virtualizing relational databases as graphs: a multi-model approach

Smart Data for Smarter Business | © 2016 Capsenta | capsenta.com

Virtualizing Relational Databases as Graphs: a multi-model approach

Juan F. Sequeda, Ph.DCo-FounderCapsenta

1

(i.e. Want Graphs? Have Relational? No Problem!)

Smart Data/Graphorum Conference – February 1, 2017


Background

• Ph.D. from UT Austin Computer Science• Research on Integrating Relational Databases with Semantics and Graphs

• Editor W3C Standard on Mapping Relational Databases to Graphs

• Co-‐Founder Capsenta, spinout from UTCS

2


Takeaway

• Relational Databases can be virtualized as Graphs!

• Do you really need to create another database?

3


Graphs are Cool!

5


Flexible

6

:US_Constitution_1992/section/123

“Excessive bail shall not be required, nor

excessive fines imposed, nor cruel and unusual punishments inflicted.”

:text

:US_Constitution_1992 “United States of America 1789 (rev. 1992)”

:text

:isSectionOf

:Cruelty:hasTopic

“Prohibition of cruel or degrading treatment”

:label

“inhumane treatment”

:keyword


Integration

7


“Excessive bail shall not be required, nor excessive fines imposed, nor cruel and unusual punishments

inflicted.”

:text


:isSectionOf

:Cruelty:hasTopic


:label


:keyword

:text

:EighthAmendment_USConstitution :Farmer_vs_Brennan

:lawsApplied

“A prison official’s ‘deliberate indifference’ to a substantial risk of a serious harm to an inmate

violates the Eighth Amendment”

:holding:sameAs

:Prisons_in_Indiana

:LGBT_right_case_laws

:subject :subject


Data and Metadata are One

8



inflicted.”

:text


:isSectionOf

:Cruelty:hasTopic


:label


:keyword

:text

:Section :Constitution:Topic

:Rights_and_Duties

:Physical_Integrity_Rights

:subClass

:subClass

:subClass

:hasTopic :isSectionOf

:type

:type


Common denominator

9

<constitution id=“US_Constitution_1992”><section id="US_Constitution_1992/section/123">

<text>Excessive bail shall ...</text></section><topic>Cruelty</topic>

</constitution>

“Excessive bail shall not be required, nor excessive fines imposed, nor cruel and unusual punishments inflicted.”

id text topic123 Excessive bail shall… Cruelty



inflicted.”

:text:Cruelty

:hasTopic

XML Text

Tabular


Traversal, Navigation, Reachability

10



inflicted.”

:text


:isSectionOf

:Cruelty:hasTopic


:label


:keyword

:text

:EighthAmendment_USConstitution :Farmer_vs_Brennan

:lawsApplied

“A prison official’s ‘deliberate indifference’ to a substantial risk of a serious harm to an inmate

violates the Eighth Amendment”

:holding:sameAs

:Prisons_in_Indiana

:LGBT_right_case_laws

:subject :subject


Semantics

11



inflicted.”

:text:Cruelty

:hasTopic


:label


:keyword


:subClass

:hasTopic


(Summary) Why are Graphs Cool?

12

• Flexible• Integration• Data and Metadata are one

• Common Denominator• Traversal, Navigation, Reachability

• Semantics

ACM Computing Surveys 2008


Are Relational Databases cool?

13


Flexible

14

id attr1 attr2 attr3 attr4 … attrn …

id attribute value

id attr1 val1 attr2 val2 attr3 val3

id valueattr1

id valueattr2

id valueattr3

Copeland and Khoshafian. A decomposition storage model. SIGMOD 1985

Agrawal et al. Storage and Querying of E-‐Commerce Data. VLDB 2001


Integration

15

ExtractTransformLoad


Data and Metadata are One

16

CREATE TABLE Person (id int primary key,Name varchar not null,…

)

Company closePrice closeDate

IBM 130 1/15/2016

MSFT 50.99 1/15/2016

IBM MSFT closeDate

130 50.99 1/15/2016

closePrice closeDate130 1/15/2016

closePrice closeDate

50.99 1/15/2016

IBM MSFT

-‐ Krishnamurthy et al. Language features for interoperability of databases with schematic discrepancies. SIGMOD 1991-‐ Lakshmanan et al. SchemaSQL -‐ A Language for Interoperability in Relational Multi-‐database Systems. VLDB 1996


Common denominator

• Social Network• Hierarchical Data

17


Traversal, Navigation, Reachability

• Write a bunch of Joins• Recursion

18


Semantics

• Views• Triggers

19


(Summary) Are Relational Databases Cool?

• Flexible• Integration• Data and Metadata are one• Common Denominator• Traversal, Navigation, Reachability• Semantics

20


Ditch your Relational Database and move to Graphs!?

Feasible?

21


Query Federation

What if …

22

Virtualize Relational Databases as Graphs

Keep your legacy data in the RDBMS

Run graph queries over the virtual graph data

Add new data that doesn’t fit into the schema into a separate graph

Federate queries over Virtualized Graph and the Real Graph


Query Federation

What if …

23


What type of graphs are we talking about?

24


Property Graphs vs RDF Graphs

25

:Bob :Alicefoaf:knows

“Bob Smith”

foaf:name

“Alice Smith”

foaf:name

id1 id2

knowskey value

name Bob Smith

key value

name AliceSmith

key value

since 2005

:g1

2005

:since

http://db-‐engines.com/en/ranking/graph+dbms http://db-‐engines.com/en/ranking/rdf+store


RDF: Resource Description Framework

• Graph Data Model• Subject (Node) – Predicate (Edge) – Object (Node)• W3C Standard for data on the web• URIs

26



:isSectionOf

:text


Schemas, Taxonomies, Ontologies

27

:Cruelty


:Rights_and_Duties


:subClass

:subClass

:subClass



SELECT ?x ?titleWHERE{ ?x :hasTopic :cruelty.?x :text ?t

}

SPARQL Graph

RDF Graph

:USConst/146

:cruelty

:prerel

“Excesses bail shall not be required … “

:text

?x :cruelty:hasTopic

?text

SPARQL Protocol and RDF Query Language

28

SPARQL is a Query Language• Graph query language for RDF• Match SPARQL graph with RDF graph• Much more features in SPARQL 1.1: Property Paths

SPARQL is a Protocol• Send query over HTTP GET or POST• Response to a query is either in XML, JSON or CSV format

:text


SPARQL Graph

RDF Graph

:USConst/146

:cruelty

:prerel

“Excesses bail shall not be required … “

:text

?x :cruelty:hasTopic

?text

:text

SPARQL Protocol and RDF Query Language

29

?x ?text

:USConst/146 “Excesses bail shall not be required … “

SELECT ?x ?titleWHERE{ ?x :hasTopic :cruelty.?x :text ?t

}


Virtualizing Relational Databases as Graphs

30


Relational Database to RDF (RDB2RDF)

31

ID NAME AGE CID

1 Alice 25 100

2 Bob NULL 100

Person

CID NAME

100 Austin

200 Madrid

City

<Person/1>

<City/100>

Alice25

Austin

<Person/2>

Bob

<City/200> Madrid

foaf:namefoaf:name foaf:age

rdfs:label

rdfs:label

foaf:based_near

Mapping


W3C RDB2RDF Standards

• W3C Standards to map Relational Data to RDF

• A Direct Mapping of Relational Data to RDF– Default automatic mapping of relational data to RDF

• R2RML: RDB to RDF Mapping Language– Customizable language to map relational data to RDF

32


RDF

W3C Direct Mapping

33

RelationalDatabase

Direct MappingEngine

Input: Database (Schema and Data)Primary KeysForeign Keys

OutputRDF graph

https://www.w3.org/TR/rdb-‐direct-‐mapping/


W3C Direct Mapping Result

34

ID NAME AGE CID

1 Alice 25 100

2 Bob NULL 100

Person

CID NAME

100 Austin

200 Madrid

City

<Person/ID=1>

<City/CID=100>

Alice25

Austin

<Person/ID=2>

Bob

<City/CID=200> Madrid

Person#Name Person#Age

City#Name

City#Name

Person#ref-‐CID

Direct Mapping

Person#Name


R2RML

35

R2RMLEngine

R2RMLFile

:Cruelty


:Rights_and_Duties


:subClass:subClass

:subClass


RDF

RelationalDatabase

Target Schema

https://www.w3.org/TR/r2rml/


<TriplesMap1>a rr:TriplesMap;

rr:logicalTable [ rr:tableName”Person" ];

rr:subjectMap [ rr:template"http://www.ex.com/Person/{ID}";

rr:classfoaf:Person ];

rr:predicateObjectMap [ rr:predicate foaf:based_near ; rr:objectMap [

rr:parentTripelMap <TripleMap2>;rr:joinCondition [

rr:child “CID”;rr:parent “CID”;

]]

].

<TriplesMap2>a rr:TriplesMap;

rr:logicalTable [ rr:tableName ”City" ];

rr:subjectMap [ rr:template "http://ex.com/City/{CID}";rr:class ex:City ];

rr:predicateObjectMap [ rr:predicate foaf:name; rr:objectMap [ rr:column ”TITLE" ]

].

Example R2RML

36


Graph Data Virtualization

37

SPARQL

RDBMS Graph

SQL

SQL Results

SPARQLResults

R2RML Mappingby


Performance of Ultrawrap

• Reuse existing relational infrastructure– 30+ years of optimizations– Semantic Query Optimizations

• Result: SPARQL as fast as SQL

38

Sequeda J. Integrating Relational Databases with the Semantic Web. IOS Press. 2016http://www.iospress.nl/book/integrating-‐relational-‐databases-‐with-‐the-‐semantic-‐web/


Demo

39


Semantic Data Virtualization as a Bridge for BI

40

HIVEImpala, etc

OracleSQL

Server

PostgresUnstructured

Semi-‐Structured

Enterprise Knowledge Graph

Search ReportsAPI BI Connectors(Tableau, …)


Takeaway: Tipping Point

41

Relational Database

Graphs

• Flexible• Integration• Data and Metadata are One• Common Denominator• Traversal, Navigation, Reachability • Semantics

Do you really need another database?

Relational Databases can be virtualized as Graphs


THANK YOU

Juan Sequeda, Ph.DCo-‐Founder – [email protected]@juansequeda

42

virtualizing relational databases as graphs: a multi-model approach

Technology