The Maturity Model: Taking the Growing Pains Out of Hadoop

Grab some
coffee and
enjoy the
pre-show
banter
before the
top of the
hour!

The Briefing Room
The Maturity Model: Taking the Growing Pains Out of Hadoop

Twitter Tag: #briefr The Briefing Room
Welcome
Host:
Eric Kavanagh
eric.kavanagh@bloorgroup.com
@eric_kavanagh

  Reveal the essential characteristics of enterprise
software, good and bad
  Provide a forum for detailed analysis of today s innovative
technologies
  Give vendors a chance to explain their product to savvy
analysts
  Allow audience members to pose serious questions... and
get answers!
Mission

Topics
June: INNOVATORS
July: SQL INNOVATION
August: REAL-TIME DATA

Three Types of Evolution
Ø  Disruptive selection
Ø  Stabilizing selection
Ø  Directional selection

Types of Evolution?
WHAT GOES AROUND, COMES AROUND!

Analyst: Rick van der Lans
Rick F. van der Lans is an independent
analyst, consultant, author and
lecturer specializing in data
warehousing, business intelligence,
analytics, big data and database
technology. He is Managing Director of
R20/Consultancy. He has advised many
large companies worldwide on
defining their business intelligence
architectures. His popular IT books
have been translated into many
languages and have sold over 100,000
copies. Rick writes for TechTarget.com
and B-eye-Network. For the last 25
years, he has been presenting
professionally around the globe and at
international events.

Think Big, A Teradata Company
  Last year Teradata acquired Think Big Analytics, Inc., a
consulting and solutions company focused on big data
solutions
  Think Big has expertise in implementing a variety of open
source technologies, such as Hadoop, Hbase, Cassandra,
MongoDB and Storm, as well as experience with
Hortonworks, Cloudera and MapR
  Its consultants can assist with the planning, management
and deployment of big data implementations

Guest: Ron Bodkin
Ron Bodkin is Founder & President of Think Big, A
Teradata Company. Ron founded Think Big to help
companies realize measurable value from Big Data. The
company’s expertise spans all facets of data science
and data engineering and helps our customers to drive
maximum value from their Big Data initiative.
Previously, Ron was vice president Engineering at
Quantcast where he led the data science and engineer
teams that pioneered the use of Hadoop and NoSQL for
batch and real-time decision making. Prior to that, Ron
was Founder of New Aspects, which provided enterprise
consulting for Aspect-oriented programming. Ron was
also Co-Founder and CTO of B2B applications provider
C-Bridge, which he led to team of 900 people and a
successful IPO. Ron graduated with honors from McGill
University with a B.S. in Math and Computer Science.
Ron also earned his Master’s Degree in Computer
Science from MIT, leaving the PhD program after
presenting the idea for C-bridge and placing in the
finals of the 50k Entrepreneurship Contest.

MAKING BIG DATA COME ALIVE
 The Maturity Model: Taking the Growing Pains out of
Hadoop
 Ron Bodkin
 Founder and President, Think Big

12
•  What is Big Data?
•  Hadoop Maturity Spectrum
(from MapReduce to Integrated
Hadoop Architecture)
•  Taking the Growing Pains out of
Hadoop
Agenda

13
•  Data sets so large and
complex that they become
awkward to work with using
standard tools and
techniques
•  Google, Quantcast, Yahoo,
LinkedIn… customer
innovation in open source
•  Discover value in dark data
and relationships among
data sets
–  Customer behavior
–  Product behavior
•  Evolution:
insight -> operational ->
business innovation
Source: Hortonworks
What is Big Data?
© 2015 Think Big, a Teradata Company 6/16/15

14
Hadoop (Circa 2008)
•  HDFS - Distributed File System
•  MapReduce distributed parallel
batch processing
•  Community Open Source since 2005
many contributors
•  Industry commitment: over $1 Billion
invested in startups, all major
software companies embrace
•  3 replicas for HA
•  50 PB+ scale
•  Great for analytics, batch processing

15
SQL on Hadoop (Circa 2009 to Today)
•  Expanded access to data in
Hadoop
•  Great for exploration of
“raw” data, e.g., by data
scientists
–  Can integrate UDFs
–  Tools integration expanding
access beyond SQL power users
•  Fast moving space
•  Also used for ETL/ELT
processing

16
•  Support for many processing engines including Spark, Storm, Tez
•  Common cluster with local data access in HDFS
Hadoop (Circa 2012)
YARN - Yet Another Resource Negotiator
16

17 © 2015 Think Big, a Teradata Company 6/16/15
Apache Spark (Circa 2013)
Spark SQL
Spark
Streaming
MLib
(Machine
learning)
GraphX
(Graph)
Apache Spark Core Engine
YARN
JDBC/ODBC
HDFS

18
OLAP on Hadoop (Circa 2014)
Events
Reporting, Visualization and Analytics
Traditional BI on Hadoop
•  Day or week old data
•  Long latencies to retrieve data
•  Doesn’t scale as number of events,
aggregates, or users grow
•  Expensive, siloed event analytics
•  Real time data
•  Unlimited custom reporting
(drill to trillions of aggregates)
•  Scales to thousands of simultaneous
users
•  Subsecond latency
•  Omni-channel analytics
OLAP on Hadoop

19
Today: Integrated Hadoop Architecture

MAKING BIG DATA COME ALIVE
 How to Take the Growing Pains out of Hadoop

21
•  Picking the wrong starting point
– ho hum, boil the ocean
•  Immature Governance
•  Siloed Organization
•  Change Management
•  Skills Gap
•  Business as usual
•  Not fitting in with ecosystem
Big Data Pitfalls

22
Big Data Center of Excellence

23
Cross-Functional Collaboration
Data
Weak intersection between Data, Systems
and Business often means project failure
Current State Improved State
Collaborative integration of Analytics with
Big Data capabilities for success.
Data
Weak Area with
diminished influence
Cross-functional
collaboration
Data-driven
business alignment

24
A “Data Lake” is a centralized approach to capturing, refining, storing, and
exploring any form of raw data at scale, enabled by low cost technologies
from which downstream facilities may draw.
What is a Data Lake?
Information Sources Data Lake Downstream
Facilities
Data Variety is the driving factor in building a Data Lake.

25
Why a Data Lake?
•  Benefit from open source big data innovation and
cost savings
•  Control over enterprise data with governance
•  Offload history of operational and analytical data
platforms
•  Remix ETL, freeing capacity downstream
•  Gain ultimate flexibility in data use and access
•  Foundation for solving business problems
–  Big data solutions enable new uses of data via working with
new data sets and analytics

26
•  Corporate Repository – Data sourcing repository, possibly system of
record
•  Active Archive – Offloaded history of operational and analytical data
platforms that needs to be used infrequently
•  Discovery – Data Profiling, schema on read, wrangling, discover signals
•  ETL Offload – Move all or parts of current ETL processes into Hadoop
•  NEW ETL Development – Develop new ETL processes on the Data Lake
•  Analytics – Data Science, Custom/product built Analytics solutions (i.e.
churn, machine learning)
•  Business Intelligence/Reporting – Event analysis, dimensional rollup and
pivot reports
Data Lake Use Cases

27
Swamp Reservoir
Data Lake: Swamp or Reservoir?

28
Data Lake
Information Sources
Evaluate
Source Data
Ingest
Collect & Manage
Metadata
Profile - Structure
Sequence
Downstream
Facilities
Generate Reports
Discovery Signals
Compress
Automate
Protect
Prepare Data
for Ingest
Prepare Source
Metadata
A centralized well-governed repository of raw data and metadata into
which all data-producing streams flow from which business analysts &
downstream facilities may draw.
Data Reservoir

29
Metadata is required to facilitate location of
and entitlements to data
Schema Metadata is a foundation, but…
Operational Metadata & Business-Security is
critical to governance
•  Where did it come from? … What is the data serialization?
•  Who owns the data? Who can see the data? Who belongs to
which group?
•  What is the environment (landing zone, OS, Line of Business)?
•  What processes touched my data?
•  Did you lose any data (Checksums, etc.)?
•  When did the data get ingested, transformed?
•  Did it get exported, or archived…when, where how will it be used
(organizational)?
Governance & Security

30
High Tech Manufacturer Example
•  Enterprise-wide data access (ODS) for timely analytics and insights
•  Foundation for large scale proactive analytics
•  Reduce $Millions in scrap waste
•  Drives revenue and market share by accelerating time to market
Legacy Data Lake
Retention 3-6 months scattered DBs
then tape archive
All data online in common
data reservoir for 3+ years
Coverage Summaries, samples for
several data sets
All parametric data
captured in raw and
integrated form
Analysis Reactive resulting in missed
improvement opportunities
In daily operations and on
larger data sets to support
proactive improvements

31
•  Insight into new data (one time
analysis)
•  Decision support models (offline
insights in production)
•  Operational models (real-time
analytics)
•  Hadoop opportunities: new data,
new algorithms, different
resolution, frequency
•  Hadoop blends with MPP,
analytics, tools not the sole tool
Data Science Analytics

32
Real Time Analytics Architecture
Clickstream
Call Center
email
Brick & Mortar
Batch Storage
(HDFS)
Event Repository
(HBase)
API
Server
CIM
ML
Personali
zation
A/B
Testing
Reporting
Call
Center
Channel
Attribution

33
•  We’ve come a long way since
MapReduce 1.0 enabled by YARN
•  Don’t forget about the last mile for
business (SQL on Hadoop, OLAP for
Hadoop, training, etc.)
•  Data lakes without governance,
security, metadata management
equates to failed projects/loss of $$
•  Analytics on big data is where the
real business value lies – from
understanding to models to
operationalization
Exploiting the Power of Hadoop

34
•  100% Big Data Focus
•  Founded in 2010 with100+ engagements across 70 clients
•  Unlock value of big data with data science and data
engineering services
•  Proven vendor-neutral open source integration expertise
•  Agile team-based development methodology
•  Think Big Academy for skills and organizational development
•  Global delivery model
Who is Think Big?

Perceptions & Questions
Analyst:
Rick van der Lans

Copyright
©
1991
-‐
2015
R20/Consultancy
B.V.,
The

Hague,
The
Netherlands.
All
rights
reserved.
No
part

of
this
material
may
be
reproduced,
stored
in
a

retrieval
system,
or
transmiHed
in
any
form
or
by

any
means,
electronic,
mechanical,
photographic,
or

otherwise,
without
the
explicit
wriHen
permission
of

the
copyright
owners.

The
Next
Stage
of

Hadoop
and
Big
Data:

Simpliﬁca9on

by

Rick
F.
van
der
Lans

R20/Consultancy
BV

TwiHer
@rick_vanderlans

www.r20.nl

Copyright © 1991 - 2015 R20/Consultancy B.V., The Hague, The Netherlands 37
The
Adop9on
Stages
of
Technologies

  Stage 0: Introduction and early adoption
  Stage 1: Adoption by more traditional customers
  Stage 2: Universal adoption
Stage 0 Stage 1 Stage 2 Stage n

Adop9on
Stage
0

  Functionality: Minimal
  Scalability and performance: Promising
  Customers: Those with very specific, urgent needs;
they have tech-savvy developers
  Use cases: Specific
  Integration: Isolated solutions

Adop9on
Stage
1

  Functionality: Sufficient for some mission-critical
systems
  Scalability and performance: Medium
  Customers: Those with very specific, urgent needs;
they have tech-savvy developers
  Use cases: More general
  Integration: Isolated solutions

Adop9on
Stage
2

  Functionality: Full
  Scalability and performance: High
  Customers: Traditional; not always with tech-savvy
developers; focus on productivity and maintainability
  Use cases: Specific and general
  Integration: Integrated with traditional solutions

Diﬀerent
SQL-‐on-‐Hadoop
Solu9ons

Apache

MapReduce
API

Apache

MapReduce

Apache
HDFS

API

Apache
HDFS

Apache
HDFS

API

Apache
HDFS

Apache
HBase

API

Apache
HBase

A
SQL
Dialect

SQL-‐on-‐Hadoop

A
SQL
Dialect

SQL-‐on-‐Hadoop

Apache
HDFS

API

Any
HDFS

A
SQL
Dialect

SQL-‐on-‐Hadoop

Apache
HDFS

API

Any
HDFS

A
SQL
Dialect

SQL-‐on-‐Hadoop

Spark
API

Spark

Self-‐Service
Data
Prepara9on

  Non-technical interface for
studying data files
  Easy way of defining rules
  Data is fixed by defining
filters, not by changing data
in source systems
  Relationship with Data
Blending

Example:
Data
Prepara9on

Stage
2
=
Simpliﬁca9on

Questions on acceptance
You talked about the skills gap. Sometimes Hadoop is used to develop operational systems, and
sometimes to develop BI systems. Now, many specialists working in BICCs are good ETL programming,
reporting, data modeling, but not so good at low-level programming. For example, many of them
have never programmed Java. What would you recommend them to do to be able to use Hadoop?
Should organizations use Hadoop for developing new systems, or for migrating existing systems? How
easy do you think it is to integrate a new Hadoop-based system with the current infrastructure?
Normally, when a technology becomes more and more popular, it becomes clearer and clearer to
everyone what the costs and the risks are. For example, how much man-hours should be spend on
managing a Hadoop environment? So, what’s the productivity of a MapReduce programmer versus a
SQL programmer?
Questions on architectural
Now that so much data is produced in a distributed fashion, is the concept of physically centralizing
data for integration really realistic? Isn’t big data too big to move?
Over the years, large multi-national organizations have always struggled to develop one large
centralized enterprise-wide data warehouse. In most cases, the problems were organizational. Who
owns that data, and so on? Why would the data lake or data reservoir succeed today?

Questions on technical aspects
It was clear that for you YARN is a key module in the Hadoop platform. How mature is YARN today?
How good is it, for example, in avoiding one massive query that monopolizes the entire platform and
slows down all other applications?
My feeling is still that HDFS is not good at joins of files. The SQL-on-Hadoop engines are not strong in
joins either. Do you think this will change? For example, can you take a star model from a SQL-based
data mart, consisting of 7 dimensions and 1 fact, migrate it to Hadoop and expect a great
performance?
Any idea what the tipping point is, when to forget about a proprietary SQL and move to Hadoop with
SQL-on-Hadoop?
You mentioned OLAP on Hadoop using Spark and you indicated that such a solution scales to thousands
of simultaneous users. Is that really true? Has that been proven? Or does that work when the reporting
tools extract data and load it in memory of the client machine?
Almost every vendor of some big data tool publishes benchmark results showing how fast their
product is. But many of these benchmarks are single user, single query benchmarks, which I think is
not that useful. Why are they not showing multi-users benchmark results? Do you think they’re hiding
something?

On one of your slides you said that Hadoop is 50 PB+ scale. But realistically, how many companies on
this planet want to store that much data? And if it is 50PB+ scale, doesn’t that mean for the majority
of the companies it’s overkill of functionality? In general, how big should big be to justify the Hadoop
platform?
To end this list of questions and the finalize the briefing, you indicated that analytics on big data is
where the real business value lies. I fully agree with that, but can you elaborate on this a little more?
Questions on technical aspects, cont’d.

Upcoming Topics
www.insideanalysis.com
June: INNOVATORS
July: SQL INNOVATION
August: REAL-TIME DATA

THANK YOU
for your
ATTENTION!
Some images provided courtesy of Wikimedia Commons and
"Selection Types Chart" by Azcolvin429 - Own work. Licensed under CC BY-SA 3.0 via Wikimedia Commons -
https://commons.wikimedia.org/wiki/File:Selection_Types_Chart.png#/media/
File:Selection_Types_Chart.png and http://sagamer.co.za/img/evolution-phone.jpg

The Maturity Model: Taking the Growing Pains Out of Hadoop

Recommended

More Related Content

What's hot (20)

Similar to The Maturity Model: Taking the Growing Pains Out of Hadoop (20)

More from Inside Analysis (20)

Recently uploaded (20)

The Maturity Model: Taking the Growing Pains Out of Hadoop