1. What’s New in Oracle Database 12c Graph Database
Xavier Lopez, Ph.D. Senior Director
Zhe Wu, Ph.D. Architect

2. Agenda
Graph Database Strategy
Customer Use Cases
Oracle Spatial and Graph RDF Graph Features
Future Plans

3. Graph Database Strategy
Support Graph Data Types…
…On all enterprise platforms
Oracle Database
Oracle NoSQL Database
Oracle Big Data Appliance
Oracle Cloud

4. What Sets Us Apart? Scalability: Trillions of triples
Transactional: Concurrent loading and updates with ACID properties
Security: OLS security labels at “triple” level (OLS).
Standards based: W3C
Manageable: Use existing DB tools, utilities and expertise
Multi-type support: graph, relational, search, geospatial …
Multi-platform: Relational database, NoSQL, Hadoop 4

5. RDF Graph v. Property Graph
Property Graphs
RDF Semantic Graphs
Use Case:
Social network analysis
Analytics:
Clustering, centrality, page rank, path finding
Analytics Execution
In-memory, In-database
Use Case:
Linked data, semantic metadata layer
Analytics:
pattern matching, Inferencing
Analytics Execution
In-database

6. RDF Semantic Graph feature of Oracle Spatial and Graph
For Oracle Database 12c

7. Two Application Use Cases
Linked Data
Entity Analytics
Find related content & relations by navigating connected entities
“Reason” across entities
Find related content & relations by navigating connected entities
“Reason” across entities
SPARQL pattern matching
Detecting related entities across large, sparse, disparate collections of data
Inferencing: Applying rules on asserted data
Unified metadata model for distributed data sources
Flexible model for sparse and evolving data
Validate semantic and structural consistency

8. Linked Data in Support of Distributed Data
Mid-Tier Server
Application 2
Application 3
Graph-based Metadata Layer
Application 1
SPARQL
W3C standard, flexible model for sparse and evolving data
Common vocabulary enables data integration & app development
Relational data stays in place, apps don’t need to change
Shared Ontologies
RDF Graph
SQL
Inventory Graph
Sales Graph
HR Schema
Inventory Schema
Sales Schema
HR Database
Inventory Database
Sales Database
Database Server

9. Linked Data in Enterprise
Access & Presentation Layer
Index
Semantic Graph model
Data Servers
Event Server
Hadoop Appliance
Content Mgmt
BI Server
Data Warehouse
Data Sources / Types
Human Sourced
Information
Machine Generated Data
Social Media
Subscription Services
Transaction Systems

10. Linked Data / Enterprise Metadata
Industries
Life Sciences
Finance
Media
Networks & Communications
Defense & Intelligence
Police
Hutchinson 3G Austria

11. Novartis Institutes for BioMedical Research (NIBR)
Business Challenge
Link database information on genes, proteins, metabolic pathways, compounds, ligands, etc. to original sources.
Increase productivity for accessing, sharing, searching, navigating, cross-linking, analyzing internal /external data
The Novartis Institutes for BioMedical Research (NIBR) is
the global pharmaceutical research organization for Novartis
committed to discovering innovative medicines to treat diseases
with high unmet medical need
With more than 6,000 scientists, physicians and business
professionals around the world, we have an open,
entrepreneurial and innovative culture that encourages
collaboration as we work to push the boundaries of science to
change the practice of medicine
Solution
Semantic integration layer on RDF graph
Rich domain-specific terminology (biology, chemistry and medicine) 1.6 M terms
Terminology Hub: 8 GB of referential data that cross-references between data repositories.

12. RDF Semantic Graph-based Applications
Linked Data
Entity Analytics
Find related content & relations by navigating connected entities
“Reason” across entities
Find related content & relations by navigating connected entities
“Reason” across entities
SPARQL pattern matching
Detecting related entities across large, sparse, disparate collections of data
Inferencing: Applying rules on asserted data
Unified metadata model for distributed data sources
Flexible model for sparse and evolving data
Validate semantic and structural consistency

13. Knowledge Management in Intelligence Domain
Extracted Entities & Relationships
Information Extraction
Feature Extraction, Term Extraction
Person: Abduwali Abdukhadir Muse
Nationality: Somalian
Country: UK
Group: Al Shabab
Ideology: Islamist
Person: ?
Nationality: Pakistani
Country: Pakistan
Group: ?
Person: Chehab Abdouljamid Bouyaly
Country: Morocco
Group: al Qaeda
Currently resides
Member of
Supports
Link ?
Has
Search, Presentation, Report, Visualization, Query
Financial Data
Telephone Records
RDF
Intelligence Ontologies
Internet Traffic
SQL/SPARQL
Enterprise Data
Spatial
images
Documents
Data Sources
Contents Repository
Databases
Web resources
Blogs, Mails, news, RSS feeds
National Intelligence Scenario

14. RDF Semantic Graph Features
Oracle Spatial and Graph
RDF Semantic Graph Features

15. Oracle Database 12c RDF Semantic Graph Database
Exadata ready
Compression & partitioning
Parallel load, inference, query
High availability
Label security: triple-level
W3C standards compliance
Semantic Indexing of text
Enterprise Manager
Native RDF graph data store
Manages billions of triples
Optimized storage architecture
Load /
Storage
SPARQL-Jena/Joseki, Sesame
SQL/graph query, B-tree indexing
Ontology assisted SQL query
Query
So let’s briefly cover the main features of Oracle Spatial and Graph option that make these kind of solutions possible. First, on the left side of this slide you’ll observe the three key services provided by the RDF graph database: storage, query and inferencing. These are three requirements for an RDF graph database. First, we must manage large amounts of graph data – in many cases into the 100s of billions of triples. Second, we support the SPARQL query language used to query RDF data. Third, we provide a native reasoning engine – in the database.
But set’s Oracle apart from competing RDF databases is the scalability and performance of the database when handling this type of graph data. And the reason is that we can leverage all of the database services, that are highlighted on the left side of the slide. Capabilities like: Exadata, compression, partitioning, parallell load, query and inference.
And finally, for manageability, we leverage Data Guard, enterprise manager and other important Oracle database utilities that well known and understood by developers.
RDFS, OWL2 RL, EL, SKOS
User-defined rules
Incremental, parallel reasoning
User-defined inferencing
Plug-in architecture
Reasoning
Semantic indexing framework
Integration with
OBIEE, Oracle R Enterprise
Oracle Data Mining
Analytics

16. Support for Apache Jena and OpenRDF Sesame
Leverage existing investments in open source frameworks
Provides application developers with:
Easy-to-use Java APIs to access Oracle databases and RDF files
A standard-compliant SPARQL web service endpoint (Joseki, Fuseki)
Data loading (RDF/XML, N-TRIPLES, N-QUADS, TriG ,Turtle)
JSON output
Oracle-specific extensions for query execution control and management

17. Relational to RDF Mapping
RDB to RDF Mapping
RDF views on relational tables
Enables SPARQL query on distributed resources
Views: Automatic and custom
Aligns with W3C RDB2RDF standard
No duplication of data and storage
RDF Views on Relational tables
Semantic graph queries provide a powerful capability to discover relationships through pattern matching on RDF graph data. RDF views extend semantic discovery to relational tables without requiring the conversion of relational tables to RDF triples. This removes the need to duplicate data and the associated storage previously required to perform RDF graph queries on relational data sets. Semantic graph queries and RDF views can also be used to enable data integration and discovery within and across relational schema and RDF graphs in Oracle Database. This simplifies semantic discovery workflows.
RDF views on relational tables, views, and SQL query results are supported. The W3C specifications for automatic mapping (called Direct Mapping), custom mapping (using the W3C R2RML language) and materializing views are supported. RDF views present relational data in RDF triple format so it can be queried using SPARQL and connected with other linked data and RDF graphs to relate and facilitate enterprise data integration.

18. Oracle Label Security Data Classification
Fine grained security through integration with Oracle Label Security
Model level security through GRANT/REVOKE privileges
Oracle Label Security - mandatory access control
Labels assigned to both users and data
Data labels determine the sensitivity of the rows or the rights a person must posses in order to read or write the data.
User labels indicate their access rights to the data records. 18

19. Core Inferencing Features
Forward-chaining based inference engine in the database
Native rulebases: RDFS, OWL 2 RL, OWL 2 EL, SKOS
Validation of inferred data
Proof generation
User defined inferencing
- Temporal reasoning, Spatial reasoning
Ladder Based Inference
- Fine grained security for inference graph
Integration with external OWL 2 reasoners (TrOWL, Pellet)

20. RDF Semantic Graph: Graph Visualization & Modeling Support
Semantic Modeling
Cytoscape
Protégé
Oracle Confidential – Internal/Restricted/Highly Restricted 20

21. Analyzing RDF with Oracle BI and Oracle Advanced Analytics

22. Oracle Partner Tools: (IO Informatics)

23. Oracle Partner Tools: Tom Sawyer Social Network Analysis

24. Manageability of RDF Semantic Graph
Built in support from Oracle Database utilities and tools
Ingest / Replicate / Recover
Tune / Analyze
Manage
Bulk load:
Apache Jena bulk loader
Oracle external tables &
SQL*Loader (Direct Path) w/ PL/SQL Bulk Load API
Replicate & recover:
Data Guard: physical standby
Data Pump: staging tables
Recovery Manager: RMAN
Tune load/ query/ inference:
Parallelism
Btree indexing triple/quad
Typed literals indexing
SPARQL query hints
Statistics gathering
Dynamic Sampling
Analyze performance:
Enterprise Manager: view optimizer plans, monitor execution / resource usage
Control query execution:
in database & Jena client
Create & monitor graph w/ SQL Developer:
Semantic Network
Models, virtual models
Btree indexes
Rule bases
Entailments
Security data labels
Semantic index policies
Bulk Load/Replicate/Recover
Apache Jena: load 100,000s RDF/OWL data files – Turtle, RDF/XML, N-Triples, RDF/JSON, TriG, N-Quads - into Oracle Database using prepareBulk / completeBulk methods in OracleBulkUpdateHandler Java class
External Tables / SQL*Loader: load N-TRIPLE / N-QUAD input files into a staging table and loaded into the graph w/ syntax correctness & duplicates checking and default index creation (PSCGM)
Data Guard: physical standby block level replication
Data Pump: Import/Export can move staging tables created from files or results of SPARQL queries on the database between databases
Recovery Manager: RMAN backups & recovery operate at the tablespace/datafile/database level & therefore support RDF data
Tune/Analyze
Parallelism is supported for load, query and inference
Btree indexing enables non-unique indexing in any combination of subject-predicate-object-named graph- model
Typed literals indexing: index XML, Text, & Spatial OGC Well Known Text literal types to filter SPARQL queries
SPARQL query hints for index choice, type (hash/nested loop) & order of join, order of query base graph pattern evaluation, etc.
Statistics: Oracle Database optimizer analysis of statistics critical to performance of SPARQL queries & OWL inference
Dynamic sampling: may produce optimal execution plans over static sampling given inherent flexibility of the RDF data model
Enterprise Manager: Analyze optimizer query plan & performance of execution / resource usage for loading, querying, entailments
Control query execution: Timeout, abort, SPARQL2SQL translation, hints in SPARQL syntax, property path, result cache, mid-tier cache, user-defined functions
Create & monitor graph: use SQL Developer to run RDF Semantic Graph Pl/SQL APIs & issue SQL queries to view views describing graph components

25. Open Geospatial Consortium: GeoSPARQL Support
Defines a Vocabulary for Spatial Query Patterns
Classes
Spatial Object, Feature, Geometry
Properties
Topological relations
Links between features and geometries
Datatypes for geometry literals
ogc:wktLiteral, ogc:gmlLiteral
Query Functions
Topological relations, distance, buffer, intersection, …

26. Graph Support on Oracle NoSQL DB
Brings horizontal scalability to RDF graph applications
RDF Graph for Oracle NoSQL
RDF Graph support in Oracle NoSQL Database Enterprise Edition
High performance Key Value store
SPARQL 1.1 access to graph data
Jena & Joseki SPARQL Web Services
Massive horizontal scalability
Support for World Wide Web Consortium (W3C) Semantic Web standards

27. When to Consider a NoSQL Database for Graphs
Horizontal scalability, low query latency/cost, ease of install & management
RDF Graph for Oracle NoSQL
High volume, simple queries (low latency)
Queries aggregating over most of the graph (e.g. what are the hobbies of the most popular people in the network)
Frequent, large-scale updates
Large Data Centers

28. Quick Steps to Get Started

29. Quick Steps to Get Started
Using SQL/PLSQL APIs
exec create_sem_model
insert/delete triples, bulk load, run SEM_MATCH, create_entailment, …
Initialize
Creating a tablespace ‘ts’
Run as SYS in SQL*Plus
exec sem_apis.create_sem_network(‘ts’)
Run as SYS (for only) in SQL*Plus
exec mdsys.enableGeoRaster;
Install Oracle Database 12c or
Use a Prebuilt VM from OTN
Using Java APIs
Load/Query/Inference through
GraphOracleSem, DatasetGraphOracleSem,
OracleBulkUpdateHandler, …
Configure Joseki/Fuseki
web service endpoint
SPARQL Query
SPARQL Update
REST APIs

30. Quick Steps to Get Started
Using SQL/PLSQL APIs
exec create_sem_model
insert/delete triples, bulk load, run SEM_MATCH, create_entailment, …
Initialize
Creating a tablespace ‘ts’
Run as SYS in SQL*Plus
exec sem_apis.create_sem_network(‘ts’)
Run as SYS (for only) in SQL*Plus
exec mdsys.enableGeoRaster;
Install Oracle Database 12c or
Use a Prebuilt VM from OTN
Using Java APIs
Load/Query/Inference through
GraphOracleSem, DatasetGraphOracleSem,
OracleBulkUpdateHandler, …
Configure Joseki/Fuseki
web service endpoint
SPARQL Query
SPARQL Update
REST APIs

31. Quick Steps to Get Started
Using SQL/PLSQL APIs
exec create_sem_model
insert/delete triples, bulk load, run SEM_MATCH, create_entailment, …
Initialize
Creating a tablespace ‘ts’
Run as SYS in SQL*Plus
exec sem_apis.create_sem_network(‘ts’)
Run as SYS (for only) in SQL*Plus
exec mdsys.enableGeoRaster;
Install Oracle Database 12c or
Use a Prebuilt VM from OTN
Using Java APIs
Load/Query/Inference through
GraphOracleSem, DatasetGraphOracleSem,
OracleBulkUpdateHandler, …
Configure Joseki/Fuseki
web service endpoint
SPARQL Query
SPARQL Update
REST APIs

32. Performance

33. Oracle Spatial and Graph - LUBM 200K on 3-Node RAC X2-4 Load, Inference and Query Performance
The LUBM 200K Graph has 48+ Billion triples (edges)
Original graph has 26.6 Billion unique triples (quads)
Inference produced another 21.4 Billion triples
Data Loading Performance
Triples Loaded and Indexed Per Second (TLIPS): 273K
Inference Performance
Triples Inferred and Indexed Per Second (TIIPS): 327K
SPARQL Query Performance
Query Results Per Second (QRPS): K
48+ Billion edges graph
Setup:
Hardware: Sun Server X2-4, 3-node RAC
- Each node configured with 1TB RAM, 4 CPU 2.4GHz 10-Core Intel E7-4870)
- Storage: Dual Node 7420, both heads configured as: Sun ZFS Storage 7420  4 CPU 2.00GHz 8-Core (Intel E7-4820)
256G Memory 4x SSD SATA2 512G (READZ) 2x SATA 500G 10K. Four disk trays with 20 x 900GB 4x SSD 73GB (WRITEZ)
Software: Oracle Database , SGA_TARGET=750G and PGA_AGGREGATE_TARGET=200G
Note: Only one node in this RAC was used for performance test. Test performed in April 2013.

34. Degrees of Parallelism
Oracle Spatial and Graph – LUBM 4400K on Exadata X4-2 Load, Inference and Query Performance
Degrees of Parallelism
Data set
Load
(B triples/hr)
OWL Inference
Query
(B answers/hr)
256*
LUBM 4400K
605.4B / 115.2hrs
B / 86hrs 30m
92.5B / 22.5 hrs
Data Loading Performance
Triples Loaded and Indexed Per Second (TLIPS): 1.420M
Inference Performance
Triples Inferred and Indexed Per Second (TIIPS): M
SPARQL Query Performance
Query Results Per Second (QRPS): M
1.08 Trillion edges graph
Setup:
Open cursors = 1000
Processes = 1000
SGA = 132GB, PGA = 100GB
32K blocksize was given to all graph tablespaces
TEMP group was created with 3 bigfile tablespaces
Test performed in Aug/Sept 2014.
Exadata X4-2 High capacity full rack
ZS3-2 with 2 controllers, 8 trays of disk
Eight compute nodes of Exadata
Oracle DB standard install of Exadata
* A mix of DOP used: 296, 256, 192
Oracle Confidential – Internal

35. Best Practices in Solving Performance Issues
When there is an underperforming SQL in RDF data loading, inference, or query operations, check:
Have you gathered statistics?
APIs: export_model_stats,export_entailment_stats, export _network_stats, import_model_stats, import_entailment_stats, import_network_stats
Have you tried parallel execution?
Balanced hardware is key.
Have you tried dynamic sampling? (Level 6, 8, 11)
Is there a lack of indexes (including text index)?
DO NOT just add indexes without careful & thorough testing

36. Best Practices in Solving Performance Issues (2)
When there is an underperforming SQL in RDF data loading, inference, or query operations, check:
Have you looked at the plan?
Is it possible to write the same query in a different way?
Is it possible to simplify?
Simpler queries  Better chance to find more efficient ways to execute
Tweak plan through hints
Send a small, reproducible test case with the execution plan to Oracle Support or post it on the Forum

37. Best Practices in Solving Performance Issues (3)
Find the top thread(s) in Java VM
Are there excessive GC activities?
Try –XX:+UseParallelGC, -XX:+UseConcMarkSweepGC, …
Has the heap size been set properly?
Try larger heap size, analyze heap by performing a heap dump
Send a small, reproducible test case with the thread dump to Oracle Support or post it on the Forum

38. Cool Ongoing Activities:
Enable Oracle Cloud Services: Oracle Social Network
Integration with Oracle business applications and middleware
Ongoing support for RDF Graph on all major platforms
Relational Database
NoSQL Database
Big Data (Hadoop)
Cloud

40. Appendix

41. W3C Semantic Technology Stack
Core Technologies
URI
Uniform resource identifier
RDF
Resource description framework
RDFS
RDF Schema
OWL
Web ontology language
RIF and SPARQL are also standards 41

42. Subject Predicate Object
What is RDF
A graph data model for web resources and their relationships
The graph can be serialized into
- RDF/XML, N3, N-TRIPLE, …
Construction unit: Triple
(or assertion, or fact)
< <:produces> <:mp3>
Quads (named graphs) add context, provenance, identification, etc. to assertions
< <:produces> <:mp3 > <:ProductGraph>
“CA”
Subject Predicate Object