level - Midstream

Enable Midstream Partners, LPinvestors.enablemidstream.com/sites/enablemidstream.investorhq... · Enable Midstream Partners, LP ... regarding Enable Midstream Partners’ (“Enable”)

Partial Key Grouping - arXivIndex Terms—Load balancing, stream processing, power of both choices, stream grouping. F 1 INTRODUCTION D ISTRIBUTED stream processing engines (DSPEs)

DTE Midstream and DTE Gas Midstream Services · PDF fileDTE Midstream and DTE Gas Midstream Services 2017 Customer Meeting in ... Ain’t Nothing Like the Real Thing. ... Ain’t No

Linked Stream Data Processing Engines: Facts and Figuresiswc2012.semanticweb.org/sites/default/files/76500294.pdf · Linked Stream Data Processing Engines: Facts and Figures? Danh

LNCS 7650 - Linked Stream Data Processing Engines: Facts ...€¦ · Linked Stream Data Processing Engines: Facts and Figures Danh Le-Phuoc1, Minh Dao-Tran2, Minh-Duc Pham3, Peter

Skoolsake - Midstream College

05/10/18 DCP Midstream LP Midstream DCP $ 150 $ 4.71 3.14% 0.00% 0.00% BAML 7 06/28/18 CNX Midstream Partners Midstream CNXM $ 119 $ 1.78 1.50% 10.51% 11.86% Citi 2 06/20/18 Tellurian

The Power of Both Choices: Practical Load Balancing for Distributed Stream Processing Engines

Database Systems 13 Stream Processing - GitHub Pages · Distributed stream processing engines, and “unified” batch/stream processing Proprietary systems: Google Cloud Dataflow,

COOPERATIVE RESEARCH - GPA Midstream · EnLink Midstream Enterprise Products Operating ExxonMobil/XTO ... August 2017 meeting. For 2018, GPA Midstream has directed $50,000 from GPSA

Hess Midstream

Stream Engines in Howarth 2014

EnLink Midstream Partners, LPd18rn0p25nwr6d.cloudfront.net/CIK-0001179060/... · EnLink Midstream Partners, LP and EnLink Midstream, LLC conducted an earnings call on November 7,

Progress Apama & Event Processing · The Elements of Event Stream Processing The EPL and Stream Processing Engines Management ESB, Reuters, ALE Correlation Engine Real-Time Dashboards

Heaven: Supporting Systematic Comparative Research of RDF Stream Processing Engines

Changing Engines in Midstream: A Java Stream Computational

Midstream Final DELOITTE

Delfin Midstream