Big Graph Data

BIG GRAPH DATAUnderstanding a Complex WorldMatthias Broecheler, CTO@mbroecheler AURELIUSNovember XIII, MMXII THINKAURELIUS.COM

IGraph Foundation AURELIUS THINKAURELIUS.COM

name: Neptune name: Alcmene type: god type: godVertex Property name: Saturn name: Jupiter name: Hercules type: titan type: god type: demigod name: Pluto name: Cerberus type: god type: monster Graph

name: Neptune name: Alcmene type: god type: godEdge brother mother name: Saturn name: Jupiter name: Hercules type: titan type: god type: demigod father father Edge battled brother Property time:12 name: Pluto name: Cerberus type: god type: monster Edge Type pet Graph

name: Neptune name: Alcmene type: god type: god brother mothername: Saturn name: Jupiter name: Herculestype: titan type: god type: demigod father father battled brother time:12 name: Pluto name: Cerberus type: god type: monster pet Path

name: Neptune name: Alcmene type: god type: god brother mothername: Saturn name: Jupiter name: Herculestype: titan type: god type: demigod father father battled brother time:12 name: Pluto name: Cerberus type: god type: monster pet Degree

IConnected World AURELIUS THINKAURELIUS.COM

HEALTH

ECONOMY

SocialSystems

IIITitan Graph Database AURELIUS THINKAURELIUS.COM

Titan Features  Numerous Concurrent Users  Many Short Transactions   read/write  Real-time Traversals (OLTP)  High Availability  Dynamic Scalability  Variable Consistency Model   ACID or eventual consistency  Real-time Big Graph Data

Storage Backends PartitionabilityConsistency Availability

Titan FeaturesI.  Data ManagementII.  Vertex-Centric Indices

Titan FeaturesIII.  Graph PartitioningIV.  Edge Compression

Titan Ecosystem  Native Blueprints Graph Server Implementation Graph  Gremlin Query Algorithms Language Object-Graph Mapper  Rexster Server Traversal Language   any Titan graph can be exposed as a REST endpoint Dataﬂow Processing Generic Graph API

IVGithub Network AURELIUS THINKAURELIUS.COM

Setup$ ./titan-0.1.0/bin/gremlin.sh! ! ! !,,,/! (o o)!-----oOOo-(_)-oOOo-----!gremlin> g = TitanFactory.open(/tmp/titan)!==>titangraph[local:/tmp/titan]!

Titan Storage Model  Adjacency list in one 5 column family  Row key = vertex id  Each property and edge in one column 5   Denormalized, i.e. stored twice  Direction and label/key as column preﬁx   Use slice predicate for quick retrieval

created USER edited opened pushedCOMMENT PAGE on ISSUE COMMIT on on to in REPOSITORY

Defining Property Keysgremlin> g.makeType().name(‘username’).! ! ! ! dataType(String.class).! ! ! ! functional().! ! ! ! indexed().unique().! ! ! ! makePropertyKey()!gremlin> g.makeType().name(‘time’).! ! ! ! dataType(Long.class).! ! ! ! functional().makePropertyKey()!

Defining Edge Labelsgremlin> g.makeType().name(‘on’).! ! ! ! makeEdgeLabel()!gremlin> g.makeType().name(‘pushed’).! ! ! ! primaryKey(time).! ! ! ! makeEdgeLabel()!gremlin> g.makeType().name(‘in’).! ! ! ! unidirected().! ! ! ! makeEdgeLabel()!

Create & Retrievegremlin> v = g.addVertex([username: ‘okram’])!==>v[4]!gremlin> v.map!==>{username=okram}!gremlin> g.V(username,okram)!==>v[4]!

Titan Locking  Locking ensures consistency when it is needed name : Hercules 5  Titan uses time stamped consistent reads and writes 9 on separate CFs for locking  Uses name :   Property uniqueness: .unique() name : Hercules Jupiter   Functional edges: .functional() father   Global ID management x name : father Pluto

Titan Indexing  Vertices can be retrieved by property key + value name : Hercules 5  Titan maintains index in a separate column family as name : Jupiter 9 graph is updated  Only need to deﬁne a property key as .index()

Basic Queriesgremlin> v.out(‘pushed’)!gremlin> v.out(‘pushed’).out(‘to’).name!gremlin> v.out(‘pushed’).out(‘to’).dedup.name!gremlin> v.out(‘pushed’).out(‘to’).dedup.! ! ! ! name.sort{it}!gremlin> v.outE(‘pushed’).has(‘time’,T.gt,1000).inV!

Basic Queriesgremlin> v.out(‘pushed’)!gremlin> v.out(‘pushed’).out(‘to’).name!gremlin> v.out(‘pushed’).out(‘to’).dedup.name!gremlin> v.out(‘pushed’).out(‘to’).dedup.! ! ! ! name.sort{it}!gremlin> v.outE(‘pushed’).has(‘time’,T.gt,1000).inV! Query Optimization

Vertex-Centric Indices  Sort and index edges per vertex by primary key   Primary key can be composite  Enables efﬁcient focused traversals   Only retrieve edges that matter  Uses push down predicates for quick, index-driven retrieval

battled battled battled time: 1 time: 3 time: 5 mother battled v v.query()! time: 9 father fought fought

battled battled battled time: 1 time: 3 time: 5 mother battled v v.query()! time: 9 .direction(OUT)! father

battled battled battled time: 1 time: 3 time: 5 battled v v.query()! time: 9 .direction(OUT)! .labels(‘battled’)!

battled battled time: 1 time: 3 v v.query()! .direction(OUT)! .labels(‘battled’)! .has(‘time,T.lt,5)!

Recommendation Enginegremlin> v.out(pushed).out(to)[0..9].! ! ! ! in(to).in(pushed)[0..500].! ! ! ! except([v]).name.! ! ! ! groupCount.cap.next().sort{-it.value}[0..4]!

Recommendation Enginegremlin> v.out(pushed).out(to)[0..9].! ! ! ! in(to).in(pushed)[0..500].! ! ! ! except([v]).name.! ! ! ! groupCount.cap.next().sort{-it.value}[0..4]!v = g.V(‘username’,’okram’):!==>lvca=175!==>spmallette=56!==>sgomezvillamor=36!==>mbroecheler=33!==>joshsh=20!

Recommendation Enginegremlin> v.out(pushed).out(to)[0..9].! ! ! ! in(to).in(pushed)[0..500].! ! ! ! except([v]).name.! ! ! ! groupCount.cap.next().sort{-it.value}[0..4]!v = g.V(‘username’,’torvalds’):!==>iksaif=90!==>rjwysocki=22!==>kernel-digger=20!==>giuseppecalderaro=16!==>groeck=15!

Titan Embedding  Rexster RexPro   lightweight Gremlin Server   based on Grizzly  Titan Gremlin Engine  Embedded Storage Backend   in-JVM method calls

Graph PartitioningGoal: Vertex Co-location  Titan maintains multiple ID Pools  Ordered Partitioner in Storage Backend  Dynamically determines optimal partition and allocates corresponding ID Pool IDs

What’s coming  Full-text indexing   external index system integration  Bulk Loading   integration with storage backend utilities and Hadoop ingestion  240 Billion Edge Benchmark   performance analysis and improvements across the entire stack

VFaunus Graph Analytics AURELIUS THINKAURELIUS.COM

Faunus Features  Hadoop-based Graph Computing Framework  Graph Analytics  Breadth-ﬁrst Traversals  Global Graph Computations  Batch Big Graph Data

Faunus Architecture g._()!

Faunus Work Flowg.V.out .out .count() hdfs://user/ubuntu/ output/job-0/ output/job-1/ graph* output/job-2/ { sideeffect*Compressed HDFS Graphs  stored in sequence ﬁles  variable length encoding  preﬁx compression

Faunus Setup$ bin/gremlin.sh ! ,,,/! (o o)!-----oOOo-(_)-oOOo-----!gremlin> g = FaunusFactory.open(bin/titan-hbase.properties)!==>faunusgraph[titanhbaseinputformat]!gremlin> g.getProperties()!==>faunus.graph.input.format=com.thinkaurelius.faunus.formats.titan.hbase.TitanHBaseInputFormat==>faunus.graph.output.format=org.apache.hadoop.mapreduce.lib.output.SequenceFileOutputFormat!==>faunus.sideeffect.output.format=org.apache.hadoop.mapreduce.lib.output.TextOutputFormat!==>faunus.output.location=dbpedia!==>faunus.output.location.overwrite=true!gremlin> g._() !12/11/09 15:17:45 INFO mapreduce.FaunusCompiler: Compiled to 1 MapReduce job(s)!12/11/09 15:17:45 INFO mapreduce.FaunusCompiler: Executing job 1 out of 1:MapSequence[com.thinkaurelius.faunus.mapreduce.transform.IdentityMap.Map]!12/11/09 15:17:50 INFO mapred.JobClient: Running job: job_201211081058_0003!

Graph Analyticsgremlin> g.E.has(label,’followed).keep.! ! ! !V.sideEffect({it.degree = it.outE.count()}).! ! ! !degree.groupCount!gremlin> g.E.has(label,pushed).keep.! ! ! !V.sideEffect({it.degree = it.outE.count()}).! ! ! !degree.groupCount!

Follow Degree Distribution

Follow Degree Distribution P(k) ~ k-γ γ = 2.2

Pushed Degree Distribution

Global Recommendationsgremlin> g.E.has(label,pushed,to).keep.! ! ! !V.out(pushed).out(to).! ! ! !in(to).in(pushed).! ! ! !sideEffect({it.score =it.pathCounter}).! ! ! !score.order(F.decr,name)!# Top 5:!Jippi ! ! ! !60892182927!garbear ! ! !30095282886!FakeHeal ! ! !30038040349!brianchandotcom !24684133382!nyarla ! ! !15230275746!

What’s coming  Faunus 0.1  Bulk Loading   loaded graph into Titan   loading derivations into Titan  Extending Gremlin Support   currently only a subset is of Gremlin implemented  Operational Tools

IGraph = Relationship Centric

IIGraph = Agile Data Model

IIIGraph = Algebraic Data Model

Aurelius Graph Cluster Apache 2 Map/Reduce Load & Compress Analysis results back into Titan Stores a massive-scale Batch processing of large Runs global graph algorithmsproperty graph allowing real- graphs with Hadoop on large, compressed, time traversals and updates in-memory graphs

Speed of Traversal/Process The Graph LandscapeIllustration only, not to scale Size of Graph

TINKERPOP.COM

Thanks! Vadas Gintautas Marko Rodriguez @vadasg @twarko Stephen Mallette Daniel LaRocque @spmallette AURELIUS THINKAURELIUS.COM

AURELIUSTHINKAURELIUS.COM

XVXBenchmark Results AURELIUS THINKAURELIUS.COM

XVX - ITitan Performance Evaluation onTwitter-like Benchmark AURELIUS THINKAURELIUS.COM

Twitter Benchmark  1.47 billion followship edges and 41.7 million users   Loaded into Titan using BatchGraph   Twitter in 2009, crawled by Kwak et. al  4 Transaction Types   Create Account (1%)   Publish tweet (15%)   Read stream (76%)   Recommendation (8%)   Follow recommended user (30%) Kwak, H., Lee, C., Park, H., Moon, S., “What is Twitter, a Social Network or a News Media?,” World Wide Web Conference, 2010.

Benchmark Setup  6 cc1.4xl Cassandra nodes   in one placement group   Cassandra 1.10  40 m1.small worker machines   repeatedly running transactions   simulating servers handling user requests  EC2 cost: $11/hour

Benchmark ResultsTransaction Type Number of tx Mean tx time Std of tx timeCreate account 379,019 115.15 ms 5.88 msPublish tweet 7,580,995 18.45 ms 6.34 msRead stream 37,936,184 6.29 ms 1.62 msRecommendation 3,793,863 67.65 ms 13.89 ms Total 49,690,061 Runtime 2.3 hours 5,900 tx/sec

Peak Load ResultsTransaction Type Number of tx Mean tx time Std of tx timeCreate account 374,860 172.74 ms 10.52 msPublish tweet 7,517,667 70.07 ms 19.43 msRead stream 37,618,648 24.40 ms 3.18 msRecommendation 3,758,266 229.83 ms 29.08 ms Total 49,269,441 Runtime 1.3 hours 10,200 tx/sec

Benchmark ConclusionTitan can handle 10s of thousands of concurrent users with short response 5mes even for complex traversals on a simulated social networking applica5on based on real-‐world network data with billions of edges and millions of users in a standard EC2 deployment. For more informa5on on the benchmark: hDp://thinkaurelius.com/2012/08/06/5tan-‐provides-‐real-‐5me-‐big-‐graph-‐data/

http://www.scoop.it	181
http://skilldev.cloudapp.net	32
http://formacionmestreacasa.gva.es	21
https://twitter.com	13
http://circleoflightspiritualcentre.wordpress.com	3
http://www.skill-player.com	2
http://lectrio.com	2
http://twitter.com	1
https://si0.twimg.com	1

Big Graph Data

by Matthias Broecheler on Nov 14, 2012

Accessibility

Categories

Upload Details

Usage Rights

9 Embeds 256

Statistics

Big Graph Data Presentation Transcript