Upload
prasad-billahalli
View
213
Download
0
Embed Size (px)
Citation preview
7/29/2019 BigDataReport Final
1/35
ExploitingBig DataSeies Iei ih Hd
Deie Bsiess Isihs
By Wayne eckerson
Dirctr f Rarc, Bin Alicatin and Arcitctr Mdia Gr, TcTargt, Jn 2012
7/29/2019 BigDataReport Final
2/35
ExploItIng BIg Data: StratEgIES or IntEgratIng wItH HaDoop to DElIvEr BuSInESS InSIgHtS 2
e s
The big daTa movemenT is transorming the business intelligence (BI) indus-
try. The term not only reers to large volumes o multi-structured data but
also new technologies, specically Hadoop and analytical databases. The
combination expands the horizon or analytical inquiries, providing exception-
ally high price-perormance and new, fexible data processing rameworks.Analytical databases oer exceptionally high price-perormance or ana-
lytical workloads. Analytical databases come as appliances, sotware-only
products or cloud-based services. The ormer are easy to deploy and man-
age, while the latter integrate seamlessly into an existing data center, giv-
ing administrators fexible deployment
options. Companies are implementing
analytical databases to provide new ana-
lytical capabilities or augment or replace
aging data warehousing platorms. Theyare also using them to serve as analytical
sandboxes that ofoad new or existing
analytical workloads rom overburdened
data warehouses.
Hadoop adds large-scale storage o
multi-structured dataWeb server logs,
video, audio, systems logs and sensor
datato the mix o content that organiza-
tions can capture and rene or business analysis. It also increases agility byletting data proessionals land data without rst having to transorm it into
an ocially sanctioned schema. And because o its open source heritage,
Hadoop reduces the cost o storing and transorming large volumes o data.
About 11% o surveyed organizations have implemented Hadoop, and around
hal o these early adopters are just experimenting with the system. Nonethe-
less, the hype over Hadoop has grown to a deaening crescendo, and many BI
Hadoop adds large-scale
storage o multi-structureddata to the mix o content
that organizations can
capture and refne or
business analysis.
executive
summary
research
background
he new analytical
ecosystem
big data trends
integrating
with hadoop
Future oF hadoop
recommendations
7/29/2019 BigDataReport Final
3/35
ExploItIng BIg Data: StratEgIES or IntEgratIng wItH HaDoop to DElIvEr BuSInESS InSIgHtS 3
executive summary
proessionals are bullish about its uture. Hadoop can provide the leverage
we need to take our BI environment to the next level, said one BI pro.
The growing popularity o Hadoop presents a curious challenge to estab-
lished BI vendors. On one hand, Hadoop represents a new source o data andopportunity or established vendors to drive revenue rom new and existing
products in an increasingly mature market. On the other, Hadoop, in its ull-
grown incarnation, could possibly supplant established commercial database
and data integration products with comparable open source products. But
today, both Hadoop and SQL database vendors are working together to build
a robust analytical ecosystem. There are our types o technical integration
vendors are pursuing, each with its advantages and disadvantages.
Whatever the uture holds, its clear that SQL tools and Hadoop will be
joined at the hip, giving users more options or storing, processing and analyz-ing data. n
executive
summary
research
background
he new analytical
ecosystem
big data trends
integrating
with hadoop
Future oF hadoop
recommendations
7/29/2019 BigDataReport Final
4/35
ExploItIng BIg Data: StratEgIES or IntEgratIng wItH HaDoop to DElIvEr BuSInESS InSIgHtS 4
rh Bgd
The purpose of this report is to describe the emerging big data ecosystem in
which Hadoop and analytical databases play an increasingly important role.
It will then examine ways that traditional BI, extract, transorm and load (ETL)
and database vendors integrate with Hadoop. Finally, it will speculate on
how Hadoop and analytical ecosystems will evolve and aect traditional BIapproaches, technologies and vendors.
The research is based on interviews with BI practitioners, briengs with BI
providers, including sponsors o this report, and a survey o BI proessionals.
The ve-minute survey was promoted to the BI Leadership Forum, an online
group o about 1,000 BI directors and managers, and my 2,000-plus Twitter
ollowers in April 2012. The survey was started by 170 people and completed
by 152 people or an 89.4% completion rate. Survey results are based on 158
respondents who completed the survey and indicated their positions as BI
or IT proessional, BI sponsor or user or BI consultant. Responses rompeople who selected the position o BI vendor or Other or who didnt com-
plete the survey were excluded rom the results. Among this set o qualied
BaSED on 158 rESponDEntS, BI LeADeRshIp FoRuM, January 2012.
75%BIprofessional
18%BIconsultant
7%BIsponsororuser
fgure 1.
Dgph:
rpd P
executive
summary
research
background
he new analytical
ecosystem
big data trends
integrating
with hadoop
Future oF hadoop
recommendations
http://www.bileadership.com/http://www.bileadership.com/http://www.bileadership.com/7/29/2019 BigDataReport Final
5/35
ExploItIng BIg Data: StratEgIES or IntEgratIng wItH HaDoop to DElIvEr BuSInESS InSIgHtS 5
researcH BackGrounD
respondents, most are BI or IT proessionals (75%) rom large companies
with more than $1 billion in annual revenues (53%). The industries with the
highest percentage o respondents are consulting (14%) and banking (11%)
(see fgures 1, 2 and 3). n
fgure 2.
Dgph: cp sz
fgure 3.
Dgph: id
53%Large($1B+)
22%Medium($100Mto$1B)
25%Small(ware
HealthcareRetail
Transpora0on
Telecommunica0ons
Internet
Government/Nonprofit
Educa0on
Media
Other
http://www.bileadership.com/http://www.bileadership.com/http://www.bileadership.com/http://www.bileadership.com/http://www.bileadership.com/http://www.bileadership.com/7/29/2019 BigDataReport Final
6/35
ExploItIng BIg Data: StratEgIES or IntEgratIng wItH HaDoop to DElIvEr BuSInESS InSIgHtS 6
th nw a e
Theres a revoluTion aoot in the way organizations architect the fow o data
to support reporting and analysis applications. The ringleader o the revolu-
tion is big data. There are two types o big data products in the market today.
There is open source sotware, centered largely on on Hadoop, which elimi-
nates up-ront licensing costs or managing and processing large volumes odata. And there are new analytical engines, including appliances and column
stores, which provide signicantly higher price-perormance than general-
purpose relational databases.
Both sets o big data sotware deliver higher returns on investment than
previous generations o data management technology, but in vastly dierent
ways. Organizations are gradually imple-
menting both types o systems, among oth-
ers, to create a new analytical ecosystem.
undersTanding hadoop
For many, big data has become syn-
onymous with Hadoopan open source
ramework or parallel computing that runs
on a distributed le system (such as the
Hadoop Distributed File System, or HDFS.)
For the rst time, Hadoop enables orga-
nizations to cost-eectively store all theirdata as well as multi-structured data, such as Web server logs, sensor data,
email and extensible markup language (XML) data. Because multi-structured
data consists o 80% o all data by most estimates, the advent o Hadoop
heralds the dawn o a new age o data processing in which organizations can
pluck the needle out o the data haystack, which consists o terabytes, i not
petabytes, o inormation.
Because multi-structureddata consists o 80% o all
data by most estimates,
the advent o Hadoop
heralds the dawn o a new
age o data processing.
executive
summary
research
background
he new analytical
ecosystem
big data trends
integrating
with hadoop
Future oF hadoop
recommendations
7/29/2019 BigDataReport Final
7/35
ExploItIng BIg Data: StratEgIES or IntEgratIng wItH HaDoop to DElIvEr BuSInESS InSIgHtS 7
tHe neW analytical ecosystem
B. Hadoop, an open source project hosted by the Apache Sotware
Foundation, brings three major benets to the world o analytical processing:
1.lw . Because Hadoop is open source sotware that runs on com-modity servers, it radically alters the nancial equation or storing and
processing large volumes o dataat least in terms o up-ront licensing
costs. With Hadoop, organizations can nally store all the data they gen-
erate in its raw orm without having to justiy its business value up ront.
This creates a low-cost staging and rening area and osters greater data
exploration and reuse.
2.ld d g. Compared with relational databases, Hadoop does not
require developers to convert data to a specic ormat and schema (suchas elds with xed data types, lengths, labels and relationships) to load
and store it. Rather, Hadoop is a load-and-go environment that handles
any data ormat and speeds load cycles, which is critical when ingesting
terabytes o data.
3. Pd d. Finally, Hadoop MapReduce enables developers to use
Java, Python or other programming languages to process the data. This
means developers dont have to learn SQL, which is not as expressive as
procedural code or certain types o analytical work, such as data miningoperations or inter-row calculations. MapReduce is also implemented in a
number o analytical databases to augment the amiliarity and enterprise
adoption o SQL.
Given these benets, the data management industry is understandably
abuzz about big data. Organizations are now exploiting creative ways to use
Hadoop. For example, Vestas Wind Systems, a wind turbine maker, uses
Hadoop to model larger volumes o weather data so it can pinpoint the opti-
mal placement o wind turbines. And a nancial services customer usesHadoop to improve the accuracy o its raud models by addressing much larg-
er volumes o transaction data.
Dwb. But Hadoop is not a data management panacea. Its essentially
a 1.0 product that is missing many critical ingredients o an industrial-proo
data processing environmentrobust security, a universal metadata catalog,
executive
summary
research
background
he new analytical
ecosystem
big data trends
integrating
with hadoop
Future oF hadoop
recommendations
7/29/2019 BigDataReport Final
8/35
ExploItIng BIg Data: StratEgIES or IntEgratIng wItH HaDoop to DElIvEr BuSInESS InSIgHtS 8
tHe neW analytical ecosystem
backup sotware, rich management utilities and in-memory parallel pipelining.
Moreover, Hadoop is a batch-processing environment that doesnt support
speed-o-thought, iterative querying. And although higher-level languages
exist, Hadoop today requires data scientists who know how to write Javaor other programs to run data processing jobs, and these olks are in scarce
supply.
To run a production-caliber Hadoop environment, you need to get sot-
ware rom a mishmash o Apache projects, which have colorul names like
Flume, Sqoop, Oozie, Pig, Hive and ZooKeeper. These independent projects
oten have competing unctionality and separate release schedules, and they
arent always tightly integrated. And each project evolves rapidly. Thats why
there is a healthy market or Hadoop distributions that package these compo-
nents into a reasonable set o imple-mentable sotware. Its also why the
jury is still out on the total cost o
ownership or Hadoop. Today the
cost o putting Hadoop into produc-
tion is high, although this will improve
as the environment stabilizes and
administrative and management utili-
ties are introduced and mature.
Many BI vendorsincluding data-base, data integration and reporting
and analysis vendorsare working to
integrate with Hadoop, making it eas-
ier to access, manipulate and query
Hadoop data than is currently possible with native Apache sotware. The
rapid merging o traditional BI and Hadoop environments creates a new ana-
lytical ecosystem that expands the ways organizations can eectively exploit
data or business gain.
adp . Leading-edge companies rom a variety o industries have
already implemented Hadoop, and many more are considering it. A survey o
the BI Leadership Forum in April 2012 shows that 10% o organizations are
implementing Hadoop or have already put it into production, while 20% are
experimenting with it in house and 32% are considering it. On the downside,
38% have no plans to use Hadoop (see fgure 4, page 9).
Today the cost o putting
Hadoop into production is
high, although this will im-
prove as the environment
stabilizes and administrative
and management utilitiesare introduced and mature.
executive
summary
research
background
he new analytical
ecosystem
big data trends
integrating
with hadoop
Future oF hadoop
recommendations
7/29/2019 BigDataReport Final
9/35
ExploItIng BIg Data: StratEgIES or IntEgratIng wItH HaDoop to DElIvEr BuSInESS InSIgHtS 9
tHe neW analytical ecosystem
Many BI directors have heard the buzz around Hadoop and are starting to
explore its value. Most are ocused on use cases that Hadoop can support,
ranging rom technical workloads such as data processing, reporting and data
mining to business applications such as raud detection, risk modeling, churn
analysis, sentiment analysis and cross-selling. But most havent sucientlyexamined how Hadoop ts into and extends their current BI architecture.
undersTanding analyTical daTabases
The other type o big data predates Hadoop by several years. It is less a
movement than an extension o existing relational database technology
optimized or query processing or advanced analytics. These analytical plat-
orms span a range o technology, rom appliances and columnar databases
to shared nothing, massively parallel processing (MPP) databases. And theyhave unique extensions that in some cases include a MapReduce processing
ramework or advanced analytics. The common thread among them is that
most are read-only environments that deliver exceptional price-perormance
compared with general-purpose relational databases originally designed to
run transaction processing applications.
Teradata laid the groundwork or the analytical platorm market when it
fgure 4.
adp Hdp
38%Noplans
20%Experimenting
4%
Inproduction
BaSED on 158 rESponDEntS,BI LeADeRshIp FoRuM, aprIl 2012.
executive
summary
research
background
he new analytical
ecosystem
big data trends
integrating
with hadoop
Future oF hadoop
recommendations
http://www.bileadership.com/http://www.bileadership.com/http://www.bileadership.com/7/29/2019 BigDataReport Final
10/35
ExploItIng BIg Data: StratEgIES or IntEgratIng wItH HaDoop to DElIvEr BuSInESS InSIgHtS 10
tHe neW analytical ecosystem
launched the rst analytical appliance in the early 1980s. Sybase was also an
early orerunner, shipping the rst columnar database in the mid-1990s. IBM
Netezza kicked the current market into high gear in 2003 when it unveiled
a popular analytical appliance and was soon ollowed by dozens o startups.Oracle launched the Exadata appliance in 2008, and it has been one o the
companys most successul products ever. Aster Data (now part o Tera-
data) tightly integrated MapReduce with SQL processing. Startup ParAccel,
which oers a sotware-only analytical database, has also achieved consider-
able traction, especially among nancial services and other companies that
require the highest level o perormance or complex queries and analytical
workloads. And several BI vendors, notably
MicroStrategy, Oracle and Tableau, have
been architecturally designed to exploitthese big data sources.
B. Although the price tag o ana-
lytical databases can oten exceed $1 mil-
lion including hardware, customers nd that
they are well worth the cost. Some use the
technology to ofoad complex analytical
workloads rom existing data warehouses
and avoid costly upgrades and time-con-suming rewrites o existing applications.
Others use SQL-MapReduce capabilities
embedded in some analytical databases to analyze multi-structured data at
scale and increase the accuracy o customer behavior models. But most use
analytical databases to replatorm aging data warehouses in an eort to elimi-
nate perormance bottlenecks and run a backlog o critical applications.
For example, Virginia-based XO Communications recovered $3 million in
lost revenue rom a new revenue assurance application it built on an analytical
appliance, even beore it had paid or the system. Gilt Groupe, an online luxuryretailer, uses SQL-MapReduce processing to tie together clickstream inorma-
tion with email logs, ad viewing inormation, Twitter and operational inorma-
tion to discover consumer preerences and improve the shopping experience.
Australian Finance Group, the largest provider o mortgage services in Austra-
lia, implemented an analytical appliance to accelerate query and log-in peror-
mance o a critical application used by its mortgage brokers, achieving a 42%
Although the price
tag o analytical data-
bases can oten exceed
$1 million including
hardware, customers
fnd that they are well
worth the cost.
executive
summary
research
background
he new analytical
ecosystem
big data trends
integrating
with hadoop
Future oF hadoop
recommendations
7/29/2019 BigDataReport Final
11/35
ExploItIng BIg Data: StratEgIES or IntEgratIng wItH HaDoop to DElIvEr BuSInESS InSIgHtS 11
tHe neW analytical ecosystem
return on investment in three years.
chg. Given the up-ront costs o analytical databases, organizations
usually do a thorough evaluation o these systems beore jumping on board.First, companies determine whether an analytical database outperorms their
existing data warehouse databases to a degree that warrants migration and
retraining costs. This requires a proo o concept in which a customer tests
the systems in its own data center using its own data across a range o que-
ries. The good news is that the new analyti-
cal databases usually deliver jaw-dropping
perormance or most queries tested. In
act, many customers dont believe the ini-
tial results and rerun the queries to makesure that the results are valid.
Second, companies must choose rom
more than two dozen analytical databases
on the market today. For instance, they
must decide whether to purchase an appli-
ance or a sotware-only system, a columnar
database or an MPP database, or an on-premises system or a Web-based
cloud service through vendors, such as Amazon Web Services. Evaluating
these options takes time, and many companies create a short list that doesntalways contain comparable products.
Finally, companies must decide what role an analytical platorm will play in
their data warehousing architectures. Should it serve as the data warehous-
ing platorm? I so, does it handle multiple workloads easily and scale linearly
to handle the growth in numbers o concurrent users? Or is it best used as
an analytical sandbox? I so, how well does it handle complex queries? Does
it have a built-in library o analytical algorithms and an extensibility rame-
work to add new ones? What kinds o data make sense to ofoad to the new
system? How do you rationalize having two data warehousing environmentsinstead o one?
Today, small and medium-sized companies which have tapped out their
Microsot SQL Server or MySQL data warehouses usually replace them with
analytical appliances to get better perormance. But large companies that have
invested millions o dollars in a data warehouse typically augment the data
warehouse with outboard systems designed to handle unique workloads or
Companies must decide
what role an analyticalplatorm will play
in their data warehous-
ing architectures.
executive
summary
research
background
he new analytical
ecosystem
big data trends
integrating
with hadoop
Future oF hadoop
recommendations
7/29/2019 BigDataReport Final
12/35
ExploItIng BIg Data: StratEgIES or IntEgratIng wItH HaDoop to DElIvEr BuSInESS InSIgHtS 12
tHe neW analytical ecosystem
types o data. Here the data warehouse becomes the hub that eeds multiple
downstream systems with clean, integrated data. This hub-and-spoke archi-
tecture helps organizations avoid a costly upgrade to their legacy data ware-
houses, which might exceed the cost o a new analytical database.
The new analyTical ecosysTem
fgure 5 depicts an architecture or a new analytical ecosystem that has the
ngerprints o big data all over it. The objects in blue represent the traditional
data warehousing environment, while those in pink represent new architec-
tural elements made possible by new big data and analytical technologies.
By extending the traditional BI architecture with Hadoop, analytical data-
bases and other analytical components, organizations gain the best o top-down reporting and bottom-up analytics in a single ecosystem. Traditionally,
pera ona sys ems(structureddata)
CasualuserOperationalExtract,transform and load
(Batch,nearreal timeorreal time)
BI
sys em
Operational
system
CEPengine
Machinedata
Hadoopcluster
Dept.datamart
Datawarehouse
Topdownarchitecture
Web data
Virtualsandboxes Bottomuparchitecture
In-
memorysandbox
Free
standing
sandbox
Audio/video
data
Analyticalplatformornonrelationaldatabase
PoweruserDocumentsandtext
Externaldata
fgure 5.
th nw a e
executive
summary
research
background
he new analytical
ecosystem
big data trends
integrating
with hadoop
Future oF hadoop
recommendations
7/29/2019 BigDataReport Final
13/35
ExploItIng BIg Data: StratEgIES or IntEgratIng wItH HaDoop to DElIvEr BuSInESS InSIgHtS 13
tHe neW analytical ecosystem
reports and analyses run in separate environmentsreports using data ware-
houses and analyses using Excel, Microsot Access and renegade data marts.
But the new analytical ecosystem brings business analysts into the main-
stream, enabling them to conduct reeorm analyses inside the corporate datainrastructure using a variety o analytical sandboxes.
tp-dw pg. The top-down world o reporting and dashboard
(depicted in the top hal o fgure 5) delivers answers to questions that users
dene in advance. BI developers gather requirements, stamp questions into
a multidimensional data model, nd relevant data, map it to the target model
and build interactive reports that answer the predened questions with
some options or urther exploration. Although the top-down world generally
processes data in large batches at night, it can also handle continuous datausing an operationalized data warehousing environment or a complex event
processing (CEP) engine when ultra low-
latency analytics is required. CEP systems
examine patterns within high-volume
event streams and trigger alerts and
workfows when predened thresholds
are exceeded.
B-p pg. The bottom-upworld o analysis (depicted in the lower
hal o fgure 5) answers new and unan-
ticipated questions using ad hoc queries,
visual exploration and predictive analyti-
cal tools. In the new analytical ecosystem,
business analysts have many options or
exploring and analyzing data. While they
used to be let to their own devices to collect, standardize and integrate data,
now they can access raw data in Hadoop or standardized, corporate data inthe data warehouse and mix it with their own data. They can then model and
explore data however they want in ocially sanctioned sandboxes without
inciting the ire o corporate IT managers. When theyre ready to share insights
with others, they publish their analyses to a corporate server and turn over
responsibility to the enterprise BI team to produce the output.
The new analytical ecosys-
tem brings business ana-
lysts into the mainstream,
enabling them to conduct
reeorm analyses inside
the corporate data inra-
structure using a variety
o analytical sandboxes.
executive
summary
research
background
he new analytical
ecosystem
big data trends
integrating
with hadoop
Future oF hadoop
recommendations
7/29/2019 BigDataReport Final
14/35
ExploItIng BIg Data: StratEgIES or IntEgratIng wItH HaDoop to DElIvEr BuSInESS InSIgHtS 14
a db. New analytical sandboxes let power users explore a
combination o enterprise and local data in a sae haven that wont interere
with top-down workloads. Dierent types o analysts use dierent sand-
boxes. Data scientists query raw, multi-structured data in Hadoop using Java,Python, Perl, Hive or Pig. Business analysts query virtual sandboxes in the data
warehouse or ree-standing analytical appliances designed or complex query
processing using SQL, built-in analytical algorithms or a combination o SQL
and MapReduce capabilities. Business analysts and superusers explore sub-
sets o data using in-memory visualization tools and immediately publish their
ndings as dashboards or departmental colleagues to consume.
D fw. In the new ecosystem, a lot o the source data fows through
Hadoop, which acts as a staging area and online archive or all o an organiza-tions data. This is especially true or multi-structured data, such as log les
and machine-generated data, but also or some structured data that com-
panies cant cost-eectively store and process in SQL engines (such as call
detail records in a telecommunications company). From Hadoop, much o the
data is ed into a data warehousing hub, which oten distributes data to down-
stream systems, such as data marts, operational data stores, analytical appli-
ances and in-memory visualization applications, where users can query the
data using amiliar SQL-based reporting and analysis tools. Some data is also
ed directly into ree-standing analytical databases and appliances. To builddata fows, organizations use graphical data integration and data quality tools
that hide the technical complexity o Hadoop and MapReduce code genera-
tion. The big data revolution brings major enhancements to the BI landscape.
First and oremost, it introduces new technologies, such as Hadoop, that
make it possible or organizations to cost-eectively consume and analyze
large volumes o multi-structured data. Second, it complements traditional,
top-down data-delivery methods with more fexible, bottom-up approaches
that promote ad hoc exploration and rapid application development.
But combining the top-down and bottom-up worlds is not easy. BI proes-sionals need to assiduously guard data semantics while opening access to data.
For their part, business analysts and data scientists need to commit to adhering
to corporate data standards in exchange or getting the keys to the kingdom. n
tHe neW analytical ecosystem
executive
summary
research
background
he new analytical
ecosystem
big data trends
integrating
with hadoop
Future oF hadoop
recommendations
7/29/2019 BigDataReport Final
15/35
ExploItIng BIg Data: StratEgIES or IntEgratIng wItH HaDoop to DElIvEr BuSInESS InSIgHtS 15
Bg D td
according To surveys o the BI Leadership Forum, user organizations have
invested signicantly in analytical databases while early adopters are taking
the plunge with Hadoop.
Almost hal o respondents to a 2011 BI Leadership Forum survey said they
have either implemented an analytical database (i.e., a sotware-only system)or an analytical appliance (i.e., a hardware-sotware combination). In contrast,
only 11% have implemented Hadoop (see fgure 6).
Companies have deployed all three types o platorms in a variety o ways.
One signicant standout is that 79% o companies that have deployed ana-
lytical appliances have implemented them to run data warehouses. In most
cases, this is to replace a legacy relational database or a SQL Server or MySQL
implementation that ran out o gas. For instance, many Oracle customers
have migrated their Oracle database licenses to the Oracle Exadata Database
46%Analyticaldatabase
49%Analyticalappliance
11%Hadoop
fgure 6.
Dp td
BaSED on 302 rESponDEntS,BI LeADeRshIp FoRuM, aprIl 2011.
executive
summary
research
background
he new analytical
ecosystem
big data trends
integrating
with hadoop
Future oF hadoop
recommendations
http://www.bileadership.com/http://www.bileadership.com/7/29/2019 BigDataReport Final
16/35
ExploItIng BIg Data: StratEgIES or IntEgratIng wItH HaDoop to DElIvEr BuSInESS InSIgHtS 16
BiG Data trenDs
Machine. Hadoop is also more likely to be used as a prototyping area than
anything else, which refects the act that most companies today are experi-
menting with Hadoop rather than putting it in production. Analytical databas-
es tend to be used as data marts, both dependent and independent, more thananalytical appliances. This refects their use as standalone systems to handle
unique workloads outside o the data warehouse (see fgure 7).
hadoop aTTracTing inTeresT
Although the analytical ecosystem depicted in fgure 5 (see page 12) portrays
Hadoop as a staging area or enterprise data warehouses and other down-
stream analytical systems, some big data and BI proessionals think Hadoop
may attract a growing percentage o analytical queries in the uture.Netfix, or example, uses Hadoop to stage all its operational data beore
AnalyticalDatabase AnalyticalAppliance Hadoop
57%
46%
79%
38%
38%
Stagingarea
Datawarehouse47%
45%
42%
16%
6%Dependentdatamart
Inde endent data mart
30%39%
31%
19%
Development/test
22%36%
44%Prototyping
fgure 7.
u c
BaSED on 302 rESponDEntS,BI LeADeRshIp FoRuM, aprIl 2011.
executive
summary
research
background
he new analytical
ecosystem
big data trends
integrating
with hadoop
Future oF hadoop
recommendations
http://www.bileadership.com/http://www.bileadership.com/http://www.bileadership.com/7/29/2019 BigDataReport Final
17/35
ExploItIng BIg Data: StratEgIES or IntEgratIng wItH HaDoop to DElIvEr BuSInESS InSIgHtS 17
BiG Data trenDs
aggregating and loading it into its Teradata data warehouse, where users can
access it using MicroStrategy and other SQL-based tools, according to Kurt
Brown, director o business intelligence platorms at the company. Business
analysts can also query the raw data in Hadoop using Java, MicroStrategy(through ree-orm HiveQL) or other languages i they need access to the
atomic-level data or have an urgent request or inormation that cant wait or
the data to be loaded into the data warehouse.
Almost all organizations today that have implemented Hadoop are using it
as a staging area (92%) to land and transorm data, presumably or loading
into a downstream data warehouse, where users can analyze it using standard
BI tools. They are also using it as an online archive (92%) so they never have
to delete data or summarize it and move it ofine where it is less accessible
(see fgure 8). In some cases, organizations move data rom their data ware-house into Hadoop or saekeeping instead o into an archival system.
Today In18Months92%
92%
92%
92%
StagingareaOnlinearchive
83%
58%
92%
67%
TransformationEngineAdhocqueries
42%
25%
58%
67%
67%
ScheduledreportsVisualexploration
83%
fgure 8.
Wh a F D Hdp spp y ogz?
BaSED on rESponDEntS wHo HavE ImplEmEntED HaDoop ( BI LeADeRshIp FoRuM, aprIl 2012).
executive
summary
research
background
he new analytical
ecosystem
big data trends
integrating
with hadoop
Future oF hadoop
recommendations
http://www.bileadership.com/http://www.bileadership.com/7/29/2019 BigDataReport Final
18/35
ExploItIng BIg Data: StratEgIES or IntEgratIng wItH HaDoop to DElIvEr BuSInESS InSIgHtS 18
BiG Data trenDs
Far ewer organizations are using Hadoop or running queries or reports or
exploring or mining data. But this will change in the next 18 months, accord-
ing to the survey. The percentage o companies that plan to use Hadoop or
ad hoc queries will jump rom 58% to 67%; reporting will climb rom 42% to67% and data mining rom 58% to 83%. But the biggest growth will come in
visual exploration, rom 25% to 67%, as companies use visualization tools
such as Tableau to explore data in Hadoop.
Hdp d. Organizations are using
Hadoop to store and process all kinds o data,
not just multi-structured data. In act, almost
all respondents are using Hadoop today (92%)
to store transaction data, and that gure willclimb to 100% in 18 months. In act, Netfix
lands all its operational data rom its consumer
websites and networking streaming devices into Hadoop. Very little opera-
tional data gets into Netfixs data warehouse beore rst passing through
Hadoop.
Ater transaction data, Hadoop is most oten used to store Web and sys-
tems logs (67% each), ollowed by social media and semi-structured data
(58% each). Some companies are using Hadoop to store documents (33%),
email (25%) and sensor data (17%), but ew are using it to store audio orvideo data yet.
Whats interesting is that these same companies plan to expand the types
o data they store in Hadoop over the next 18 months. The astest growing
data types stored in Hadoop among early adopters will be audio and video
data, sensor data and email (42% each) and documents (50%). More com-
panies also plan on capturing social media data in Hadoop (75%) as well as
Web logs (75%), transaction data (100%) and semi-structured data (67%)
(see fgure 9, page 19).
Despite some o the rhetoric coming out o the Hadoop community, earlyHadoop adopters see the technology as serving a complementary role to the
data warehouse. Most are using Hadoop to handle new workloads (66.7%)
that dont run in their data warehouse or they are ofoading existing work-
loads that are straining the current data warehouse (50%). Another third
(33%) use Hadoop to share existing data warehousing workloads, while
another quarter (25%) use it to share new workloads (see fgure 10, page 19).
Organizations are using
Hadoop to store and
process all kinds o data.
executive
summary
research
background
he new analytical
ecosystem
big data trends
integrating
with hadoop
Future oF hadoop
recommendations
7/29/2019 BigDataReport Final
19/35
ExploItIng BIg Data: StratEgIES or IntEgratIng wItH HaDoop to DElIvEr BuSInESS InSIgHtS 19
BiG Data trenDs
67%
58%
75%
67%
75%
e
ogs
Systemlogs
Socialmedia
92%
58%
100%
67%
Transactiondata
Semistructureddata
0%
25%
42%
42%
42%
Sensordata
Audioorvideo
33%50%
Documents
fgure 9.
Wh D D Hdp spp?
BaSED on rESponDEntS wHo HavE ImplEmEntED HaDoop,The BI LeADeRshIp FoRuM, aprIl 2012.
0%
50%
ReplacesitOffloadsexistingworkloads
67%
33%
HandlesnewworkloadsSharesexistingworkloads
25%
8%
SharesnewworkloadsDon'tknow
fgure 10.
ip Hdp y D Wh
BaSED on rESponDEntS wHo HavE ImplEmEntED HaDoop,The BI LeADeRshIp FoRuM, aprIl 2012.
executive
summary
research
background
he new analytical
ecosystem
big data trends
integrating
with hadoop
Future oF hadoop
recommendations
http://www.bileadership.com/http://www.bileadership.com/http://www.bileadership.com/http://www.bileadership.com/7/29/2019 BigDataReport Final
20/35
ExploItIng BIg Data: StratEgIES or IntEgratIng wItH HaDoop to DElIvEr BuSInESS InSIgHtS 20
BiG Data trenDs
non-implemenTers
Companies that have yet to implement Hadoop are airly bullish about their
plans or the technology. For us, Hadoop is a given, said Mike King, technical
ellow at FedEx Services, who develops database architectures or the ship-ping and logistics company. Were looking at big data with an eye towards
[implementing] a ull analytics stack. We see Hadoop as part o the puzzle
and believe its a good place to land raw data sources or analysis.
While FedEx has a robust data warehouse, it is used primarily to run report-
ing and dashboarding applications on structured data sets, according to King.
Hadoop will support new data sources, such as server log data, system peror-
mance data, Web clickstream data and reerence dataincluding addresses,
shipments, invoices and claims. And King expects people will analyze much o
this data in place, using MapReduce jobs to query the data.Like FedEx, 40% o companies plan to implement Hadoop within 12 months,
and another 22% will do so within two years, according to the survey. About a
third (30%) are not sure o their plans, and surprisingly, no one said they will
never implement Hadoop (see fgure 11).
Clearly, Hadoop has xed itsel in the current imaginations o BI proes-
sionals who dont want to get let out in the coldits one o the biggest new
fgure 11.
epd adp r Hdp b n-ip
40%
22%
Within12monthsWithin24months
5%
3%
Within36monthsIn3+years
30%
0%
NotsureNever
BaSED on 76 rESponDEntS wHo HavE not yEt ImplEmEntED HaDoop,The BI LeADeRshIp FoRuM, aprIl 2012.
executive
summary
research
background
he new analytical
ecosystem
big data trends
integrating
with hadoop
Future oF hadoop
recommendations
http://www.bileadership.com/http://www.bileadership.com/http://www.bileadership.com/http://www.bileadership.com/7/29/2019 BigDataReport Final
21/35
ExploItIng BIg Data: StratEgIES or IntEgratIng wItH HaDoop to DElIvEr BuSInESS InSIgHtS 21
BiG Data trenDs
developments in data management. (The last big trend was the advent o ana-
lytical appliances and platorms, which are still in the early adoption phase.)
Hdp . Companies that have yet to implement Hadoop havea dierent idea about how theyll use the new technology. The largest per-
centage o potential Hadoop users plans to use it or data mining (57%), ol-
lowed by ad hoc queries (45%). A smaller percentage sees Hadoop as a stag-
ing area (37%), online archive (23%) or transormation engine (39%), which
diers sharply rom organizations that have already implemented Hadoop
(see fgure 12).
Much o the disconnect between current and uture adopters o Hadoop
comes rom the hype emanating out o the big data community, which gener-
ally portrays Hadoop as an analytical system or multi-structured data insteado a staging and pre-processing area. Analytics is the promise o Hadoop, but
this wont happen until there is a viable market o easy-to-use graphical tools
or accessing and analyzing Hadoop data. The BI industry has spent the last
20 years developing such tools or SQL databases, and its only now getting
the ormula right. Fortunately, Hadoop vendors will be able to piggyback on
the lessons learned by BI vendors, i not the actual tools and techniques them-
fgure 12.
epd u Hdp b n-ip
37%
23%
39%
StagingareaOnlinearchive
TransformationEngine45%
5%
27%
AdhocqueriesScheduledreportsVisualexploration
57%
23%
5%
DataminingNotsure
OtherBaSED on 76 rESponDEntS wHo HavE not yEt ImplEmEntED HaDoop (tHE BI LeADeRshIp FoRuM, aprIl 2012,www.BIlEaDErSHIp.com).
executive
summary
research
background
he new analytical
ecosystem
big data trends
integrating
with hadoop
Future oF hadoop
recommendations
http://www.bileadership.com/http://www.bileadership.com/http://www.bileadership.com/7/29/2019 BigDataReport Final
22/35
ExploItIng BIg Data: StratEgIES or IntEgratIng wItH HaDoop to DElIvEr BuSInESS InSIgHtS 22
selves (see Integrating with Hadoop, page 25).
Hdp d. Non-implementers also share a similar prole as Hadoop
implementers when it comes to the kind o data they expect to store inHadoop with one exception: Non-implementers are less bullish about storing
transaction data in Hadoop (see fgure 13). In act, only 47% said they plan to
store transaction data in Hadoop; thats compared with 100% o implement-
ers. For non-implementers, Hadoop is synonymous with non-structured data.
ip d wh. Non-implementers share much the same pro-
le in the way Hadoop will aect their existing data warehouses. First, no one
plans to replace a data warehouse with Hadoop (see fgure 11, page 20). The
largest percentage o respondents plans to use it to handle new workloads(58%), ollowed by sharing new workloads (37%), ofoad existing workloads
(27%) and share existing workloads (24%) (see fgure 14, page 23).
A majority o non-implementers view Hadoop as a means to analyze new
multi-structured data in its native orm or process that data and load it into
their data warehouses or analysis. One respondent wrote, I need more gory
processing o semi-structured and unstructured data in an MPP environ-
ment prior to getting that content linked up to our relational data [in the data
BiG Data trenDs
fgure 13.
D th n-ip W s Hdp
53%
33%
47%
Weblogs
Systemlogs
Socialmedia
44%50%
24%
8%
Transactiondata
Semistructureddata
Sensordata
Audioorvideo
18%
18%
11%
Documents
Notsure
executive
summary
research
background
he new analytical
ecosystem
big data trends
integrating
with hadoop
Future oF hadoop
recommendations
7/29/2019 BigDataReport Final
23/35
ExploItIng BIg Data: StratEgIES or IntEgratIng wItH HaDoop to DElIvEr BuSInESS InSIgHtS 23
warehouse.] Another said, Hadoop will augment our current business intel-
ligence and data warehousing environment with large unstructured data sets,
mostly social networking logs. It will in no way replace our existing data ware-
house and data marts or ETL environments.
hadoop hype
s p. For some, Hadoop is a curiosity that they eel obligated
to explore because o all the hype. Were looking at Hadoop to see what all
the excitement is about and whether it oers any benets or us, said John
Rome, deputy chie inormation ocer and BI strategist at Arizona State Uni-
versity. All we hear about is big data, big data, big data.
Rome said Hadoop might be an appropriate place to house the universitysvoluminous Web log, swipe card and possibly social media data as well as
data rom its learning management system, which generates our million
records daily. But he wonders whether Hadoop would be overkill or these
applications since its a good possibility IT could fow this data directly into
the universitys data warehouse. We look orward to applying these emerg-
ing tools to administrative tasks and gaining a better understanding o how we
BiG Data trenDs
fgure 14.
th ip Hdp D Wh n-p
0%
27%
ReplacesitOffloadsexistingworkloads
58%
24%
37%
HandlesnewworkloadsSharesexistingworkloadsSharesnewworkloads
12%
3%
Don'tknowDoesn'tapply
BaSED on 76 rESponDEntS wHo HavE not yEt ImplEmEntED HaDoop (tHE BI LeADeRshIp FoRuM, aprIl 2012,www.BIlEaDErSHIp.com).executive
summary
research
background
he new analytical
ecosystem
big data trends
integrating
with hadoop
Future oF hadoop
recommendations
http://www.bileadership.com/http://www.bileadership.com/http://www.bileadership.com/7/29/2019 BigDataReport Final
24/35
ExploItIng BIg Data: StratEgIES or IntEgratIng wItH HaDoop to DElIvEr BuSInESS InSIgHtS 24
can help those at ASU who are working on big data challenges and research
problems, he said.
A senior vice president o analytics at a major online publishing company
concurs with Rome. I am not convinced Hadoop is the right solution or theproblem they are trying to solve; however, our team is taking a look. My per-
sonal belie is that the team partially just wants to play with Hadoop because
it is a hot technology.
Nonetheless, both managers plan to evaluate Hadoop in the near uture.
Rome plans to carve out time this summer or all to evaluate Hadoop, while
the analytics manager is awaiting the results o his teams experimental usage
o Hadoop. n
BiG Data trenDs
executive
summary
research
background
he new analytical
ecosystem
big data trends
integrating
with hadoop
Future oF hadoop
recommendations
7/29/2019 BigDataReport Final
25/35
ExploItIng BIg Data: StratEgIES or IntEgratIng wItH HaDoop to DElIvEr BuSInESS InSIgHtS 25
igg wh Hdp
vendors of all sTripes are working to integrate their products with Hadoop so
users can do any kind o workload with any tools they want. In many respects,
there seems to be a race between members o the Apache Sotware Founda-
tion and established BI and data warehousing (DW) vendors to see who can
wrap Hadoop with the enterprise and analytical unctionality it currently lacks.Both sides are reaching out to the other to improve interoperability and inte-
gration between Hadoop and established BI and DW environments.
QuesT for inTeroperabiliTy
aph pj. The Apache Sotware Foundation has multiple
projects to address Hadoop shortcomings in the arena o reporting and ana-
lytics. One promising project is Hive, a higher-level data access language that
generates MapReduce jobs and is geared to BI developers and analysts. Hiveprovides SQL-like access to Hadoop, although it does nothing to overcome
its batch processing paradigm. Another is HBase, which overcomes Hadoops
latency issues but is designed or ast row-based reads and writes to support
high-perormance transactional applications. Both create table-like structures
on top o Hadoop les. Another high-level scripting language is Pig Latin,
which like Hive runs on MapReduce, but is better suited to building complex,
parallelized data fows, so its something to which ETL programmers might
gravitate.
Another emerging Apache project that addresses the lack o built-in meta-data in Hadoop is HCatalog, which takes Hives metadata and makes it more
broadly available across the Hadoop ecosystem, including Pig, MapReduce
and HBase. Soon SQL vendors will be able to query HCatalog with MapRe-
duce, Web and Java Database Connectivity (JDBC) interaces to determine
the structure o Hadoop data beore launching a query. Armed with this meta-
data, SQL products will be able to issue ederated queries against SQL and
executive
summary
research
background
he new analytical
ecosystem
big data trends
integrating
with hadoop
Future oF hadoop
recommendations
7/29/2019 BigDataReport Final
26/35
ExploItIng BIg Data: StratEgIES or IntEgratIng wItH HaDoop to DElIvEr BuSInESS InSIgHtS 26
inteGratinG WitH HaDooP
Hadoop data sources and return a unied result set without users having to
know where data lies or how to access it.
On the data-mining ront, Apache Mahout is a set o data-mining algo-
rithms or clustering, classication and batch-based collaborative ltering thatrun on MapReduce. Finally, Apache Sqoop is a project to accelerate bulk loads
between Hadoop and relational databases data ingestion, and Apache Flume
streams large volumes o log data rom multiple sources into Hadoop.
These projects are still in the early stages, although many leading-edge
adopters have implemented them and are providing ample eedback, i not
code, to the Apache community to make these sotware environments enter-
prise-ready. They are assisted by many commercial open source vendors that
contribute most o the code to Apache Hadoop projects and want to create a
rich, enterprise-ready ecosystem.
Bi d DW h. In addition, many BI and DW vendors have either
extended their tool sets to connect to or interoperate with Hadoop. For exam-
ple, many vendors now connect to Hadoop and can move data back and orth
seamlessly. Some even enable BI and ETL
developers to create and run MapReduce
jobs in Hadoop using amiliar graphical
development environments. Some startup
companies go a step urther and buildBI and ETL products that run natively on
Hadoop. Many database vendors are ship-
ping appliances that bundle relational data-
bases, Hadoop and other relevant sotware
in an attempt to oer customers the best o
both worlds.
Most o these Apache projects are led by
commercial open source vendors, typically
the Hadoop distribution vendors, whichstrive to enrich their distribution and make
it more enterprise-ready than the next one. By working under the Apache
Sotware Foundation umbrella, they accelerate the time to market o these
eatures and ensure that compatibility will be maintained throughout the
process.
Many BI and DW vendors
have either extended theirtool sets to connect to or
interoperate with Hadoop.
For example, many ven-
dors now connect to Ha-
doop and can move data
back and orth seamlessly.
executive
summary
research
background
he new analytical
ecosystem
big data trends
integrating
with hadoop
Future oF hadoop
recommendations
7/29/2019 BigDataReport Final
27/35
ExploItIng BIg Data: StratEgIES or IntEgratIng wItH HaDoop to DElIvEr BuSInESS InSIgHtS 27
inteGratinG WitH HaDooP
tp g. Since many organizations plan to add Hadoop to their
existing BI environments, its important to understand the degree to which BI
and DW products can interoperate with Hadoop. There are our types o inte-
gration, each with its own advantages and disadvantages:
c. Prebuilt data connectors enable developers to view and
move data in bulk between the two environments.
Hbd . These products blend SQL and MapReduce unctionality
into a single data processing environment.
n Hdp. These run natively on Hadoop but are accessed via a
graphical user interace that hides the complexities o the underlying pro-cessing environment.
ipb. Application programming interaces and metadata cata-
logs enable developers in one environment to design and dynamically
execute native jobs in another environment.
Type 1. connecTiviTy
A data connector enables developers to view and download data in a remotesystem. Since Hadoop stores data in a distributed le system, the data con-
nector can simply be a le system viewer that connects to Hadoops name
node, which contains a list o all les in the cluster. The connector also usually
executes Hadoop le system commands to retrieve specied les distributed
across the cluster and move them to the requesting system.
Many data connectors connect via an application programming interace
(API), like Open Database Connectivity (ODBC), and JDBC drivers, which link
client applications to relational databases. With an API approach, the con-
nector enables users to invoke unctions on the remote system, which is thesecond level o integration (see Type 4. Interoperablity, page 30). Most
vendors dont charge or connectors.
For example, ParAccel oers a bidirectional parallel connector or Hadoop
that enables developers to issue SQL calls that invoke appropriate prewritten
MapReduce unctions in Hadoop and transer les to ParAccel or urther pro-
cessing. Similarly, Oracle oers its Direct Connector or Hadoop Distributed
executive
summary
research
background
he new analytical
ecosystem
big data trends
integrating
with hadoop
Future oF hadoop
recommendations
7/29/2019 BigDataReport Final
28/35
ExploItIng BIg Data: StratEgIES or IntEgratIng wItH HaDoop to DElIvEr BuSInESS InSIgHtS 28
inteGratinG WitH HaDooP
File System, which queries HDFS data using SQL and loads it into the Oracle
Database or joins it with data already stored there.
P d . The benet o data connectors is that they are relativelyeasy to build and use. The downsides are that they require moving entire
data sets, which in a big data environment is not ideal. I the data volumes are
sizeable, the download may take a long time and create network bottlenecks.
Administrators who want to move data between Hadoop and relational data-
base management systems will use Sqoop or Flume or some proprietary bulk
load utility instead.
Type 2. hybrid sysTemsHybrid systems create a single environment that supports both Hadoop and
SQL data and unctions. There are no remote unction calls because all pro-
cessing happens locally. The single system runs both a relational database
and MapReduce and supports SQL and le system storage.
Hybrid systems vary considerably in architecture. Some embed relational
databases inside Hadoop, while others embed Hadoop and a relational data-
base inside a single data warehousing appliance.
For instance, the Teradata Aster MapReduce platorm embeds a MapRe-
duce engine inside a relational database. Administrators load structuredand multi-structured data into the platorm or data discovery. Teradatas
patented Aster Data SQL-MapReduce ramework enable users to invoke
and execute MapReduce jobs that run inside the Aster platorm using stan-
dard SQL and BI tools. As a result, SQL-MapReduce enables organizations to
store all their databoth structured and multi-structuredin a single place
and query it using standard SQL or a variety o programming languages with
MapReduce. This strategy eectively puts Hadoop and unstructured data into
the amiliar land o SQL-based processing.
Hadapt takes the opposite approach in its cloud-based product, which is setto launch this summer. Hadapt installs a Postgres relational database on every
node in a Hadoop cluster and then converts SQL calls to MapReduce, where
needed, to process both sets o data at once. Like Aster SQL-MapReduce,
Hadapt lets users store all their data in one place and use a single query para-
digm to access all data. These specialized environments are ideal or support-
ing applications, such as customer analytics, that require querying both struc-
executive
summary
research
background
he new analytical
ecosystem
big data trends
integrating
with hadoop
Future oF hadoop
recommendations
7/29/2019 BigDataReport Final
29/35
ExploItIng BIg Data: StratEgIES or IntEgratIng wItH HaDoop to DElIvEr BuSInESS InSIgHtS 29
inteGratinG WitH HaDooP
tured and unstructured data to get a complete picture o customer behavior.
P d . The benets o a hybrid system are straightorward: (1) It
creates a unied environment or all types o data, (2) you buy and supportone system instead o two, (3) you can issue a single query that runs against
any data type and (4) all processing runs locally, providing ast perormance.
The major downside o such systems is they are somewhat redundant i you
already manage an enterprise data warehouse or Hadoop environment or
both. As such, hybrid systems make good analytical sandboxes or discovery
platorms to handle new applications that require a synthesis o data types
and analytical techniques.
Type 3. naTive hadoop
Some vendors, including a bevy o startups and some established BI and DW
players, are aggressively courting the Hadoop market by building applications
that use MapReduce as the underlying execution engine.
As a new platorm, Hadoop needs to encourage developers to write appli-
cations that run on top o it. Some stalwart BI and data integration veterans
have taken up the challenge and have built new applications rom scratch or
are porting existing applications to run on Hadoop. In most cases, the new
applications provide a graphical interace that hides the underlying process-ing rom end users, who dont know where the application runs or the nature
o its execution engine. All they know is that they can query Hadoop data
using a point-and-click interace.
For example, startup Datameer oers a native Hadoop application or
creating dashboards. The tool builds a meta catalog o Hadoop data that it
imports using 20-plus connectors to a variety o sources, including most rela-
tional databases, NoSQL databases and social media eeds, such as Twitter.
Developers use the catalog to populate a Web-based spreadsheet with a sub-
set o data that they use to design the dashboard. Once theyre satised withthe design, they schedule a job to create the dashboard by running it against
the entire data set in Hadoop.
Some established BI and DW vendors are contemplating shipping native
Hadoop versions o their products. For instance, one leading data integration
vendor plans this year to ship a native Hadoop version o its ETL tool that will
convert existing mappings to run as MapReduce jobs on Hadoop. This capa-
executive
summary
research
background
he new analytical
ecosystem
big data trends
integrating
with hadoop
Future oF hadoop
recommendations
7/29/2019 BigDataReport Final
30/35
ExploItIng BIg Data: StratEgIES or IntEgratIng wItH HaDoop to DElIvEr BuSInESS InSIgHtS 30
inteGratinG WitH HaDooP
bility is part o a wider set o unctionality that pushes down processing to the
most ecient platorm in the BI and DW ecosystem.
P d . The benet o this native approach is that it provides seam-less access to Hadoop data and optimizes MapReduce unctions. Presumably,
the new tool would better stay abreast o new developments in the Hadoop
world and optimize MapReduce processing better than SQL vendors. The
downside is that Hadoop MapReduce is a batch environment that does not
support iterative queries, at least today. Users need to design the entire dash-
board beore they hit the execute button, since there is sizable latency when
running MapReduce jobs. Also, the application would need to make ODBC or
JDBC calls via MapReduce to access SQL data in relational databases when
joining this data with Hadoop data or use data services to retrieve this datadynamically.
Type 4. inTeroperabiliTy
As mentioned above, interoperability takes a data connector one step urther
and lets users execute native unctions on the remote system, oten using
a graphical interace that doesnt require users to write code. This is appli-
cation-level interoperability using custom or standard APIs and a metadata
catalog rather than data-level connectivity that links directly to a database orle system. Obviously, writing code to an API requires more time and testing
than does creating a data connector.
For example, Tableau uses Clouderas ODBC driver or Hive to query
Hadoop data directly. The driver shields users rom having to write HiveQL
statements. Since Hive does not support iterative queries, Tableau users can
view, connect and download the data they want to an in-memory cache that
users can query in an ad hoc manner. MicroStrategy also uses Clouderas
ODBC driver but goes urther by embedding Hive metadata into its semantic
layer (such as Projects) and Query Builder, dynamically generating HiveQLunder the covers. It also lets developers execute predened HiveQL and Pig
scripts and HiveQL using the ree-orm query capability.
Hortonworks, a startup that provides a sotware distribution o Apache
Hadoop along with training, service and support, establishes deep levels
o integration with established BI and DW products. Its developers, most
o whom helped build and manage Yahoos 42,000-node Hadoop cluster,
executive
summary
research
background
he new analytical
ecosystem
big data trends
integrating
with hadoop
Future oF hadoop
recommendations
7/29/2019 BigDataReport Final
31/35
ExploItIng BIg Data: StratEgIES or IntEgratIng wItH HaDoop to DElIvEr BuSInESS InSIgHtS 31
contribute the majority o code to Apache Hadoop. One o the key pieces o
Apache Hadoop that it is building is HCatalog, a SQL-like metadata catalog
accessible via Hive, MapReduce and various REST-based APIs. Talend, an
open source ETL vendor, already uses HCatalog to design MapReduce work-fows and transormations. HCatalog is likely to become Hadoops de acto
metadata repository and key piece in any interoperability ramework.
P d . In the case o Hadoop, the benet o an application-level
interace is signicant. Users o SQL-based tools can issue a query against
Hadoop and retrieve just the data they want without having to download a
much larger data set. In other words, they can use SQL to locate Hadoop data
and invoke a MapReduce job to process or query that data. In addition, rather
than execute an import-export unction, users query Hadoop on demand aspart o a single SQL query, which may also join the Hadoop data with other
data to ulll the SQL request. One downside: these APIs are not bidirection-
alin other words, a Hadoop developer would need to work with a SQL API
to query data rom a relational database as part o a MapReduce job. n
inteGratinG WitH HaDooP
executive
summary
research
background
he new analytical
ecosystem
big data trends
integrating
with hadoop
Future oF hadoop
recommendations
7/29/2019 BigDataReport Final
32/35
ExploItIng BIg Data: StratEgIES or IntEgratIng wItH HaDoop to DElIvEr BuSInESS InSIgHtS 32
F Hdp
cooperaTion or compeTiTion? Vendors have been quick to jump on the big
data bandwagon largely because it opens up a new market or their database,
data integration and BI products but also because Hadoop represents a poten-
tial threat. Established sotware vendors stand to lose signicant revenue i
Hadoop evolves without them or gains robust data management and analyti-cal unctionality that reduces the appeal o their existing products. Most are
hedging their bets. They are working to interoperate with Hadoop, in case it
becomes a major new data source, and ensure it remains subservient to their
existing products.
Both sides are eager to partner and work together. Hadoop vendors benet
as more applications run on Hadoop, including BI, ETL and relational database
management system (RDBMS) products. And commercial vendors benet i
their existing tools have a new source o data to connect to and plumb. Its a
big new market whose honey attracts a hive ull o bees.Some customers are already asking whether data warehouses and BI tools
will eventually be olded into Hadoop environments or the reverse. As one
survey respondent put it, We are still evaluating between two schools o
thought: pull rom all sources including Hadoop clusters into the RDBMS or
pull rom all, including the RDBMS, into the Hadoop cluster.
Its no one wonder BI proessionals are conused. In the past year, they
have been bombarded with big data hype, some o which makes it sounds like
the days o the data warehouse and relational database are numbered. They
think, Why spend millions o dollars on a new analytical RDBMS i we cando that processing without paying a dime in license costs using Hadoop? Or
Why spend hundreds o thousands o dollars on data integration tools i our
data scientists can turn Hadoop into a huge data staging and transormation
layer? or Why invest in traditional BI and reporting tools i our power users
can exploit Hadoop using reely available programs, such as Java, Python, Pig,
Hive or Hbase?
executive
summary
research
background
he new analytical
ecosystem
big data trends
integrating
with hadoop
Future oF hadoop
recommendations
7/29/2019 BigDataReport Final
33/35
ExploItIng BIg Data: StratEgIES or IntEgratIng wItH HaDoop to DElIvEr BuSInESS InSIgHtS 33
Given the hype it shouldnt be surprising that some user organizations are
opting or Hadoop when a NoSQL database or more traditional BI solution
would suce. For example, several customers o Splunk, a NoSQL vendor that
applies search technology to log data, planned to implement Hadoop untilthey came across Splunk. Even though Splunk is commercial sotware, the
customers ound the sotware much easier to use than Hadoop, making its
total cost o ownership much lower. Also, a customer that has large volumes
o XML data was considering Hadoop but recently discovered MarkLogic,
another NoSQL database vendor that uses an XML schema to process data. It
is now investigating whether MarkLogic is a better t than Hadoop.
Right now, its too early to divine the uture o the big data movement and
name winners and losers. Its possible that in the uture all data management
and analysis will run entirely on open source platorms and tools. But its justas likely that commercial vendors will co-opt (or outright buy) open source
products and unctionality and use them as pipelines to magniy sales o their
commercial products. Since Hadoop vendors are venture-backed, its likely
well discover the answer soon enough as investors get restless and push or
a sale. In this case, a commercial vendor will gain access to the major Hadoop
distributions and more than likely pursue a hybrid strategy.
More than likely, well get a mlange o open source and commercial capa-
bilities that interoperate to create the big data ecosystem as described in
section one o this report. Ater all, 30 years ater the mainrame revolution,mainrames are still an integral component o many corporate data process-
ing architectures. And multidimensional databases didnt replace relational
databases as the building blocks o data warehousing environments; rather,
multidimensional databases became an external data mart or cubing technol-
ogy geared to intensive, multidimensional analytical workloads. Our corporate
computing environments have an amazing ability to ingest new technologies
and nd an appropriate niche or each within the whole. In IT, nothing ever
dies; it just nds its niche in an evolutionary ecosystem.
Accordingly, Hadoop will most likely serve as a staging area and pre-pro-cessing environment or multi-structured data beore it is loaded into ana-
lytical databases. Like any staging area, some users with requisite skills and
permission will query and report o this layer, but the majority o users will
wait until the data is saely integrated, aggregated, cleansed and loaded into
high-perormance analytical databases, where it can be queried and analyzed
with SQL-based tools.
Future oF HaDooP
executive
summary
research
background
he new analytical
ecosystem
big data trends
integrating
with hadoop
Future oF hadoop
recommendations
7/29/2019 BigDataReport Final
34/35
ExploItIng BIg Data: StratEgIES or IntEgratIng wItH HaDoop to DElIvEr BuSInESS InSIgHtS 34
rd
1.ip db. I you havent already, its time to
implement an analytical database, either to augment or replace your data
warehousing engine or create an analytical sandbox or discovery platorm or
business analysts or both. As part o the process, identiy a critical applica-
tion or the new system that can kick-start adoption and build support amongexecutives to expand the analytics program.
2.r -d d. Identiy multi-structured data in yourorganization and how harvesting it might provide business value. Talkto business-side people and ask what they would do i they could query this
multi-structured data and what it would be worth to them.
3.ig Hdp. Understand the pros and cons o using Hadoop to
manage your multi-structured data versus other approaches. Identiya critical application or Hadoop, such as improving the accuracy o raud or
cross-sell models, to justiy the commitment o time and money and keep it
rom being labeled a science experiment.
4.ig hw g wh Hdp. Understand the our types ointegration with Hadoop rom SQL environments. Identiy which levelsyou need or the various Hadoop applications you have identied.
5.udd d bw Hdp d db.When you should process data in Hadoop versus new analytical data-
bases is not always clear. Obviously, Hadoop is geared more toward unstruc-
tured data because its based on a le system, while analytical databases are
geared more toward structured data. But that doesnt mean you cant do any
kind o processing in either environment. It simply depends on data volumes,
internal expertise and the nature o the consuming application. n
executive
summary
research
background
he new analytical
ecosystem
big data trends
integrating
with hadoop
Future oF hadoop
recommendations
7/29/2019 BigDataReport Final
35/35
aBout tHe autHor
w ek hs bee hh
ede i he d ehsi, bsi-
ess ieiee (BI) d ee
ee feds sie 1995. He hs
ded es i-deh esehsdies d is he h he bes-
sei bk prfrmanc Dabard:
Maring, Mnitring, and Managing Yr Bin. He is
ed kee seke d be d he ss d
ds kshs bsiess is, ee
dshbds d BI, he is. es,
Ekes seed s die edi d eseh
the D wehsi Isie, hee he es he
s e d ii s d hied is
BI Eeie Si.
Ekes is die eseh tehte, hee
he ies eek b ed wes wd,hih ses ids eds d eies bes -
ies i he ii BI (see www.b-eye-network.
com/blogs/eckerson). Ekes is s eside BI
lede csi (www.bileader.com) d de BI
ledeshi (www.bileadership.com), ek
BI dies h ee e ehe ides b
bes ies i BI d ede he e BI i.
He be ehed [email protected].
executive
summary
research
background
he new analytical
ecosystem
big data trends
integrating
with hadoop
Future oF hadoop
recommendations
exliting Big Data: stratgi fr
Intgrating wit had t Dlivr
Bin Inigt is
BeyeNETWORK
e-bii.
h st
Edii Die
J smi Edi, E-pbiis
J s
Edi i chie
l K
Die oie Desi
mk b
pbishe
e lt
Die Ses
2012 tehte I. n his b-ii be sied eded
i b es ih ieeissi he bishe. tehte e-is e ibe hh The YGS Group.
at TTt: tehte bishesedi ii eh es-sis. me h 100 sed ebsies
ebe qik ess dee se es,die d sis b he ehies,ds d esses i jb.o ie d i ees ie die
ess ideede ee ed die. a It Kede Ehe, si i, e die d
she sis ih ees d ees.
http://www.b-eye-network.com/blogs/eckersonhttp://www.b-eye-network.com/blogs/eckersonhttp://www.bileader.com/http://www.bileadership.com/mailto:[email protected]://www.b-eye-network.com/http://searchmanufacturingerp.techtarget.com/http://searchbusinessanalytics.techtarget.com/mailto:mbolduc%40techtarget.com%20?subject=http://reprints.ygsgroup.com/m/techtargethttp://reprints.ygsgroup.com/m/techtargetmailto:mbolduc%40techtarget.com%20?subject=http://searchbusinessanalytics.techtarget.com/http://searchmanufacturingerp.techtarget.com/http://www.b-eye-network.com/mailto:[email protected]://www.bileadership.com/http://www.bileader.com/http://www.b-eye-network.com/blogs/eckersonhttp://www.b-eye-network.com/blogs/eckerson