NOSQLEU - Graph Databases and Neo4j

Graph Databases
and Neo4j
twitter: @thobe / #neo4j
Tobias Ivarsson email: tobias@neotechnology.com
web: http://www.neo4j.org/
Hacker @ Neo Technology web: http://www.thobe.org/

NOSQL - Why now?
Four trends

2

Trend 1: Data size
ExaBytes (10¹⁸) of data stored per year
988
1000
Each year more and
more digital data is
created. Over t wo
750 years we create more
digital data than all 623
the data created in
history before that.
500
397

253
250 161

0
2006 2007 2008 2009 2010
Data source: IDC 2007 3

Trend 2: Connectedness
Giant
Global
Graph (GGG)

Over time data has evolved to Ontologies
be more and more interlinked
and connected.
RDF
Hypertext has links,
Blogs have pingback,
Tagging groups all related data Folksonomies
Information connectivity

Tagging

Wikis User-generated
content
Blogs

RSS

Hypertext

Text documents
web 1.0 web 2.0 “web 3.0”

1990 2000 2010 2020 4

Trend 3: Semi-structure
๏ Individualization of content
• In the salary lists of the 1970s, all elements had exactly one job
• In Or 15? lists of the 2000s, we need 5 job columns! Or 8?
the salary

๏ All encompassing “entire world views”
• Store more data about each entity
๏ Trend accelerated by the decentralization of content generation
that is the hallmark of the age of participation (“web 2.0”)

5

Trend 4: Architecture

1980s: Mainframe applications

Application

DB

6


1990s: Database as integration hub

Application Application Application

DB

7


2000s: (moving towards) Decoupled services
with their own backend

Application Application Application

DB DB DB

8

Why NOSQL Now?

๏Trend 1: Size
๏Trend 2: Connectedness
๏Trend 3: Semi-structure
๏Trend 4: Architecture

9

RDBMS performance
Salary List Relational database

Requirement of application
Performance

Majority of
Webapps

Social network
We are building

}
applications today that
Semantic Trading
have complexity
requirements that a
Relational Database
cannot handle with
sufficient performance
custom

Data complexity 10

Scaling to size vs. Scaling to complexity
Size
Key/Value stores

Bigtable clones

Document databases

Graph databases
Billions of nodes
and relationships

> 90% of use cases

Complexity

11

Graph Databases focuses on structure of data
Graph databases focus
on the structure of the
data, scaling to the
complexity of the data
and of the application.

12

What is Neo4j?
๏ Neo4j is a Graph Database
• Non-relational (“#nosql”), transactional (ACID), embedded
• Data is stored as a Graph / Network
‣Nodes and relationships with properties
‣“Property Graph” or “edge-labeled multidigraph”
• Schema free, bottom-up data model design
๏ Neo4j is Open Source / Free (as in speech) Software
Prices are available at
http://neotechnology.com/

• AGPLv3
Contact us if you have
questions and/or special
license needs (e.g. if you

• Commercial (“dual license”) license available
want an evaluation license)

‣First server is free (as in beer), next is inexpensive 13

More about Neo4j
๏ Neo4j is stable
• In 24/7 operation since 2003
๏ Neo4j is in active development
• Neo Technology received VC funding October 2009
๏ Neo4j delivers high performance graph operations
• traverses 1’000’000+ relationships / second
on commodity hardware

14

The Neo4j Graph data model

•Nodes
•Relationships bet ween Nodes
•Relationships have Labels
•Relationships are directed, but traversed at
equal speed in both directions
•The semantics of the direction is up to the
application (LIVES WITH is reflexive, LOVES is not)
•Nodes have key-value properties
•Relationships have key-value properties 15


LIVES WITH
LOVES

OWNS
DRIVES

•Nodes


LOVES

LIVES WITH
LOVES

OWNS
DRIVES

•Nodes

name: “Mary”
LOVES
name: “James” age: 35
age: 32 LIVES WITH
twitter: “@spam” LOVES

OWNS
DRIVES

•Nodes
•Relationships have Labels brand: “Volvo”
•Relationships are directed, but traversed at model: “V70”

name: “Mary”
LOVES
name: “James” age: 35
age: 32 LIVES WITH
twitter: “@spam” LOVES

OWNS
item type: “car” DRIVES

•Nodes
•Relationships have Labels brand: “Volvo”
•Relationships are directed, but traversed at model: “V70”

Graphs are all around us
A B C D ...
1 17 3.14 3 17.79333333333

2 42 10.11 14 30.33

3 316 6.66 1 2104.56

4 32 9.11 592 0.492432432432

5 Even if this spreadsheet looks
like it could be a fit for a RDBMS
2153.175765766
it isn’t:
•RDBMSes have problems with
... extending indefinitely on both
rows and columns
•Formulas and data
dependencies would quickly lead
to heavy join operations

16

Graphs are all around us
A B C D ...
1 17 3.14 3 = A1 * B1 / C1

2 42 10.11 14 = A2 * B2 / C2

3 316 6.66 1 = A3 * B3 / C3

4 32 9.11 592 = A4 * B4 / C4

5 = SUM(D2:D5)
With data dependencies
... the spread sheet turns
out to be a graph.

17

Graphs are all around us If we add external data
sources the problem
becomes even more
interesting...

17 3.14 3 = A1 * B1 / C1

42 10.11 14 = A2 * B2 / C2

316 6.66 1 = A3 * B3 / C3

32 9.11 592 = A4 * B4 / C4

= SUM(D2:D5)

18

Graphs are whiteboard friendly An application domain model
outlined on a whiteboard or piece
of paper would be translated to
an ER-diagram, then normalized
to fit a Relational Database.
With a Graph Database the model
from the whiteboard is
implemented directly.

Image credits: Tobias Ivarsson 19


*
1
*
* 1

* 1
*

1
*


thobe

Joe project blog

Wardrobe Strength

Hello Joe

Modularizing Jython

Neo4j performance analysis

Query Languages
๏ Traversal APIs
• Neo4j core traversers
• Blueprint pipes
๏ SPARQL - “SQL for linked data” - query by graph pattern matching
SELECT ?person WHERE { Find all persons that
?person neo4j:KNOWS ?friend . KNOWS a friend that
?friend neo4j:KNOWS ?foe . KNOWS someone named
“Larry Ellison”.
?foe neo4j:name "Larry Ellison" .
}

๏ Gremlin - “perl for graphs” - query by traversal
./outE[@label='KNOWS']/inV[@age > 30]/@name

Give me the names of all the people I know that are older than 30. 20

Data manipulation API
GraphDatabaseService graphDb = getGraphDbInstanceSomehow();

// Create Thomas 'Neo' Anderson
Node mrAnderson = graphDb.createNode();
mrAnderson.setProperty( "name", "Thomas Anderson" );
mrAnderson.setProperty( "age", 29 );

// Create Morpheus
Node morpheus = graphDb.createNode();
morpheus.setProperty( "name", "Morpheus" );
morpheus.setProperty( "rank", "Captain" );
morpheus.setProperty( "occupation", "Total bad ass" );

// Create relationship representing they know each other
mrAnderson.createRelationshipTo( morpheus, RelTypes.KNOWS );
// ... similarly for Trinity, Cypher, Agent Smith, Architect

21

Data manipulation API
GraphDatabaseService graphDb = getGraphDbInstanceSomehow();
Transaction tx = graphDb.beginTx();
try {
// Create Thomas 'Neo' Anderson
Node mrAnderson = graphDb.createNode();
mrAnderson.setProperty( "name", "Thomas Anderson" );
mrAnderson.setProperty( "age", 29 );

// Create Morpheus
Node morpheus = graphDb.createNode();
morpheus.setProperty( "name", "Morpheus" );
morpheus.setProperty( "rank", "Captain" );
morpheus.setProperty( "occupation", "Total bad ass" );

// Create relationship representing they know each other
mrAnderson.createRelationshipTo( morpheus, RelTypes.KNOWS );
// ... similarly for Trinity, Cypher, Agent Smith, Architect
tx.success();
} finally {
tx.finish(); 21
}

Graph traversals

name: “The Architect”
disclosure: “public”
name: “Thomas Anderson”
age: 29 name: “Cypher”
last name: “Reagan”
KNOWS name: “Morpheus”
KNOWS KNOWS
rank: “Captain” CODED BY
LOVES occupation: “Total badass” KNOWS
KNOWS
name: “Trinity” disclosure: “secret”
name: “Agent Smith”
version: “1.0b”
since: “meeting the oracle” since: “a year before the movie”
language: “C++”
cooperates on: “The Nebuchadnezzar”

22

Graph traversals name: “The Architect”
KNOWS KNOWS
KNOWS
version: “1.0b”
language: “C++”
import neo4j
class Friends(neo4j.Traversal): # Traversals ! queries in Neo4j
types = [ neo4j.Outgoing.KNOWS ]
order = neo4j.BREADTH_FIRST
stop = neo4j.STOP_AT_END_OF_GRAPH
returnable = neo4j.RETURN_ALL_BUT_START_NODE
for friend_node in Friends(mr_anderson):
print "%s (@ depth=%s)" % ( friend_node["name"],
friend_node.depth )
23

KNOWS KNOWS
KNOWS
version: “1.0b”
language: “C++”
import neo4j
types = [ neo4j.Outgoing.KNOWS ] Morpheus (@ depth=1)
order = neo4j.BREADTH_FIRST
friend_node.depth )
23

KNOWS KNOWS
KNOWS
version: “1.0b”
language: “C++”
import neo4j
order = neo4j.BREADTH_FIRST Trinity (@ depth=1)
friend_node.depth )
23

KNOWS KNOWS
KNOWS
version: “1.0b”
language: “C++”
import neo4j
Cypher (@ depth=2)
friend_node.depth )
23

KNOWS KNOWS
KNOWS
version: “1.0b”
language: “C++”
import neo4j
Cypher (@ depth=2)
Agent Smith (@ depth=3)
friend_node.depth )
23

Finding a place to start
๏ Traversals need a Node to start from
• QUESTION: How do I ﬁnd the start Node?
• ANSWER:You use an Index
๏ Indexes in Neo4j are different from Indexes in Relational Databases
• RDBMSes use them for Joining
• Neo4j use them for simple lookup
IndexService index = getGraphDbIndexServiceSomehow();

Node mrAnderson = index.getSingleNode( "name",
"Thomas Anderson" );

performTraversalFrom( mrAnderson );
24

Indexes in Neo4j
๏ The Graph *is* the main index
• Use relationship labels for navigation
• Build index structures *in the graph*
‣Search trees, tag clouds, geospatial indexes, et.c.
‣Linked/skip lists or other data structures in the graph
‣We have utility libraries for this
๏ External indexes used *for lookup*
• Finding a (number of) points to start traversals from
• Major difference from RDBMS that use indexes for everything
25

A domain object implemented in Neo4j
public interface Person {
String getName();
void setName( String firstName, String lastName );
}

public final class PersonImpl implements Person {
private final Node underlyingNode;
public PersonImpl( Node underlyingNode ) {
this.underlyingNode = underlyingNode;
}
public String getName() {
return String.format("%s %s",
underlyingNode.getProperty("first name"),
underlyingNode.getProperty("last name") );
}
public String setName(String firstName, String lastName) {
underlyingNode.setProperty("first name", firstName);
underlyingNode.setProperty("last name", lastName);
}
} 26

Neo4j as Software Transactional Memory
๏ Implement objects as wrappers around Nodes and Relationships
• Neo4j is fast enough to allow you to read all state from the
Node/Relationship
๏ Mutating operations require transactions
• The changes are isolated from all other threads until committed
• Multiple mutations can be committed atomically
๏ Nested transactions are ﬂattened
• Makes it possible to have methods open their own transaction
๏ Fits nicely with the OO paradigm
• More focus on data than on objects (comp. Object DBs) 27

Why not use an O/R mapper?
๏ Model evolution in ORMs is a hard problem
• virtually unsupported in most ORM systems
๏ SQL is “compatible” across many RDBMSs
• data is still locked in
๏ Each ORM maps object models differently
• Moving to another ORM == legacy schema support
‣except your legacy schema is a strange auto-generated one
๏ Object/Graph Mapping is always done the same way
• allows you to keep your data through application changes
• or share data between multiple implementations 28

What an ORM doesn’t do

๏Deep traversals
๏Graph algorithms
๏Shortest path(s)
๏Routing
๏etc.
29

Path exists in social network
๏ Each person has on average 50 friends The performance impact
in Neo4j depends only on
the degree of each node. in
Tobias an RDBMS it depends on
the number of entries in
the tables involved in the
join(s).
Emil

Johan
Peter

Database # persons query time
Relational database 1 000 2 000 ms
Neo4j Graph Database 1 000 2 ms
Neo4j Graph Database 1 000 000 2 ms
Relational database 1 000 000 way too long...
30

On-line real time routing with Neo4j
๏ 20 million Nodes - represents places
๏ 62 million Edges - represents direct roads between places
• These edges have a length property, for the length of the road
๏ Average optimal route, 100 separate roads, found in 100ms
๏ Worst case route we could ﬁnd:
• Optimal route is 5500 separate roads
• Total length ~770km There’s a difference

• Found in less than 3 seconds
bet ween least
number of hops and
least cost.

๏ Uses A* “best ﬁrst” search
31

Routing with Neo4j - using Neo4j Graph-Algos
# The cost evaluator - for choosing the best next node
class GeoCostEvaluator
include EstimateEvaluator
def getCost(node, goal)
straight_path_distance(
node.getProperty("lat"), node.getProperty("lon"),
goal.getProperty("lat"), goal.getProperty("lon") )
end
end

# Instantiate the A* search function
path_finder = AStar.new( Neo4j::instance,
RelationshipExpander.forTypes(
DynamicRelationshipType.withName("road"),
Direction::BOTH ),
DoubleEvaluator.new("length"), GeoCostEvaluator.new )

# Find the best path between New York City and San Francisco
best_path = path_finder.findSinglePath( NYC, SF )
32

Newest addition: Neo4j lets you REST
๏ Hello Neo4j REST server - Neo4j no longer needs to be embedded
๏ Opens up Neo4j to your favorite platform (even if that isn’t Java)
• PHP, .NET, et.c. - libraries already exists!
• http://wiki.neo4j.org/content/Getting_Started_REST
๏ Uses JSON for state transfer + browsable HTML for introspection
๏ Atomic modiﬁcation operations
๏ Brand new declarative traversal framework
• Extensible using your favorite scripting language
‣javascript is included. Jython, JRuby, et.c. supported
33

Other cool Graph Databases
๏ Sones GraphDB
• Graph Query Language - a SQL-like query language for graphs
๏ Franz Inc. AllegroGraph
๏ HypergraphDB
๏ InfoGrid
๏ Twitter’s FlockDB
• Optimized for the Twitter use case - one level relationships
๏ Interestingly we all have different approaches
34

Up until recently there was
only one Database, the
RDBMS.
The days of a single database
that rules all is over.

One database to rule them all

Image credits: The Lord of the Rings, New Line Cinema

35

Use best suited storage for each kind of data
The era of using
RDBMSes for all
problems is over.
Instead we should use
the database most
suited for the problem
at hand.

Image credits: Unknown :’( 36

Polyglot persistence
... we could even use
multiple databases in
conjunction, and let
each database handle
the things it does best.

Document
{...}

{...}

{...}
37

SQL && NOSQL

Document
{...}

{...}

All databases are welcome!
SQL and NOSQL - it is Not Only SQL! {...}
38

Finding out more
๏ http://neo4j.org/ - project website
‣http://api.neo4j.org/ and http://components.neo4j.org/
‣http://wiki.neo4j.org/ - HowTos, Tutorials, Examples, FAQ, et.c.
‣http://planet.neo4j.org/ - aggregation of blogs about Neo4j
๏ http://neotechnology.com/ - commercial licensing
๏ http://twitter.com/neo4j/team - follow the Neo4j team
๏ http://nosql.mypopescu.com/ - good source for news on NOSQL
monitors Neo4j and other NOSQL solutions
๏ http://highscalability.com/ - has published a few articles about Neo4j
39

Buzzword summary http://neo4j.org/

Semi structured
SPARQL
AGPLv3
ACID transactions
Open Source

Object mapping Gremlin Shortest path
In-Graph indexes NOSQL
A* routing
whiteboard friendly
RESTful
Traversal
Query language

Embedded
Beer
Schema free
Software Transactional Memory
Right tool for the right job
Scaling to complexity
Free Software

40

http://neotechnology.com

NOSQLEU - Graph Databases and Neo4j

More Related Content

What's hot (20)

Viewers also liked (20)

Similar to NOSQLEU - Graph Databases and Neo4j (20)

More from Tobias Lindaaker (9)

Recently uploaded (20)

NOSQLEU - Graph Databases and Neo4j