This document discusses best practices for using Hadoop as an enterprise data hub. It provides an overview of how big data is driving new analytical workloads and the need for deeper customer insights. It discusses challenges with analyzing new sources of structured, unstructured and multi-structured data. It introduces the concept of a Hadoop enterprise data hub and data refinery to simplify access to new insights from big data. Key components of the data hub include a data reservoir to capture raw data from various sources, a data refinery to cleanse and transform the data, and publishing high value insights to data warehouses and other systems.
2. 2
About Mike Ferguson
Mike Ferguson is Managing Director of Intelligent
Business Strategies Limited. As an analyst and
consultant he specialises in business
intelligence, data management and enterprise
business integration. With over 32 years of IT
experience, Mike has consulted for dozens of
companies, spoken at events all over the world
and written numerous articles. Formerly he was
a principal and co-founder of Codd and Date
Europe Limited – the inventors of the Relational
Model, a Chief Architect at Teradata on the
Teradata DBMS and European Managing
Director of DataBase Associates.
www.intelligentbusiness.biz
mferguson@intelligentbusiness.biz
Twitter: @mikeferguson1
Tel/Fax (+44)1625 520700
3. The Hadoop Data Refinery and Enterprise Data
Hub
Mike Ferguson
Managing Director
Intelligent Business Strategies
June 2014
4. 4
Topics
! Data warehousing and the evolution of ETL processing
! New data and new analytical workloads
! Big data use cases driving business agendas
! The unprecedented demand for customer insight
! Challenges with new big data sources
! Beyond the data warehouse – new platforms for new analytical
workloads
! The role of Hadoop in the modern analytical ecosystem
! Introducing the Hadoop enterprise data hub and data refinery
! Simplifying access to new big data insight using SQL on Hadoop
! Integrating Hadoop into your analytical ecosystem
5. 5
For Many Years The Traditional Data Warehouse and BI
Environment Has Been Used For Analysis & Reporting
Operational
systems
web
P
o
r
t
a
l
Employees
Partners
Customers
BI
Tools
Platform
Data
Integration/DQ
Reports &
analytics
Data warehouse
& data marts
DW
6. 6
The Evolution of Data Integration in Data
Warehousing – From Hand Coded to ETL to ELT
Hand coded ETL programs
DW
Hand
coded
programs
ETL Servers
DW
ETL
Servers
ELT processing
Generated
SQL ELT
processing
DWEvolution of Data Warehousing
MPP RDBMS systems
7. 7
Sales
Product line n
Product line 4
Product line 3
Product line 2
Product/
service line 1
Marketing
Service
Credit
Verification
HR
Finance
Planning
Procurement
SupplyChain
Suppliers
Front Office BackOffice
Operations
Customers
New Data Sources Have Emerged Inside And Outside
The Enterprise That Business Now Wants To Analyse
E.g. RFID tag
sensor
networks
weather data
Data volume
Data variety
Number of sources
Data volume
Data velocity
8. 8
Popular Big Data Analytic Applications – Web Data
! Clickstream analytics
• Site navigation behaviour (session) analysis
– Paths to buy, paths to abandonment, what else
they looked at
– Improve customer experience and conversion
– Associate clicks with customers & prospects
! Social network influencer analysis
• Graph analytics for influencer behavioural impact
analysis
• ‘Target the influencer’ marketing campaign
effectiveness
9. 9
Popular Big Data Analytic Applications – Sensor Data
For Improving Process Efficiency and Optimisation
! Sustainability analytics e.g. energy optimisation
! Supply/distribution chain optimisation
! Asset management and field service optimisation
! Manufacturing production line optimisation
! Location based advertising (mobile phones)
! Grid health monitoring
• Electricity, water, mobile phone cell network…
! Smart metering (collect data every 15 minutes)
! Fraud
! Healthcare – ITC vital signs, fit bits,….
! Traffic optimisation
" WHAT ARE YOU PREPARED TO INSTRUMENT?
E.g. RFID tag
10. 10
Popular Big Data Analytic Applications
– Unstructured Data
! Case management
! Fault management and field
service optimisation
! “Voice of the customer”
! Sentiment analytics
! Competitor analysis
! Media coverage analysis
! Improve pharma drug trials
" Unstructured content is hard to
analyse
How much is TEXT worth to
your business?
11. 11
Big Data Analytics - Industry Use Case Examples
Industry Use Case Examples
Financial
Services
Improved risk decisions, KYC customer insight, auto programmatic
trading, 360 view of financial crime, pre-trade decision support,
real-time trade & corp action tagging for compliance and RT P&L,
grow security services outsourcing, Reference Data Exchange
Utilities Smart meter data analysis, pricing elasticity analysis, customer
loyalty, sustainability, asset management
Telecommunic
ations
Customer Churn, Network optimization analysis from device,
sensor and GPS inputs, monetization of GPS and data
Manufacturing Sensor data for next generation ‘smart’ products, production line
optimisation, improved customer service and improved field
service, distribution chain optimization, asset management
Insurance “How you drive” insurance (sensors to reduce risk), broker
document analysis (risk assessment)
Government Smart cities (e.g. transportation optimisation), anti-terrorism, law
enforcement
Logistics Distribution optimisation, route optimisation,
12. 12
More Data Is Required To Get A Deeper
Understanding of Customers
! We now need
• Transaction data
• Data from touch points you own
• Data from the touch points you don’t own
• Interaction data
– Need to look at Inbound interactions Vs outbound interactions
– Social interactions
• Master data
• Professional data e.g. profiles on LinkedIn
• Internal and external event data
• Competition data…..
! Then use analytics to understand and predictive desire and
propensity e.g. propensity to churn
13. 13
Top Priorities - Improving Customer Experience Via
Time Series Analysis of All Customer Interactions
OMNI channel – analyse all customer
interactions across all channels
identity
data
behavioural
data
social
data
Customer “DNA”
14. 14
identity
data
behaviou
ral data
social
data
Customer “DNA”
Customer Experience Management - Understanding Customer
On-Line Behaviour is Mission Critical to Retention and Growth
! Important new data sources for analysis for customer ‘DNA’
• Clickstream data from web logs
• Sentiment and social network influencer data
New competitors
More choice
Voice of the customer
On the web the
customer is king
On the
move
Easy to find
15. 15
Today Both Structured And Multi-Structured Data Are
Needed For Deeper Insight
Multi-
structured
data
Click stream web log data
Customer interaction data
Social interaction data
Sensor data
Rich media data (video, audio)
External content
Documents
Internal web content
Seismic data (oil & gas)
Structured
data
OLTP system data
Data warehouse data
Personal data stores e.g.
Excel, Access
Often un-modelled and may
not be well understood
Often a schema is defined
and data is well understood
Data characteristics are changing
- Companies must deal with volume,
variety and velocity
16. 16
Big Data Analytics Challenges Include The Analysis of
Unstructured, Semi-structured and Structured Data
{ "firstName": ”Wayne",
"lastName": ”Rooney",
"age": 25,
"address": {
"streetAddress": "21 Sir Matt Busby Way",
"city": ”Manchester”,
“country”: “England”,
"postalCode": “M1 6DY”
},
"phoneNumbers": [
{ "type": "home”,
"number": ”0161-123-1234”
},
{
"type": ”mobile",
"number": ”07779-123234”
}
]
} JSON data
Text data
Image Data
Makes analysis more complex with new analytics and visualisations needed
17. 17
Increased Data and Analytical Complexity Has Created
A Need For A New Role – The Data Scientist
Image source: Wikipedia
Data Science is the process of investigative / exploratory analysis of
multi-structured data to discover and produce new business insights
Image source:
www.computing.co.uk
18. 18
People In Different Roles In The Analytical Landscape
Need To Work Together To Deliver Business Value
Exploratory analysis
Predictive / statistical
model producer
Business Analyst
Business Manager /
Operations worker /
Customer
Data Scientist
Model consumer
Data visualisation
Information Producer
• Build reports
• Build and publish
dashboards
Information consumer
Decision maker
Action taker
Strategic
Business
Objective
Priority KPI Current
KPI
Value
What is
+1%
worth?
KPI
Target
Executive
Accountable
Business
Initiatives
(projects)
Budget
Allocation
Action
Plan
1 $$$ Project
Project
Project
£ x Million
2
3
4
Business Strategy – strategic objectives and targets including sustainability targets
sandbox
19. 19
Data Science Produces New Insights For Business Analysts
Who Produce Actionable BI For Front Office Decision Makers
Business Analyst
Marketing Manager /
Marketing, Sales and
Service workers
Data Scientist
Data Quality
Forecasting
Segmentation
Models
Customer Lifetime
Value
Social
Network
Strategy
Creation
Performance
& Effectiveness
Reporting
Direct Mail
Understand
Customer
Behavior
& Navigation
Marketing
Performance &
Reporting
Campaign
Planning
Financial
Planning
Creative
Materials
Marketing
Attribution
Operations
Management
Channel
Efficiency
Sentiment
& Influence
Dynamic
Content
Re-marketing
Web
Call Center
Live Event
Broadcast Media
Mobile/ SMS
Social
Email
Industry Specific
Big Data Analytics
Traditional DW/BI
Workflow
& Approvals
New insights Actionable BI
20. 20
Big Data Analytics Has Taken Us Beyond The
Traditional DW – New Big Data Analytical Workloads
1. Analysis of data in motion
2. Complex analysis of structured data
3. Exploratory analysis of un-modeled multi-structured data
4. Graph analysis e.g. social networks
5. Accelerating ETL and analytical processing of un-
modeled data to enrich data in a data warehouse or
analytical appliance
6. The storage and re-processing of archived data
21. 21
The Changing Landscape – We Now Have Different
Platforms Optimised For Different Analytical Workloads
Big Data workloads result in multiple platforms now being needed for
analytical processing
Streaming
data
Hadoop
data store
Data Warehouse
RDBMS
NoSQL
DBMS
EDW
DW & marts
NoSQL DB
e.g. graph DB
Advanced Analytic
(multi-structured data)
mart
DW
Appliance
Advanced Analytics
(structured data)
Analytical
RDBMS
Graph
analysis
Investigative
analysis,
Data refinery
Traditional
query,
reporting &
analysis
Real-time
stream
processing &
decision m’gmt
Data mining,
model
development
22. 22
Hadoop Is A Key Platform In Big Data Analytics
– Data Can Be Accessed Via Multiple APIs
Java MapReduce
APIs to HDFS,
HBase, Cascading
file file file file file
file file file file file
file file
file file
webHDFS
(An HTTP
interface to
HDFS has
REST APIs)
HDFS
file
file
file
file
YARN
PIG latin
scripts
SQL
Vendor SQL on
Hadoop engine
MapReduce
Application
index
indexIndex
partition
SQL
BI Tools &
Applications
Storm
Application
YARN
Tez or SparkMapReduce HBase
HDFS API
23. 23
Defacto Standard APIs Allow Hadoop Components To Be
Replaced e.g. Faster, More Secure File System Than HDFS
Java MapReduce
APIs to HDFS,
HBase, Cascading
webHDFS
(An HTTP
interface to
HDFS has
REST APIs) file file file file file
file file file file file
file file
file file
file
file
file
file
Vendor Specific File System (e.g. )
YARN
HDFS API
PIG latin
scripts
index
indexIndex
partition
Storm
Application
YARN
MapReduce HBase
MapReduce
Application
SQL
Vendor SQL on
Hadoop engine
SQL
BI Tools &
Applications
Tez or Spark
24. 24
Apache Hadoop Components
Component Description
Hadoop HDFS A distributed file system that partitions files across multiple machines for high-throughput
access to application data – HDFS API allows vendors to replace HDFS with an alternative
Hadoop YARN" A framework for job scheduling and cluster resource management"
Hadoop
MapReduce
A programming framework for distributed batch processing of large data sets distributed
across multiple servers
Avro A serialization system that creates & reads files in a format containing both JSON data
definitions & the data itself for dynamic interpretation of the data by applications
Hive A data warehouse system for Hadoop that facilitates data summarization, ad-hoc queries,
and the analysis of large datasets stored in Hadoop-compatible file systems. Hive provides
a mechanism to project structure onto this data and query it using a SQL-like language
called HiveQL. HiveQL programs are converted into MapReduce programs
HBase HBase is an open-source, distributed, versioned, column-oriented store modeled after
Google' Bigtable.
Pig A high-level data-flow language for expressing Map/Reduce programs for processing and
analysing large HDFS distributed data sets
Mahout A scalable machine learning and data mining library
Oozie A service for running and scheduling workflows of Hadoop jobs (including Map-Reduce,
Pig, Hive, and Sqoop jobs)
Spark A general purpose engine for large scale data processing in-memory. It supports analytical
applications that wish to make use of stream processing, SQL access to columnar data and
analytics on distributed in-memory data
Zookeeper A high-performance coordination service for distributed applications
25. 25
The Role of Hadoop - Data Is Arriving Faster Than We
Can Consume It – How Good Is Your Filter?
F
D I
A L
T T
A E
R
Enterprise
Enterprise
systems
26. 26
New Requirement
– The Managed Hadoop Enterprise Data Hub
Parse & Prepare Data in Hadoop (MapReduce)
Transform & Cleanse Data in Hadoop (MapReduce)
Discover data in Hadoop
ELT
work
-flow
sandbox
other data
sandbox sandbox
Data Reservoir
(raw data)
Load data into Hadoop
Data
Refinery
New high
value Insights
(pub/sub)
EDW
Graph
DBMS
DW
appliance
contains clean,
high value data
XML,%
JSON%
Web
logs
27. 27
What’s In An Enterprise Data Hub?
! A managed data reservoir (raw data)
• Organised capture of multi-structured data
• Includes real-time data capture
• May include operational reporting
! A governed data refinery
• Data integration and cleansing at scale
• Analytical sandboxes to discover high value data
! Published, protected and secure high value insights
! Long-term storage of archived data from data warehouses
28. 28
file file file
file file file
file file
file file
file
file
Real-time Data Capture – E.g. MapR Allows Web Log
Data To Be Directly Streamed/Stored in Hadoop
MapR Direct Access NFSs allows
Web log files to be stored directly on
their Hadoop File System so that
click stream is captured in real-time
MapR Distribution
for Hadoop
Web Server
Direct Access NFS
web log
fileweb log
file
# mount localhost:/mapr /mapr
HDFS
Web Server
Web Server
29. 29
High Volume Data Capture
- Column Family Databases
! Suitable for fast capture of large amounts of sparse, volatile data
• Very fast capture and can hold vast amounts of data
• Billions of rows containing thousands or millions of columns
! Provide column-centric storage and wide de-normalised big
tables can also help simplify operational reporting if used with
SQL-on-Hadoop e.g. SQL access to HBase
! Allow you to
• Group together related columns into column families
• Design column families to optimize the most common queries
• Retrieve columnar data for multiple entities by iterating through a
column family
• Shard rows in a column family and distribute across many servers
• Create indexes and secondary indexes
• Support schema variance - columns in a column family can vary for
every row
30. 30
NoSQL Column Family Databases - HBase
Row 1 # Column A = value
Column B = value
Column C = value
Row 2 # Column X = value
Column Y = value
Column Z = value
Hbase Storage Architecture
Hmaster and several HRegionServers
Regions (partitions) created automatically as tables grow
Hbase allows applications to directly read and write data
31. 31
Column Families Can Be Stored In Different Files And
Queries Will Only Retrieve The Column Family Needed
Source: Data Access for Highly-Scalable Solutions : Using SQL, NoSQL, and Polyglot Persistence, McMurtry, Oakley, Sharp, Subramanian, Zhang
Portfolio.* means all
columns in the Portfolio
column family
Data about a customer and their
stock purchases are partitioned
vertically by column family
Column
family data
can also be
compressed
32. 32
Fast Data Capture – MapR-DB Is A High Speed
Version of HBase Built Into The MapR Data Platform
HBase API
Source: MapR
33. 33
Enterprise Data Hub – We Need A Data Refinery To
Process And Clean Complex Data
Image source: http://www.hollyfrontier.com/navajo/
34. 34
Evolution of Big Data Integration Is Following The
Same Cycle as it Did in Data Warehousing
Hand coded ETL programs
Hadoop
Hand
coded
programs
ETL Servers
Hadoop
ETL
Servers
ELT processing
Generated
MapReduce ELT
processing
HadoopEvolution of Big Data Integration
35. 35
Data Cleansing and Integration Tool
Scaling ETL In A Data Refinery By Generating Pig, Hive or 3GL
MapReduce Code for In-Hadoop ELT Processing
Extract Parse Clean Transform AnalyseLoad Insights
Option 1
ETL tool generates HQL
or convert generated
SQL to HQL
Option 2
ETL tool generates
Pig Latin
(compiler converts
every transform to
a map reduce job)
Note - Generating native MapReduce code instead of HiveQL or Pig Latin would
likely perform faster because there is no need to translate into MapReduce
Also HiveQL is a subset of SQL so check how ETL tools generating HiveQL do
complex transformations – HiveQL on its own may not be enough e.g. Hive UDFs?
Option 3
ETL tool generates
3GL MapReduce
code
36. 36
Need to Parse & Extract From Multi-Structured Data While
Integrating Data In A Big Data Environment
E-mail (semi-structured)
Text (unstructured)
ExtractParse TransformLoad …
37. 37
Sandboxes In The Data Refinery - Data Science Teams Need
To Conduct Exploratory Analysis on Multi-Structured Data
Click stream web log data
Customer interaction data
Social interaction data (e.g.
Twitter, Facebook)
Sensor data
Rich media data (video, audio)
External web content
Documents
Internal web content
Seismic data (oil & gas)
Investigative /
Exploratory
Analysis
C
R
U
D
Asset
Customer
Product
MDM System
EDW
mart
new
business
insights
sandbox
Multi-structured
data
Historical Data
archived DW datamaster data
Data Scientists
38. 38
In-Hadoop Analytics In A Data Refinery
– Example Technologies
! Hadoop MapReduce, Tez or Spark analytic
applications with custom analytics
• Pig, Java, Python, Scala, Cascading…..
! Hadoop MapReduce, Tez or Spark analytic
applications using pre-built Hadoop analytics e.g.
Mahout, Spark MLlib
• Several analytical algorithms for use in analysis
! Revolution Analytics RevoScaleR
! SAS Analytics and In-Memory Statistics for Hadoop
! … many more
Analytical
tools
Data
management
tools
39. 39
In-Hadoop Analytics:
- Mahout Supports A Number Of Analytic Techniques
! Collaborative Filtering
! User and Item based recommenders
! K-Means and Fuzzy K-Means clustering
! Mean Shift clustering
! Dirichlet process clustering
! Latent Dirichlet Allocation
! Singular value decomposition
! Parallel Frequent Pattern mining
! Complementary Naive Bayes classifier
! Random forest decision tree based classifier
https://cwiki.apache.org/confluence/display/MAHOUT/Algorithms
Now runs
on Spark as
well as
MapReduce
40. 40
Expediting The Data Refinery Process On Hadoop With
Automated Analysis – From ETL to Analytical Workflows
Parse & Prepare Data in Hadoop (MapReduce)
Transform & Cleanse Data in Hadoop (MapReduce)
Discover data in Hadoop
ELT
work
-flow
other data
Raw data
Load data into Hadoop
Data
Refinery
EDW
Graph
DBMS
DW
appliance
Automated Invocation of Custom Built & Pre-built
Analytics on Hadoop
contains clean,
high value data
New high
value Insights
(pub/sub)
41. 41
High Value Insights Produced In A Hadoop Data Hub Can Be
Brought Into A DW to Enrich What We Already Know
Cloud Data
HDFS
Extract
DW
D
IMap/ Reduce data
transformation
and analytics
applications
Transform
e.g. PIG, IBM JAQL
Cloud Data e.g. Deriving insight from huge
volumes of social web content on
sites like twitter, facebook. Digg,
mySpace, tripAdvisor, Linkedin….for
sentiment analytics
Hundreds of
terabytes up
to petabytes
new
insights
Operational
systems
42. 42
Making New Insights Available To Business Analysts
Via SQL Access To Big Data - Options
SQL
SQL access to
big data in
Hadoop
SQL
DW
data virtualisation server
SQL access to
big data via data
virtualisation
SQL
Analytical
RDBMS
SQL access to big
data in an
analytical RDBMS
streaming
data
SQL
SQL access to
streaming data in
motion
43. 43
Self-Service BI
BI Tool(s)
e.g, Visual Discovery tools
Business Analyst
or ‘budding’ Data
Scientist
personal &
office data
Predictive
models
community
Publish / Share
Consume /
Enhance /
Re-publish
Transaction
systems
DW
SQL Access to Hadoop Is Needed To Allow Hadoop Data To
Be Accessed By Users With Self-Service BI Tools
collaborate
HDFS / Hbase/ Hive
e.g. Hive interface
44. 44
SQL access
to Big Data?
Key Questions That May Influence If SQL Access to Big
Data Is A Good Choice or What SQL Option to Take
What kind of analysis?
Text analysis, Graph analysis,
Machine Learning, reporting
What kind of data type(s)
do you need to analyse?
- structured, unstructured, semi-
structured,
What kind of data volumes
do you want to analyse?
Is the data at rest or is it real-
time streaming data in motion?
What analytical functions
can you invoke on big
data from SQL?
Join with other data in
another data store?
How many concurrent users?
Performance and scalability
of complex queries and
analytical functions
(need parallelism)
Is the requirement for
interactive, exploratory,
or real-time analysis?
Data
Analytical Workload
45. 45
SQL On Hadoop Initiatives
Key Questions
What analytic functions
are provided?
How can analytic
functions be extended
Can you join to data
outside of Hadoop?
Are these SQL on
Hadoop options
suitable for reporting
and analysis, interactive
discovery, exploratory
analysis or all of these?
Vendor SQL on Hadoop Initiative
AMPlab (UC Berkeley) Shark (Forked Hive at V0.9) or SparkSQL
Apache Hadoop Hive
Actian Vortex (Actian Vector on Hadoop data nodes)
CitusDB CitusDB (uses external tables)
Cloudera Impala / Parquet
Concurrent Lingual (SQL on Cascading)
Hadapt Schemaless SQL
Hortonworks Stinger / ORC (Hive 13)
HP Vertica on Hadoop
IBM BigSQL (SQL on HDFS & HBase)
InfiniDB InfiniDB on Apache Hadoop
Jethro Data JethroData
MapR Apache Drill
Microsoft Hive 13
Pivotal HawQ (uses external tables via PFX)
Teradata SQL-H
Splice Machine Splice Machine (SQL Engine on HBase)
Salesforce.com Phoenix (SQL engine on HBase)
Attivio Active Intelligence Engine (SQL access to
search indexes on Hadoop data)
46. 46
SQL on Hadoop
– Apache Drill Can Access HDFS And HBase Data
BI Tool(s)
e.g, Visual Discovery tools
Business Analyst
or’ Data Scientist
Drill
Analytic Application
SQL SQL
Data Scientist
HDFSHBase
MapR Distribution for Hadoop
Apache Drill does not use MapReduce
MongoDB/
Cassandra
sensors
XML,%
JSON%
Data
entering
HBase
47. 47
Apache Drill Distributed Query Processing
– A Storage Independent Drillbit MPP Architecture
Each drillbit is capable of receiving queries from applications and BI tools
- there is no master in this architecture
Multiple drillbits are involved in parallel query processing on distributed data
Supports Apache HDFS, Apache HBase, MapR-FS, MapR-DB, Amazon S3
48. 48
SQL on Hadoop Example – Apache Drill Supports
Query of Self-Describing Data Without a Schema
JSON
Source: MapR
50. 50
Hadoop Storage Is Independent of Any SQL Engine Accessing
HDFS - Multiple SQL Engines Can Coexist On The Same Data
file file file file file
file file file file file
file file
file file
HDFS
file
file
file
file
YARN
Batch
(MapReduce)
Interactive
(Tez)
On-line
(HBase)
Streaming
(Storm,..)
Graph
(Giraph)
In-memory
(Spark)
HPC MPI
(OpenMPI)
Other
(Search,.)
file
file
file
file
SQLSQLSQL SQL
Storage is independent
of any SQL engine! Key points about Hadoop
• It is possible to have MULTIPLE SQL engines on the same data
• Different SQL engines run on different Hadoop frameworks (M/R, Tez,
Spark) or on no framework at all i.e. directly access HDFS or HBase data
51. 51
Relational DBMS / Hadoop Integration – Several Vendors Have
Integrated RDBMS with Hadoop to Run Analytics
Relational DBMS
External
Polymorphic
table function(s)
HDFS / Hbase/ Hive
SQL, XQuery
RDBMS optimizer handles
transparent access to external
analytical platforms on behalf
of the user
RDBMS and Hadoop could
be deployed on the same
hardware cluster
(preferred) or on different
hardware clusters
Allows join across data in a
single RDBMS and Hadoop
52. 52
Relational DBMS / Hadoop Integration Example
- HP Vertica and MapR
Source: MapR
53. 53
Self-Service BI
Self-service Data
Discovery & Visualisation
or Dashboard Server
Business
analyst
Data Virtualization and Optimization
personal
& office
data Predictive
models
Transaction
systems
Data Management Tools (ETL, DQ, etc.)
DW
Self-Service Access To Big Data Via Data Virtualization
BUT what about optimization?
Can the data virtualisation server push
down analytics to underlying platforms
to make them do the work?
54. 54
New Insights Can Be Added Into A Data Warehouse To Enrich
What You Already Know
DW
D
I
new
insights
Operational
systems
e.g. Deriving insight from social web sites like for sentiment analytics
sandbox
Data Scientists
social
Web
logs
web cloud
ELT
55. 55
Alternatively New Insights In Hadoop Can Integrated With A
DW Using Data Virtualization To Provide Enriched Information
DW
D
I
e.g. Deriving insight from social web sites like for sentiment analytics
new
insights
OLTP systems
sandbox
Data Scientists
social
Web
logs
web cloud DataVitualisation
SQL on
Hadoop
56. 56
Using Hadoop As A Data Archive Means Data Can Be Kept
On-line, Analysed And Still Integrated With Data In The DW
DW
D
I
OLTP systems
DataVitualisation
SQL on
Hadoop
Archived data
Archiveunused
ordata>nyears
57. 57
SQL on
Hadoop
Big Data Governance – Data Sources, Sandboxes,
People, Data Access Security, Results Lineage….
Graph DBMS
MPP Analytical
RDBMS
Social
graph data Unstructured / semi-
structured content
DW
RDBMSFiles
clickstream%
Web logs
governance
governance
governance
governance
governance
governance
governancegovernancegovernance
58. 58
Issues: Siloed Analytics - Different Tools to Manage and
Integrate Data For Each Type of Analytical Data Store
Analytical
tools
Data
management
tools
EDW
mart
Structured data
CRM ERP SCM
Silo
DW & marts
Streaming data
(markets, sensors
Analytical
models
Silo
Analytical
tools/apps
Data
management
tools
Multi-structured
data
Silo
DW
Appliance
Advanced Analytics
(structured data)
Data
management
tools
Structured data
CRM ERP SCM
Analytical
tools
Silo
Analytical
tools/apps
Data
management
tools
NoSQL DB
e.g. graph DB
Silo
Multi-structured &
structured data
59. 59
EDW
MDM SystemDW & marts
NoSQL DB
e.g. graph DB
Advanced Analytic
(multi-structured data)
mart
DW
Appliance
Advanced Analytics
(structured data)
Need to Manage The Supply of Consistent Data Across
The Entire Analytical Ecosystem
Common Enterprise Information Management Tool Suite
Stream
processing
C
R
U
D
Prod
Asset
Cust
actions
feedssensors
XML,%
JSON%
RDBMS Files office docssocial Cloud
clickstream%
Web logs
web services
New
New
New
New
New New New New NewNew
New
New
C
R
U
D
Prod
Asset
Cust
New data types need to be supported by EIM tool suites
60. 60
BI tools platform &
data visualisation
tools
Search
based
BI tools
Custom
MapReduce
applications
Map
Reduce
BI tools
Graph
Analytics
tools
A New Architecture for Analytics - The Intelligent
Business Strategies Extended Analytical Ecosystem
Enterprise Information Management Tool Suite
feedssensors
XML,%
JSON%
RDBMS Files office docssocial Cloud
clickstream%
Web logs
web services
Event
processing
C
R
U
D
Prod
Asset
Cust
EDW
MDM SystemDW & marts
NoSQL DB
e.g. graph DB
Advanced Analytics
(multi-structured data)
mart
DW
Appliance
Advanced Analytics
(structured data)
actions
Filtered
data
Data Virtualisation and optimization
61. 61
Conclusions
! Business demand for new more complex, high volume data is driving
the need for new analytical workloads beyond the data warehouse
! Hadoop is a low cost analytical platform capable of supporting new
analytical workloads on multi-stuctured data
! A key role for Hadoop is as an data hub and data refinery
! The data refinery process requires data integration and cleansing to
scale to handle the volume, variety and velocity of complex multi-
structured data
! Data scientists analyse big data as part of the data refining process to
produce new insights that can be added to what you already know
! Hadoop is part of an extended analytical ecosystem with data
management tools supplying consistent data across all data stores
! Data scientists, business analysts and information consumers need to
work together to deliver new insight for competitive advantage