INTELLIGENT

BUSINESS
STRATEGIES
WHITE PAPER

Architecting A Platform For Big Data

Analytics
2nd Edition

By Mike Ferguson
Intelligent Business Strategies
March 2016

Prepared for:
Architecting a Platform For Big Data Analytics – 2nd Edition

Table of Contents
Introduction ....................................................................................................................... 3

Business Data Requirements For Understanding Customer DNA .................................... 5

On-line Click Stream Data .................................................................................... 5
Social Network Data ............................................................................................. 5
Open Government Data ....................................................................................... 6
Sensor Data ......................................................................................................... 6

Technical Requirements to Understand The Customer .................................................... 7

Multiple Analytical Workload Requirements ......................................................... 7
Multiple Analytical Data Platforms ........................................................................ 7
Scalable Data Capture, Preparation And Integration ........................................... 8
Scalable Analytics Requirements ......................................................................... 9
Data Governance Requirements ........................................................................ 10
Easy Access to Data .......................................................................................... 11

Integrating Big Data Into Your DW/BI Environment – The Logical Data Warehouse ...... 12

Agile Big Data Development ........................................................................................... 13

Start In The Cloud .............................................................................................. 13
Logical Data Warehouse .................................................................................... 13
Organise For Success Using Publish And Subscribe ........................................ 14

Vendor Example: IBM’s End-to-End Platform For Big Data Analytics ............................ 15
IBM BigInsights and Open Data Platform with Apache Hadoop ........................ 15
Apache Spark ..................................................................................................... 16
IBM Technology Integration With Apache Spark ................................... 16
Spark-as-a-Service on IBM Bluemix ...................................................... 17
IBM PureData System for Analytics ................................................................... 17
IBM dashDB ....................................................................................................... 18
IBM Data Integration For The Big Data Enterprise ............................................. 18
IBM BigInsights BigIntegrate and BigInsights BigQuality ...................... 18
IBM Streams – Real-time Optimisation Using Big Data ..................................... 19
Accessing The Logical Data Warehouse Using IBM Big SQL And IBM Fluid
Query.................................................................................................................. 20
IBM Big SQL .......................................................................................... 20
IBM Fluid Query ..................................................................................... 21
Analytical Tools ..................................................................................... 21

Conclusions..................................................................................................................... 22

Copyright © Intelligent Business Strategies Limited, 2016, All Rights Reserved 2

Architecting a Platform For Big Data Analytics – 2nd Edition

INTRODUCTION
In this digital age, ‘disruption’ is a term we hear a lot about. It can be defined as
Data and analytics
provide the insights for
a the key to business
“A disturbance that interrupts an event, activity, or process”
disruption
In the context of this paper it means competition appearing from unexpected
quarters that rapidly takes business away from traditional suppliers. It is made
possible because of data, that when collected, cleaned, integrated and
analysed, provides sufficient insights to identify new market opportunities and
prospective customers. Those that have the data can see the opportunities and
cause disruption. Those that don’t have data, or only a subset of it, cannot.

Disruption is The speed at which disruption can occur is accelerating. People with mobile
accelerating fuelled by devices can easily search for products and services, find available suppliers,
people who are compare products and services, read reviews, rate products and collaborate
increasingly informed
with others to tell them about what they like and dislike all while on the move.
before they buy
This kind of power puts prospective customers in a very powerful position when
making buying decisions because they are increasingly informed before they
buy and, armed with this information, they can switch suppliers at the click of a
mouse if a more personalised, better quality product or service is available.
Companies are having The result is that, with this information at hand, loyalty is becoming cheap.
to fight harder to retain People spread the word on social networks and disruption takes hold in the
existing customers market. In this kind of market, no one is safe. Companies have to fight harder
while also trying to to retain existing customers while also trying to grow their customer base.
grow
Given this backdrop, it is not surprising that many companies are therefore
Companies are trying focusing on improving quality and customer service to try to retain their
to offer improved customers. Process bottlenecks are being removed and process errors that
quality of service and impact on the customers’ experience are being fixed to make sure that
personalisation to everything runs smoothly. In addition, many companies are trying to improve
retain customers customer engagement and ensure the same customer experience across all
physical and on-line channels. That requires all customer facing employees
and systems to have access to deeper customer insight and to know about all
customer interactions. The objective (often referred to as an Omni-channel
initiative) is to create a smart front office with personalised customer insight
Analysing enriched and personalised customer marketing recommendations available across all
customer data using channels. This is shown in Figure 1 where personalisation is made possible by
predictice and analysing enriched customer data using predictive and prescriptive analytics.
prescriptive analytics is
the key to Until recently, the way in which we produced customer insight and customer
personalisation recommendations was to simply analyse transaction activity in a data
warehouse. However the limitation is that only transaction activity is analysed.
It does not include analysis of other high value data that when pieced together
Insights produced by offers a much more complete understanding of a customer’s “DNA”. Therein,
analysing of
lies the problem. Analysis of traditional transaction data in a data warehouse is
transactional activity is
insufficient for a company to gain disruptive insight. More data is needed.
no longer enough
tocause disruption Therefore new data requirements and new technical requirements need to be
defined to identify, process and analyse all the data needed to enable
companies to become ‘disrupters’. Let’s take a look at what those
requirements are.

Copyright © Intelligent Business Strategies Limited, 2016, All Rights Reserved 3

Architecting a Platform For Big Data Analytics – 2nd Edition

Consistent Customer Treatment Across All Channels –

The Smart Omni-Channel Front-Office

Customers
Insights produced by Improve customer
engagement
analysing transactional Digital channels are
and non-transactional generating big data
data is now needed Front-Office Operations
E-commerce M-commerce Social commerce Sales Force Customer facing Customer
across all traditional application Mobile apps applications automation apps bricks & mortar apps service apps
and digital channels e.g. In-store apps
In-branch apps

OMNI-CHANNEL

Predictive Common Personalised Personalised Prescriptive

analytics transaction customer customer marketing analytics
services Insight recommendations

Improve marketing, sales and service via on-demand access to

insights and recommendation services relevant to each customer
Copyright © Intelligent Business Strategies 1992-2015!

Figure 1

Copyright © Intelligent Business Strategies Limited, 2016, All Rights Reserved 4

Architecting a Platform For Big Data Analytics – 2nd Edition

BUSINESS DATA REQUIREMENTS FOR

UNDERSTANDING CUSTOMER DNA
Given the limitations of traditional transaction data, businesses are now
Data from digital
defining new data requirements to attempt to strengthen and enrich what they
channels also needs to
be analysed to gain a know about customers in order to see new opportunities and disrupt
complete marketplaces. In addition to the aforementioned transaction data, business
understanding of departments now want to analyse data generated in the digital channels
customer behaviour (highlighted in red in Figure 1), and data from external data sources to gain
much more complete understanding of customers. This includes:
o On-line click stream data
o Social network data and in-bound email
o Unstructured data e.g. loan documents, application forms
o Open government data
o Sensor data emitted from smart products being used by customers

ON-LINE CLICK STREAM DATA

Click stream data
On-line click stream data is the data generated from visitors clicking on your
reveals how customers website. They may land on your home page from a search engine and then
navigate your website start browsing your site moving from web page to web page looking at products
and services. Your web servers record every click a visitor makes in a visitor
‘session’ in one or more weblog files. By accessing these files and analysing
clickstream data, it becomes possible to understand:
o Where visitors come from (via geocoding of IP addresses)
o How they reached your site
o Navigation paths taken through your site before buying products
o Navigation paths taken through your site before abandoning it
o What they looked at en route to buying products on your site
The richer the data, the This data can be used to improve customer experience and conversion
more effective the especially if clicks can be associated with customers and prospects. Doing this
customer insight will be requires additional data to be integrated with clickstream data. The richer the
data, the more detailed questions can be answered.
For example analysing transaction data may provide insights about loyal, low
risk customers that are not generating much revenue. Clickstream analysis
may identify high revenue products on your website buried under too many
clicks. Combining the two may identify new ways to digitally market these
products to loyal customers that would result in increased revenue generation.

SOCIAL NETWORK DATA

Social networks can Popular social networks like Twitter, Facebook, LinkedIn and YouTube can
provide new data and also provide valuable new customer data and insights. This includes:
customer insights o Additional customers attributes e.g. about employment, hobbies, interests
o Previously unknown relationships e.g. household members
o Likes and dislikes
o Influencers
o Product and brand sentiment

Copyright © Intelligent Business Strategies Limited, 2016, All Rights Reserved 5

Architecting a Platform For Big Data Analytics – 2nd Edition

Social networks Sentiment can also come from in-bound email and CRM system notes.
provide new data, Identifying influencers in social networks is important because it allows
sentiment and an marketing departments to run ‘target the influencer’ marketing campaigns to
understanding of who see if they can recruit new customers, cause revenue uplifts and therefore
influencers are
improve marketing effectiveness.

OPEN GOVERNMENT DATA

Open government data (e.g. Data.gov, Data.gov.uk, Data.gove.be etc.) is any
data produced or commissioned by public bodies and published for download.
It is free to use and includes information about:
o Businesses e.g. patents, trademarks and public tender databases
o Geographic information (including address information, aerial photos,
Open government data
geodetic networks, geology)
is hugely valuable to o Planning information
understanding o Legal decisions (of national, foreign and international courts)
customers and risk
o Weather information (e.g. Climate data and models and weather forecasts)
o Social data (including various types of statistics on economics,
employment, health, population, public administration)
o Crime data (including various types of statistics)
o Transport information (including traffic congestion, work on roads, public
transport and vehicle registration)
This kind of data can be hugely valuable in understanding customers and risk.

SENSOR DATA
Sensors can also be used to gain a deeper understanding of customers in that
they can be embedded in products those customers own. As a result data can
Sensor data in smart be collected to understand how products are used and to capture data about
products is another people’s health. The use of sensor data is particularly relevant to customer
good data source that location, which can be revealed using smart phone GPS sensors. This
can reveal new combined with clickstream data allows telecommunications companies, for
business opportunities
example, to monitor what people are browsing on-line and also where they are
while browsing. This allows Telecommunications companies to disrupt the
advertising industry by offering location-based mobile advertising - a totally
new line of business. Sensors can also be used to monitor customer
movement and product usage.

Sensor data can help Beyond customer, sensor data is frequently used to help optimise operations,
optimise business reduce risk and provide insights that may lead to new products. Sensors allow
operations and avoid companies to monitor live operations, prevent problems happening to keep
unplanned operational processes optimised and avoid unplanned costs. Typical use cases include:
cost o Supply/distribution chain optimisation
o Asset management and field service optimisation
o Manufacturing production line optimisation
o Location based advertising (mobile phones)
o Grid health monitoring e,g. Electricity, water, mobile phone cell network
o Oil and gas drilling activity monitoring, well integrity and asset management
o Usage / consumption monitoring via smart metering
o Healthcare
o Traffic optimisation

Copyright © Intelligent Business Strategies Limited, 2016, All Rights Reserved 6

Architecting a Platform For Big Data Analytics – 2nd Edition

TECHNICAL REQUIREMENTS TO UNDERSTAND THE

CUSTOMER
Looking at these data requirements, a lot of new internal and external data is
needed, over and above traditional transaction data, to create a more complete
view of a customer and to optimise business operations. However we need to
recognise that the characteristics of the data needed have gone beyond the
Structured, semi- structured transaction data in traditional data warehouses.
structured and
unstructured data are
The new data required includes structured, semi-structured (e.g. JSON, XML)
all needed to create a
and unstructured (e.g. text) data, all of which needs to be captured, processed
more complete view of
the customer and and analysed to produce new insights. Some of this data (e.g. clickstream and
optimise operations social media data) can be very large in volume and some of it is created in
real-time at very high rates (e.g. sensor data). It is not surprising therefore that
new technical requirements naturally fall out of the need capture, process and
analyse this data. These are shown below and categorised for easy reading.

MULTIPLE ANALYTICAL WORKLOAD REQUIREMENTS

Having already established that additional insight is needed beyond what is
produced from analysis of transaction data, it should be possible to:

New kinds of analytical • Support new kinds of analytical workloads to enable disruption including:
workloads are needed
to enable disruption o Real-time analysis of data in motion (e.g. analyse clickstream while
the visitor is on your website, or sensor data to predict asset failure)
o Exploratory analysis of un-modeled, multi-structured data e.g.
Social network text, open government data, sensor data
o Graph analysis e.g. community analysis, social network influencer
Real-time analysis of analysis
streaming data, graph
o Machine learning to:
analysis, machine
learning and § Develop predictive models on large volumes of structured
exploratory analysis of data e.g. clickstream data
multi-structured data
are all needed § Score and predict propensity to buy, trends in behavior and
customer churn using a more comprehensive set of data
§ Score and predict asset or network failure that would impact
negatively on revenue, customer experience, risk or
operational cost

MULTIPLE ANALYTICAL DATA PLATFORMS

In order to capture, prepare and integrate data at scale, and run multiple
analytical workloads to enable disruption support is now needed for multiple
analytical platforms including:
Multiple platforms are o NoSQL databases e.g. a Graph DBMS
now needed to support
different analytical o Hadoop
workloads o Analytical RDBMSs

Copyright © Intelligent Business Strategies Limited, 2016, All Rights Reserved 7

Architecting a Platform For Big Data Analytics – 2nd Edition

These can be on the

o Streaming analytics platforms
cloud, on-premises or These could be on the cloud, on-premises or both. Also note that Apache
both Spark massively parallel in-memory processing could be used in analysis
of data at rest and so can retrieve data from any of the above listed data
stores. It can also be used for streaming analytics on data in-motion.

Different Platforms Optimised For Different

Analytical Workloads Are Now Needed
Big Data workloads result in multiple platforms now being needed for
analytical processing

Real-time Investigative Master data

stream Graph
processing &
analysis, management
analysis Data refinery
decision m’gmt

Streaming Advanced Analytic DW & marts Advanced Analytics MDM

analytics (multi-structured data) (structured data) R
C Prod Cust U
NoSQL DB EDW DW
Asset
e.g. graph DB mart Appliance
D
Streaming NoSQL Hadoop Data Warehouse Analytical
data DBMS data store RDBMS RDBMS

Traditional Data mining,

query, model
reporting & development
analysis

Copyright © Intelligent Business Strategies 1992-2015!

Figure 2

SCALABLE DATA CAPTURE, PREPARATION AND INTEGRATION

Data ingest, data To process data with these kinds of characteristics requires scalability. That
cleansing, data scalability needs to apply to data ingest (capture), cleansing, transformation,
transformation, data integration and, of course, analysis. There is no point having analytics that
integration and analysis
scale if the software needed to capture, prepare and integrate data before the
all need to scale
analysis begins, does not also scale or vice versa. Therefore the following
technical requirements are related to data connectivity, scalable data capture,
data processing and analytics.

It should be possible to:

New internal and • Connect to multiple different data sources where new big data originates.
external data sources This includes connectors to NoSQL DBMSs (e.g. MongoDB, Cassandra,
need to be supported HBase), social networks (e.g. Facebook, Twitter, LinkedIn), machine data
log files, sensor networks (with different types of sensors), web sites, email
servers, streaming data sources, cloud data sources, and Hadoop
Distributed File System (HDFS). This is in addition to ‘traditional’ data
sources such as RDBMSs, files, XML, JSON, packaged applications, web
services etc., that are typically used in data warehousing
Data preparation and
integration jobs should • Capture / ingest big data in parallel into a scalable analytical platform either
be defined once and be on-premises or on the cloud
able to run on all
necessary platforms • Develop data preparation and integration jobs independently of the platform
they execute on and then to be able to run them on the platform required
with the necessary parallelism to exploit the full power of the platform
selected. This means that business and IT professional productivity should
not be slowed down by having to learn new user interfaces and new
development languages needed to exploit underlying big data technology
platforms. The software should hide that

Copyright © Intelligent Business Strategies Limited, 2016, All Rights Reserved 8

Architecting a Platform For Big Data Analytics – 2nd Edition

New data should not • Irrespective of whether development of data capture, data preparation, data
impose additional work integration and analysis jobs are done by IT professionals or business
on IT and business analysts, it should be possible to develop once, and run on the platform
analysts to process it best suited to do the work irrespective if this is on-premises or on the cloud
• Support the ability to flatten semi-structured data e.g. JSON, XML
• Support shared nothing parallel execution of each and every data
cleansing, data transformation and data integration task defined in a data
Scalable data
preparation and integration job. This is needed to be able to exploit the full
cleansing,
transformation and power of underlying hardware
integration is needed to • Support multiple types of parallelism across nodes including parallel
handle large data execution of each transformation (as described above) across partitioned
volumes
data and also pipeline parallelism whereby the output of one task
(executing in parallel) can flow to the next task while the first task is still
executing
Data cleansing, • Push data cleansing, transformation and integration tasks to execute where
transformation and the data is located and not have to take the data to a central point where
integration needs to
cleansing, transformation and integration tasks are located
happen as close to the
data as possible • Process more data simply by adding more hardware
• Undertake probabilistic matching (fuzzy matching) of sentiment with
customer master data at scale to understand customer sentiment

SCALABLE ANALYTICS REQUIREMENTS

It should be possible to:
Analytical models • Develop predictive analytical models offline and have flexibility in terms of
should be able to run deploy options to run these models in:
in-stream, in-memory
in-batch or in-database o Real-time in a scalable streaming analytical environment to analyse
data in motion
o Batch using scalable batch analytical processing e.g. on Hadoop
o Memory using scalable analytical processing e.g. on Spark
o An analytical RDBMS e.g. to analyse data in a data warehouse
This should be possible both on-premises and on the cloud
Invocation of scalable • Invoke any pre-built analytic algorithms available in-Hadoop, in-Database
analytics is needed in-Memory and in-Stream during analytical model execution. For example,
exploiting pre-built machine-learning algorithms, text analytics, graph
analytics or custom- analytics already available on Hadoop or in-Spark from
within a batch data integration job, a batch analytic application or a front-
end self-service interactive analytical tool. In the case of text analytics, this
would allow high value structured data to be extracted in parallel from large
document collections, and then integrated with other data during execution
• Support model management including support for versioning,
Champion 
Challenger and A/B Testing
 of models
• Automate model evaluation, model refresh and model re-deployment
Analytical models should • Develop predictive analytical models on-premises or on the cloud
be able to run in the
cloud or on-premises • Rapidly explore and analyse new structured, semi-structured and
unstructured big data using search

Copyright © Intelligent Business Strategies Limited, 2016, All Rights Reserved 9

Architecting a Platform For Big Data Analytics – 2nd Edition

• Create analytical pipelines by stringing together multiple analytic algorithms

in a pipeline (workflow) to analyse data at scale
Need to develop
• Develop batch analytic applications in popular languages such as Java,
analytical applications in
popular programming Python, Scala and R that can analyse data at scale using pre-built analytic
languages algorithms that run close to the data in parallel
• Easily develop custom scalable analytical algorithms to add to those
available ‘out-of-the-box’ in Hadoop, Spark and analytical RDBMSs to
analyse complex data
Use of tools to generate
scalable analytical • Make use of tools that generate code to analyse data at scale using pre-
applications should also built analytic and custom algorithms that run in parallel
be possible to reduce • Analyse streaming data in real-time by invoking in-memory analytics in a
time to value
scalable cluster computing environment
• Define rules to drive automated decisions and actions when patterns are
Automated alerting and
detected and conditions predicted during streaming analytics
recommendations are
needed to guide and • Automate alerting, recommendations, transaction invocation and process
optimise operations invocation as actions during real-time stream analytics processing

DATA GOVERNANCE REQUIREMENTS

It should be possible to:
An information catalog is • Register new datasets / data streams coming into the enterprise in an
needed to track and information catalog either via API or user interface so their existence is
govern data
known and the data can be governed
• Automate the profiling, schema detection and semantic analysis of data so
that the quality, structure, and meaning of data can be quickly assessed
Automated profiling and and determined
classification of data will
speed up data • Classify newly registered data in terms of its sensitivity/confidentiality, data
processing and help quality, trustworthiness, retention and business value
determine how data is
governed
• Define governance policies to govern data based on how it is classified
• Collaborate with others on defining these policies
It should be possible to • Define rules to enforce governance policies
define rules to govern
data irrespective of its o Irrespective of data store and location on-premises or in the cloud
location
o To support national, regional or other jurisdictional boundary
requirements for data
It should be possible to
enforce national, • Verify the quality of data and report problems
regional and other
jurisdictional regulations
• Record metadata associated with data cleansing (standardisation,
when governing data cleansing, enrichment), transformation and integration tasks irrespective of
whether the data is at rest or in-motion, on-premises or in the cloud and the
Auditing of processing tasks are performed by tools or custom applications or both
tasks is needed to
• View metadata lineage from end-to-end to ensure trustworthiness
ensure trustworthiness
Data privacy needs to • Mask and protect data classified as confidential or sensitive and enforce
be enforced irrespective any policies irrespective of whether the data is at rest in Hadoop HDFS, an
of where data is stored RDBMS, a NoSQL database, or if it is data-in-motion

Copyright © Intelligent Business Strategies Limited, 2016, All Rights Reserved 10

Architecting a Platform For Big Data Analytics – 2nd Edition

• Archive and manage data movement between data stores (e.g. data
warehouse and Hadoop) ensuring all governance rules associated with
sensitivity, quality and retention are upheld irrespective of data location
• Control access to data by applications, tools and users to protect data
Governance aware • Have governance aware execution engines to enforce data governance
execution engines are
rules anywhere in the analytical ecosystem shown in Figure 2 to enforce
needed to enforce
defined policies
governance rules
anywhere that data to be • Map new insights produced by analysing new data, into a shared business
governed is located vocabulary in a business glossary so that the meaning of this data can be
understood before it is shared

EASY ACCESS TO DATA

It should be possible for IT professionals and data scientists to:

Data should be available • Publish new datasets, data integration workflows, analytical workflows, and
as a service within the insights, as part of a data curation process, to an information catalog for
enterprise and other users and applications to consume and use
documented in an
It should be possible for data scientists and business analysts to:
information catalog
• Access an information catalog to see what data exists, where it is located,
Data lineage is needed what state it is in, where it came from, how it has been transformed,
to help understand whether they can trust it and if it is available for use
where data came from
• Easily search the information catalog to find new datasets and insights so
that they can quickly see what exists
Business users should
be able to search the • Subscribe to receive new data and insights, published in an information
information catalog to catalog, for delivery to wherever they require it, in whatever format they
find and subscribe to require it subject to any governance rules being applied to enforce national,
data
regional, or other jurisdictional policies that restrict its use
• Access authorised data and insights that may exist in multiple analytical
data stores and data streaming platforms through a common SQL interface
It should be possible to from self-service tools to simplify access to data
federate queries across
streaming data, Hadoop
• Federate queries data across Hadoop, traditional data warehouses, and
and traditional data live data streams to produce disruptive actionable insights
warehouses to produce • Access Hadoop data using SQL from traditional data warehouse data
disruptive insights
stores as well as via SQL on Hadoop initiatives
It should be possible to
hide complexity of data • It should be possible to query and analyse data in a logical data warehouse
access by creating a (across multiple analytical data stores and real-time streaming platforms)
logical data warehouse using traditional and Cognitive Analytical tools irrespective of whether the
comprised of multiple data in the logical data warehouse is on-premises, on the cloud or both
underlying platforms

Copyright © Intelligent Business Strategies Limited, 2016, All Rights Reserved 11

Architecting a Platform For Big Data Analytics – 2nd Edition

INTEGRATING BIG DATA INTO YOUR DW/BI

ENVIRONMENT – THE LOGICAL DATA WAREHOUSE
In order to bring together new insights with what already exists (e.g. for
disruptive customer insight), a new integrated ‘logical data warehouse’ platform
is needed that includes multiple analytical data stores (including Hadoop, the
data warehouse, MDM, data warehouse appliances), end-to-end-information
management, high volume ingest, data governance, batch and real-time
streaming analytics, and simplified access to all data via a SQL-based data
virtualisation layer. This logical data warehouse architecture may exist on the
cloud, on premises or include data stores spread across both (hybrid). This is
shown in Figure 3. Note that access to multi-structured data is simplified
because of a ‘SQL on Everything’ data virtualisation layer that can federate
query processing across multiple data stores.
The Logical Data Warehouse Is A Multi-Platform Analytical
Ecosystem With Simplified Access To Multi-Structured Data
Graph Custom Spark Search Traditional, Self-
A federated query layer Analytics Analytic and MR based Service and Cognitive
tools applications BI tools BI tools Analytics tools
is needed to hide
Data Virtualisation / SQL on Everything
multiple data stores from Advanced Analytics DW & marts Advanced Analytics MDM System
users in a modern (multi-structured data) (structured data) R
Prod Cust U
analytical ecosystem NoSQL DB EDW DW C
e.g. graph DB mart Appliance Asset

D
actions
Filtered
Stream
Data in Hadoop can be processing data Enterprise Information Management Tool Suite
integrated with data in
Streaming
traditional data data
XML,
warehouses to produce clickstream
JSON
sensors Web logs web services RDBMS feeds social Cloud Files office docs
disruptive insights
Copyright © Intelligent Business Strategies 1992-2016!

Figure 3
Apache Spark is rapidly Furthermore, Apache Spark can sit on top of all data stores as a general
emerging as an in- purpose massively parallel in-memory analytics layer primarily aimed at data
memory analytics layer science. It could also be embedded in database products for in-database, in-
that can run on top of
memory analytics.
muliple data stores or
even inside DBMSs Figure 3 also shows an enterprise information management (EIM) tool suite to
manage data flows into and between data stores in a hybrid multi-platform
analytical ecosystem. The EIM suite includes tooling to support:
An enterpise information
management tool suite
• An information governance catalog and business glossary
enables data curation • Data modeling
and governance across
a multi-platform • Data and metadata relationship discovery
analytical ecosystem
• Data quality profiling and monitoring

The EIM tool suite works

• Data cleansing and matching at scale
across cloud on- • Data transformation and integration at scale
premises and hybrid
logical data warehouse • Data privacy and lifecycle management
environments
• Data audit and protection

Copyright © Intelligent Business Strategies Limited, 2016, All Rights Reserved 12

Architecting a Platform For Big Data Analytics – 2nd Edition

AGILE BIG DATA DEVELOPMENT

Having understood new data requirements plus the technical requirements that
need to be supported to process and analyse that data, the next question is:

“How can you enable development of big data analytics to produce disruptive
insight quickly while also dealing with a deluge of new and complex data?”

There are a few key ways to make this possible:

There are several ways • Use the cloud as a low cost way to get started in create new insights to
to introduce agility into add to what you already know
big data development
processes • Make use of data virtualisation and federated query processing to
create a ‘logical data warehouse’ across multiple analytical data stores

• Organise for success by creating a ‘publish and subscribe’ approach to

development using an information catalog to facilitate re-use

START IN THE CLOUD

Creating big data Once candidate big data projects have been defined and prioritised, and new
analytics in the cloud is data sources identified, cloud computing is a fast, low cost option to initiate
a fast and low cost way disruptive big data analytical projects. Hadoop and Spark are often available as
to get started a service in the cloud. Data like social network and open government data can
be loaded into cloud storage so exploratory analysis on that data can start
quickly. Data preparation and integration jobs can be developed to process
data at scale on the cloud before being analysed by scalable in-memory Spark
analytic applications that use Spark streaming, machine learning (MLlib) and/or
graph analysis (GraphX). This can optionally be switched to on-premises later.

LOGICAL DATA WAREHOUSE

Data virtualisation or Figure 3 shows how data virtualisation can be used to create the logical data
federated SQL query warehouse. This can either be done by a data virtualisation server or by a ‘SQL
engines can simplify on Everything’ engine that can connect to multiple sources, optimise queries
data access and
and federate them across non-relational data (e.g. Hadoop, Spark, NoSQL
introduce flexibility via
DBMSs) and relational DBMS data sources. By hiding complexity from users, it
support for multiple
virtual views of becomes possible to create the concept of a logical data warehouse with
underlying data different virtual personalised views across multiple heterogeneous data stores.
This of course requires a SQL optimiser to be able to pushdown analytics close
to the data and also to work out the best way to join Hadoop to non-Hadoop
data.

Copyright © Intelligent Business Strategies Limited, 2016, All Rights Reserved 13

Architecting a Platform For Big Data Analytics – 2nd Edition

ORGANISE FOR SUCCESS USING PUBLISH AND SUBSCRIBE

Creating a production line In addition you can organise for success, creating a ‘production line’ approach
approach to building so that people in different parts of the production line build on what others
insight clearly defines before them have created. This can be done via a publish-and-subscribe
roles and encourages development approach and an information catalog to keep track of what is
reuse of data and
available. This is shown in Figure 5 with every task building on the previous by
analytics to speed up the
production of insights taking services produced by one task as input to the next. The first task is
creating trusted data services. These services are then published for others to
pick up to develop DI/DQ flows that join data from multiple trusted data flows.
Newly developed integrated data workflows are then themselves published as
services. Other people can then pick these up as a ‘quick start’ to creating
analytical workflows. Analytical workflows take integrated trusted data and
analyse it to produce scores or other insights. These analytical workflows are
then published and made available for others to either embed in applications,
visualise the results of the analysis, use as input to a decision engine to create
decision services (e.g. next best offer recommendation) or turn into analytic
applications.

Accelera'ng*Delivery*Using*A*Publish*And*Subscribe*Approach*
trusted*data*
as*a*service*
Acquire publish*
Acquire Data*Cleansing*,* Info*
data Acquire Transform,*Enrich,*Filter catalog*
A publish and subscribe source*
production line approach New*predic've*
analy'c*pipelines*
together with an (as*a*service)*
information catalog
encourages reuse and subscribe* publish* Info* subscribe* publish* Analy'cs*
consume* Data* catalog* consume* Analyse* catalog*
can significantly reduce Integra'on* (e.g.score)*
the time to produce trusted,*integrated*
data*ad*a*service*
disruptive insights
other*e.g.*embed**
publish* New*analy'c*
analy'c*applica'ons*
subscribe* Solu'ons* applica'ons*
Visualise catalog* use*
consume*
publish* New*prescrip've*
Decide Act analy'c*pipelines*
Copyright © Intelligent Business Strategies 1992-2015!

Figure 4

Copyright © Intelligent Business Strategies Limited, 2016, All Rights Reserved 14

Architecting a Platform For Big Data Analytics – 2nd Edition

VENDOR EXAMPLE: IBM’S END-TO-END

PLATFORM FOR BIG DATA ANALYTICS
IBM has been a provider Having defined data and technical requirements together with a new logical
of big data and analytics data warehouse architecture and agile development approaches, this section
technologies for a looks at how one vendor steps up to meet the requirements defined earlier in
number of years order to enable disruption. That vendor is IBM.

IBM has a number of technology components that together can be used to

build a logical data warehouse architecture combining traditional and big data
technologies. These include:
Their portfolio of
• A Hadoop distribution
analytical technology
components includes • Apache Spark in-memory analytics execution engine
Hadoop, an analytucal
RDBMS, streaming • An analytical relational DBMS for enterprise data warehouse
analytics, a federated
SQL engine, an
• A streaming analytics platform
information management • A cloud-based analytical DBMS
tool suite and a range of
analytical tools • SQL access to Hadoop and traditional data warehouse data
• An EIM tool suite that includes scalable data cleansing and integration

IBM BIGINSIGHTS AND OPEN DATA PLATFORM WITH APACHE HADOOP

IBM’s distribution of Apache Hadoop consists of two major components. These
are:

Two major components • IBM Open Platform with Apache Hadoop

comprise the core of
IBM’s Hadoop offering • IBM BigInsights for Apache Hadoop

IBM Open Platform with Apache Hadoop is a 100% open source common
Open Data Platform (ODP) core of Apache Hadoop (inclusive of HDFS, YARN,
and MapReduce) and Apache Ambari software. The following components ship
as part of this:

Component Description
Ambari Hadoop cluster management
Apache Kafka Scalable message handling for inbound streaming
data
IBM Open Platform
with Hadoop includes Flume Web log data ingestion into Hadoop HDFS
core Hadoop HBase Column family NoSQL database for high velocity data
technologies such as ingest and operational reporting
HDFS, Hive, Pig, Spark HDFS Hadoop Distributed File System that partitions &
and HBase distributes data across a cluster
Hive SQL access to HDFS and HBase data
Knox An API Gateway for protection of Rest APIs
Lucene Java-based indexing and search technology
Oozie Scheduling
Parquet Columnar

Copyright © Intelligent Business Strategies Limited, 2016, All Rights Reserved 15

Architecting a Platform For Big Data Analytics – 2nd Edition

Pig Scripting language for data processing

Slidr A YARN application to deploy distributed applications
on YARN
Solr Search server built using Lucene Core
Spark A massively parallel in-memory execution
environment for Java, Python, Scala and R based
analytic applications doing machine learning, graph
analysis and streaming analytics. It also supports SQL
access to in-memory data
Sqoop Data mover from relational to HDFS and vice-versa. It
also supports other sources and targets
Zookeeper An open-source server which enables highly reliable
distributed coordination

IBM BigInsights includes IBM BigInsights for Apache Hadoop is a collection of value-added services that
a set of packaged
can be installed on top of the IBM® Open Platform with Apache Hadoop or any
modules aimed at
different types of user
incumbent Hadoop distribution (e.g. Cloudera, MapR etc.). It offers analytic
and enterprise capabilities for Hadoop and includes the following modules:

• BigInsights Data Scientist - includes native support for the R

Data scientists have programming language (Big R) and adds machine learning algorithms
access to a scalable that are optimised for Hadoop. It also includes various analyst
version of R and machine capabilities along with Scaling R packages to Hadoop, using declarative
learning algorithms machine language text analytics and distilling unstructured data into
optimised for Hadoop
meaningful data. It provides web-based tooling for annotation. Native
support for open source R statistical computing helps clients leverage
their existing R code or gain from more than 4,500 freely available
statistics packages from the open R community.
Business analysts can
use BigSheets and • BigInsights Analyst - helps discover data in Hadoop for analysis. The
BigSQL to join Hadoop to module includes IBMs SQL-on-Hadoop engine Big SQL and also
non-Hadoop data BigSheets, which is a spreadsheet-like visualisation tool.

Administrators have • BigInsights Enterprise Management - includes tools to help

access to tools to administrators manage, monitor, and secure their Hadoop distribution.
manage monitor and This includes tools to allocate resources, monitor multiple clusters, and
secure a Hadoop cluster optimise workflows to increase performance.

APACHE SPARK
IBM Technology Integration With Apache Spark
IBM has made a A number of IBM software products now integrate with Apache Spark. Some of
strategic commitment these are shown in the table below along with a description of how they
to using Apache Spark integrate.

IBM Product Description of Integration with Spark

IBM SPSS Can invoke Spark MLlib algorithms from IBM SPSS model
Analytic Server workflows
and Modeler
IBM BigSQL Spark analytic applications can access data in HDFS, S3,
HBase and other NoSQL data stores using IBMBigSQL
which will return an RDD for processing. Also IBM BigSQL
can chose to leverage Spark if required in answering SQL
queries

Copyright © Intelligent Business Strategies Limited, 2016, All Rights Reserved 16

Architecting a Platform For Big Data Analytics – 2nd Edition

IBM Streams Add Spark transformation functions, action functions and

Spark MLlib algorithms to existing Streams applications
IBM Cloudant Data in IBM Cloudant can be accessed and analysed in
on Bluemix Spark analytic applications on the IBM Bluemix cloud
A number of IBM IBM BigInsights Data in IBM Open Platform with Apache Hadoop can be
software products are on Bluemix accessed and analysed in BigInsights Data Scientist analytic
now using Apache applications using Spark on the IBM Bluemix cloud
Spark IBM Swift Data in IBM Swift Object Storage can be accessed and
Object Storage analysed in Spark analytic applications
IBM IoT on A fully managed, cloud-hosted service that makes it simple
Bluemix to derive value from Internet of Things (IoT) device
IBM Insights for Allow developers and entrepreneurs to search, quickly
Twitter Service explore and then mine enriched Twitter content

In fact, IBM’s entire analytics platform is built on Spark (see Figure 5)

Figure 5
Spark-as-a-Service on IBM Bluemix
IBM has also made Spark available as a service on IBM Bluemix. Analytics for
Apache Spark works with commonly used tools available in IBM Bluemix so
that you can quickly start tapping into the full power of Apache Spark. The tools
include the following:
• Jupyter Notebooks for interactive and reproducible data analysis and
visualization
• SWIFT Object Storage for storage and management of data files
• Apache Spark for data processing at scale

IBM PUREDATA SYSTEM FOR ANALYTICS

IBM PureData System for Analytics powered by Netezza technology is a
massively parallel data warehouse and analytic hardware appliance running
IBM’s Netezza analytical relational DBMS optimised for advanced analytical
workloads on structured data. In includes a range of in-database analytics so
that visual discovery tools can invoke these on the fly to get good performance
on large volumes of structured data.

Copyright © Intelligent Business Strategies Limited, 2016, All Rights Reserved 17

Architecting a Platform For Big Data Analytics – 2nd Edition

IBM PureData System for The IBM PureData System for Analytics integrates database, server, storage
Analytics is optimised for and advanced analytic capabilities into a single system. It scales from 1 TB to
advanced analytics on 1.5 petabytes includes special processors to filter data as it comes off disk so
structured data and for that only data relevant to a query is processed in the RDBMS. The IBM
some data warehouse
Netezza Analytic RDBMS requires no indexing or tuning which makes it easier
workloads
to manage. It is designed to interface with traditional BI tools including IBM
Cognos Analytics and also runs IBM SPSS developed advanced analytical
models deployed in the database on large volumes of data.

IBM PureData System for Complementing the IBM PureData System for Analytics, is an advanced
Analytics provides free in- analytics framework. In addition to providing a large library of parallelised
database analytics advanced and predictive algorithms, it allows creation of custom analytics
capabilities allowing you created in a number of different programming languages (including C, C++,
to create and apply Java, Perl, Python, Lua, R, and even Fortran) and it allows integration of
complex and leading third party analytic software offerings from companies like SAS, SPSS,
sophisticated analytics
Revolution Analytics, and Fuzzy Logix.
right inside the appliance.
IBM PureData System for Analytics allows you to create, test and apply models
to score data right inside the appliance, eliminating the need to move data and
giving you access to more of the data and more attributes than you might
otherwise be able to use if you needed to extract the data to another computer.

IBM DASHDB
IBM dashDB is a new
IBM dashDB is a fully managed, cloud-based MPP DBMS enabling IBM to offer
massively parallel
cloud-based DBMS data warehouse-as-a-service. It provides in-database analytics, in-memory
offering data columnar computing and connectivity to a wide range of analytical toolsets,
warehouse-as-a- including Watson Analytics and many third-party BI tools.
service
A second deployment option of IBM dashDB, currently in early access preview,
It can also be deployed is also available for fast deployment into private or virtual private clouds via
in a private cloud Docker container.

IBM DATA INTEGRATION FOR THE BIG DATA ENTERPRISE

In terms of scalable data cleansing and integration, IBM provides the following
new and extended tooling capabilities.

IBM BigInsights BigIntegrate and BigInsights BigQuality

IBM has extended its IBM BigInsights BigIntegrate and BigInsights BigQuality are data integration
data integration and and data cleansing components that run on a number of Hadoop distributions.
data cleansing These include IBM BigInsights and IBM Open Platform with Apache Hadoop,
software to support big Hortonworks HDP and Cloudera. BigInsights BigIntegrate and BigInsights
data sources by BigQuality are capable of supporting large data integration workloads. This is
releasing IBM
because the execution architecture is a shared nothing massively parallel
BigIntegrate and
BigQuality
implementation. They include the following capability:

• Connectivity to multiple data sources. These include:

Copyright © Intelligent Business Strategies Limited, 2016, All Rights Reserved 18

Architecting a Platform For Big Data Analytics – 2nd Edition

o The Hadoop Distributed File System (HDFS)

IBM BigIntegrate and
o Hadoop Hive tables
BigQuality run in
parallel across a cluster o NoSQL DBMSs e.g. Apache HBase, Mongo DB and Cassandra
and can process data
in HDFS, Hive, NoSQL o JSON data
DBMSs, relational o Streaming data from IBM Streams to pump filtered event data
DBMSs, files, and
into IBM BigInsights for further analysis
packaged applications
o Java Message Service [JMS]
Streaming data can o Relational DBMSs e.g. IBM PureData System for Analytics, IBM
also be accessed Distributed, DB2 IBM DB2 z/OS data warehouse, IBM DB2
Analytics Accelerator, 3rd party RDBMSs
o Flat files
o IBM InfoSphere Master Data Management
o Popular packaged applications
In-Hadoop analytics
can also be invoked • Change data capture from big data sources
during data intrgration
processing
• Ability to invoke in-Hadoop analytics including Text Analytics via Java
APIs to invoke AQL
• Scalable data processing that includes:
o Data partitioning and distribution of data across Hadoop nodes
o Parallel execution of all data cleansing and data transformation
tasks on any hardware configuration. This includes partition and
pipeline parallelism with repartitioning of data between stages
and 
between nodes without needing to persist intermediate data
IBM BigIntegrate and to disk
BigQuality can run on a o Ability to run unaltered on single processor MPP nodes,
range of multi-processor clustered multi-core, multi-processor SMP nodes and full grid
environments
computing environments
IBM BigIntegrate and o No upper limit on data volumes, processing throughput and
BigQuality also integrate numbers of processing nodes
with IBM business
glossary and data o Ability to run natively on Hadoop YARN bypassing Hadoop
modelling software frameworks such as MapReduce, Tez and Spark

Data integration jobs can • Integration with business glossary and data modeling software in the
also be published in an same EIM platform via shared metadata
information catalog to • An information catalog and the ability to publish data integration jobs as
offer trusted information
data services in InfoSphere Information Governance Catalog so that
services to information
information consumers can see what data services are available for
consumers
them to shop for and order via InfoSphere Data Click

IBM STREAMS – REAL-TIME OPTIMISATION USING BIG DATA

IBM Streams is IBM’s IBM Streams is a platform for building and deploying continuous real-time
platform for developing analytic applications that analyse data in motion. These applications
real-time analytic continuously look for patterns in data streams. When detected, their impact is
applications on analysed and instant real-time decisions made for competitive advantage.
streaming data Examples include analysis of financial market trading behaviour, analysis of
This includes sensor RFID data for supply and distribution chain optimisation, monitoring sensor
data from supply
chains, production
lines, assets, IoT data,
Copyright © Intelligent
financial markets data, Business Strategies Limited, 2016, All Rights Reserved 19
and also unstructured
text, images and video
Architecting a Platform For Big Data Analytics – 2nd Edition

data for manufacturing process control, monitoring sensor data to understand

product performance and usage in the Internet of Things (IoT), neonatal ICU
monitoring, real-time fraud prevention and real-time multi-modal surveillance in
law enforcement. IBM Streams can simultaneously monitor multiple streams of
external and internal events whether they are machine generated or human
generated. High volume structured and unstructured streaming data sources
are supported including text, images, audio, voice, VoIP, video, web traffic,
email, geospatial, GPS data, financial transaction data, satellite data, sensors,
and any other type of digital information.

IBM Streams ships with To help expedite real-time analytic application development, IBM also ships
pre-built toolkits and with pre-built analytical toolkits and connectors for popular data sources. Third
connectors to expedite party analytic libraries are also available from IBM partners. In addition, an
development of real- Eclipse based integrated development environment (IDE) is included to allow
time analytic organisations to build their own custom built real-time analytic applications for
applications
stream processing. It is also possible to embed IBM SPSS predictive models or
analytic decision management models in IBM Streams analytic application
workflows to predict business impact of event patterns.

Events can be analysed Scalability is provided by deploying IBM Streams applications on multi-core,
and acted upon in real- multi-processor hardware clusters optimised for real-time analytics and via
time or filtered and integration with Apache Spark. Events of interest to the business can also be
stored for further filtered out and pumped to other IBM analytical data stores for further analysis
analysis and replay in
and/or replay. IBM Streams can therefore be used to continually ingest data of
IBM BigInsights
interest into IBM BigInsights to analyse. It is also possible to summarise high
volume data streams and route these to IBM Cognos Analytics for visualisation
in a dashboard for further human analysis.

ACCESSING THE LOGICAL DATA WAREHOUSE USING IBM BIG SQL AND
IBM FLUID QUERY
IBM can also simplify Users and analytic applications need access to data in a variety of data
access to multiple data
repositories and platforms without concern for the data’s location or access
stores via a federated
SQL interface
method or the need to rewrite a query. To make this possible, IBM provides Big
SQL and Fluid Query.
IBM Big SQL can
access data in both IBM Big SQL
Hadoop and relational IBM Big SQL is IBM’s flagship multiple-platform SQL query engine for
DBMSs accessing Hadoop and non-Hadoop data or both. It therefore creates a Logical
Data Warehouse layer over the top of multiple underlying analytical data stores
IBM Big SQL can be and can federate queries to make those platforms work locally on the
used to create a Logical necessary data. Business analysts can connect directly to Big SQL from self-
Data Warehouse layer service BI tools that generate SQL. Data scientists and IT developers who
over the top of multiple want to access Hadoop and non-Hadoop data using SQL from within their
underlying data stores
analytic applications can also use it.

Copyright © Intelligent Business Strategies Limited, 2016, All Rights Reserved 20

Architecting a Platform For Big Data Analytics – 2nd Edition

IBM Big SQL is a Spark When processing in-bound SQL, IBM Big SQL bypass Hadoop MapReduce,
compliant massively Tez and Spark execution environments. Instead it runs natively under YARN on
parallel SQL engine that a Hadoop cluster with direct access to HDFS and HBase data. Big SQL is fully
can be used by Spark integrated with the Hive metastore and can therefore see Hive Tables, Hive
analytic applications and SerDes, Hive partitioning and Hive Statistics. It is also fully compliant with
BI tools to query Spark but does not require Spark. By that we mean that IBM Big SQL can be
Hadoop and non- used by Spark analytic applications written in Python, Java, Scala and R to
Hadoop data access data as an alternative to Spark SQL. This works because Spark
applications can access data via Big SQL and get back Spark RDDs to analyse
IBM Big SQL provides data in memory across a cluster. The difference is that there is more
support for the full 2011 functionality in Big SQL, it is fully ANSI 2011 compliant and has optimisation
ANSI SQL standard
capability to perform query re-write to improve performance. It supports
aggregate, scalar and OLAP functions, virtual tables, JAQL UDFs for analysis
It supports complex data of unstructured data, and data types such as STRUCT, ARRAY, MAP and
types, OLAP functions BINARY to handle more complex data. In addition, it supports Hadoop
and can use UDFs to columnar file formats such as ORC, Parquet, and RCFile and has no
analyse unstructed data
proprietary storage format of its own. In terms of security, IBM Big SQL offers
role-based access plus column and row security. IBM Big SQL can also
potentially ‘push down’ query functionality into Spark to execute if it deems it
necessary (e.g. to make use of GraphX functions).

IBM Fluid Query

IBM Fluid Query is
included with IBM IBM Fluid Query is included with IBM PureData System for Analytics. It
PureData System for provides access to data in Hadoop from IBM PureData System for Analytics
Analytics appliances and enables the fast movement of data between Hadoop and IBM
PureData System for Analytics appliances. IBM Fluid Query allows queries
against PureData System for Analytics, Hadoop or both by merging results
IBM Fluid Query allows from PureData System for Analytics database tables and Hadoop data sources
BI tools and analytic thus creating powerful analytic combinations. Therefore existing queries,
applications to query reports, and analytics can run against data on Hadoop, in addition to the data
data in PureData in a PureData System for Analytics appliance.
System for Analytics,
Hadoop or both This means that IBM Fluid Query allows IBM PureData System for Analytics to
Data in PureData route a query (or even part of a query) to the correct data store. It provides IBM
System for Analytics, with the ability to combine data in PureData System for Analytics with Hadoop
Hadoop can be joined to data without having to know how to connect to Hadoop. IBM Fluid Query
data in Hadoop connectivity to a ‘back end’ Hadoop can be via Hive, IBM Big SQL, Cloudera
Impala or SparkSQL.

Analytical Tools
A wide range of tools A wide range of third party and IBM analytical tools like IBM Cognos Analytics,
can leverage scalable IBM Watson Analytics, IBM SPSS, BigSheets, IBM BigR and IBM BigSheets
analytics running in the can all make use of BigSQL and Fluid Query to access data and invoke in-
IBM Analytics Platform database, in-memory, in-Hadoop and in-stream scalable analytics running in
the IBM Analytics Platform.

Copyright © Intelligent Business Strategies Limited, 2016, All Rights Reserved 21

Architecting a Platform For Big Data Analytics – 2nd Edition

CONCLUSIONS
With technology advancing rapidly companies need to work out how to piece
Organisations need to
understand what they
together and integrate the necessary components to maximise their ability to
want to achieve though produce disruptive insight. They need to be able to define what they need to
the production of produce to cause disruption, understand their data and analytical requirements
disruptive insight and then select the technology components necessary to be able to get
started. It should be possible to get started quickly in the cloud and then move
on-premises if needs be.
They need to
understand data and In addition they also need to be able to produce disruptive insight in a
analytical requirements productive manner without the need for major re-training to use new
before selecting
technologies like Hadoop, Spark and streaming analytics. To that end, if
technology components
companies can make use of existing tools used in traditional data warehousing
It should be possible to to also clean, integrate and analyse data in big data environments then the
start in the cloud and time to value will come down significantly. Also, tools should be able to exploit
move on-premises or the scalability of underlying hardware, when data volumes and data velocity is
create a hybrid solution high, without users needing to know how that is done.

Exploiting existing tools It should also be possible to simplify access to data in multiple data stores and
and skillsets that can join data across multiple data stores so that complexity is kept to a minimum.
exploit traditional and This is true irrespective of whether a business analyst needs to access data
big data technologies from a self-service analytics tool or whether it an IT developer or data scientist
fuels productivity needs to access data from a custom built analytical application. As such a
common federated SQL layer is needed to create a ‘Logical Data Warehouse’.
IBM is building an
IBM is pursuing all of this both on the cloud and on-premises with the ability
architecture for Big Data
and traditional analytics
deploy BigInsights, Apache Spark Open Platform with Apache Hadoop both on
on the cloud and on- the cloud and on premises. In addition it is creating a common scalable
premises analytical RDBMS code base that works across its dashDB, Big SQL, DB2,
PureData System for Analytics Appliances and as a software version on private
Data virtualisation and cloud with dashDB for software defined environments. Also Spark is being
federated query support integrated everywhere to push down analytics to make them run as close to the
simplifies access to data as possible. In addition, Big SQL and Fluid Query simplify access to
traditional and big data traditional data warehouses and Hadoop, helping to create a logical data
stores by creating a warehouse layer. And there is more to come.
Logical Data
Warehouse Today we are beyond the point where big data is in the prototype stage. We
are entering an era where automation, integration and end-to-end solutions
need to be built rapidly to facilitate disruption. Companies need to architect a
platform for Big Data (and traditional data) analytics. Given this requirement,
IBM would have to be a short list contender to helping any organisation
become a disrupter whether it be on the cloud or on-premises.

Copyright © Intelligent Business Strategies Limited, 2016, All Rights Reserved 22

Architecting a Platform For Big Data Analytics – 2nd Edition

About Intelligent Business Strategies

Intelligent Business Strategies is a research and consulting company whose
goal is to help companies understand and exploit new developments in
business intelligence, analytical processing, data management and enterprise
business integration. Together, these technologies help an organisation
become an intelligent business.

Author
Mike Ferguson is Managing Director of Intelligent Business Strategies
Limited. As an analyst and consultant he specialises in business intelligence
and enterprise business integration. With over 34 years of IT experience, Mike
has consulted for dozens of companies on business intelligence strategy, big
data, data governance, master data management and enterprise architecture.
He has spoken at events all over the world and written numerous articles. He
has written many articles, and blogs providing insights on the industry.
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, an
independent analyst organisation. He teaches popular master classes in Big
Data Analytics, New Technologies for Business Intelligence and Data
Warehousing, Enterprise Data Governance, Master Data Management, and
Enterprise Business Integration.

INTELLIGENT
BUSINESS
STRATEGIES

Water Lane, Wilmslow

Cheshire, SK9 5BG
England
Telephone: (+44)1625 520700
Internet URL: www.intelligentbusiness.biz
E-mail: [email protected]

Architecting a Platform For Big Data Analytics – 2nd Edition

Copyright © 2016 by Intelligent Business Strategies
All rights reserved

Copyright © Intelligent Business Strategies Limited, 2016, All Rights Reserved 23