forrester dawning of age of bi dbms

18
Making Leaders Successful Every Day May 27, 2011 It’s The Dawning Of The Age Of BI DBMS by Boris Evelson for Business Process Professionals

Upload: shweta-prashad

Post on 22-Oct-2014

120 views

Category:

Documents


4 download

TRANSCRIPT

Page 1: Forrester Dawning of Age of Bi Dbms

Making Leaders Successful Every Day

May 27, 2011

It’s The Dawning Of The Age Of BI DBMSby Boris Evelsonfor Business Process Professionals

Page 2: Forrester Dawning of Age of Bi Dbms

© 2011 Forrester Research, Inc. All rights reserved. Forrester, Forrester Wave, RoleView, Technographics, TechRankings, and Total Economic Impact are trademarks of Forrester Research, Inc. All other trademarks are the property of their respective owners. Reproduction or sharing of this content in any form without prior written permission is strictly prohibited. To purchase reprints of this document, please email [email protected]. For additional reproduction and usage information, see Forrester’s Citation Policy located at www.forrester.com. Information is based on best available resources. Opinions reflect judgment at the time and are subject to change.

For Business Process Professionals

ExEcuTIvE SuMMAryWe know how to address the key business intelligence (BI) challenges of the past 20 years, such as stability, robustness, and rich functionality. Agility and flexibility challenges now represent BI’s next big opportunity. Business process professionals realize that earlier-generation BI technologies and architecture, while still useful for more stable BI applications, fall short in the ever-faster race of changing business requirements. Forrester recommends embracing Agile BI methodology, best practices, and technologies to tackle agility and flexibility opportunities. Alternative database management system (DBMS) engines architected specifically for Agile BI will emerge as one of the compelling Agile BI technologies business process pros should closely evaluate and consider for specific use cases.

TABlE OF cOnTEnTSFitting BI Into A RDBMS Is Like Putting A Square Peg Into A Round Hole

Specialized BI Databases Will come To The rescue

Learn To Navigate The Crowded Landscape Of Specialized BI DBMS

use columnar rDBMS To leverage And Turbocharge your Existing BI Applications

use In-Memory Index When nothing Else Is Fast Or Agile Enough

use Inverted Index For complex Data Structures And unstructured content

use Associative When you can’t Predict The Future But Have To Prepare For Anything

rEcOMMEnDATIOnS

Choose Wisely Between Mainstream And Specialized DBMSes For BI

WHAT IT MEAnS

BI-Specific DBMS Will Overshadow Traditional RDBMS In 10 Years

nOTES & rESOurcESForrester interviewed 18 vendors and 10 user companies for this research.

Related Research Documents“Forrester’s Business Intelligence DBMS Effort Estimation Model”May 27, 2011

“Trends 2011 And Beyond: Business Intelligence” March 31, 2011

“The Forrester Wave™: Enterprise Data Warehousing Platforms, Q1 2011”February 10, 2011

“Stay Alert To Database Technology Innovation”november 19, 2010

“Empower BI HErOes With Self-Service Tools”October 26, 2010

May 27, 2011

It’s The Dawning Of The Age Of BI DBMScomplement your rDBMS-Based BI With More Agile DBMS Technologiesby Boris Evelsonwith rob Karel, Brian Hopkins, James G. Kobielus, leslie Owens, and Allison caine

2

4

14

14

Page 3: Forrester Dawning of Age of Bi Dbms

© 2011, Forrester research, Inc. reproduction ProhibitedMay 27, 2011

It’s The Dawning Of The Age Of BI DBMS For Business Process Professionals

2

FITTINg BI INTO A RDBMS IS LIkE PuTTINg A SquARE PEg INTO A ROuND HOLE

Mainstream relational databases are an awkward fit for BI. You can use them, but it requires lots of tuning and customization and constant optimization — which is difficult, time-consuming, and costly. Unfortunately, row-based relational database management systems (RDBMSes) like IBM DB2, Microsoft SQL Server, Oracle, and Sybase ASE were originally designed and architected for transaction processing, not reporting and analysis.1 In order to tune such a RDBMS for BI usage, specifically data warehousing, architects usually:2

· Denormalize data models to optimize reporting and analysis. Data modelers usually model transactional data in what’s referred to as third normal form (3NF). 3NF best practice calls for storing each data entity and attribute in one place and then creating multiple lookup or decoding tables that map and translate keys into full item descriptions. For example, a customer record would only have a pointer, or a key, to a full address record stored in a different table. This works great when one needs to insert, update, or delete rows of data, as the amount of data for each transaction is limited. BI abhors such a structure, as it often creates very complex multiway joins of data that negatively affect query performance. To solve the problem, developers flatten or denormalize data models so that every row contains all of the attributes needed for a report. In the example above, each customer record would contain the customer address and become longer and flatter.

· Build indexes to optimize queries. Transactional processes, on the other hand, loathe many database indexes, such as pointers and data shortcuts, because inserting, updating, or deleting records will then require updating all relevant indexes, slowing down performance. The opposite holds true for BI — the more indexes the merrier! Additional indexes actually improve query performance.

· Build aggregate tables to optimize summary queries. Some SQL queries contain a “group by” clause to summarize information by time period, geographic region, line of business, or general ledger chart of accounts, for example. But calculating such summaries or aggregates at query time requires many processing cycles, especially if you’re managing millions or even billions of rows of detailed transactions. To solve this problem, data integration pros usually design a BI database schema to pre-aggregate data by the attributes they think will be used most often. ETL processes load these aggregated tables during off-peak times, so that subsequent reporting queries can instantaneously access the aggregated data without waiting for the database to pull it all together.

· Build OLAP cubes to further optimize analytic queries. Multidimensional online analytical processing (OLAP) databases, also known as cubes, represent another specialized approach to aggregate and index transactions by business-defined hierarchies like product level, region, and business unit. BI pros often use products like IBM Cognos PowerCubes, Microsoft Analysis Services, Oracle Essbase, SAS OLAP, and open-source Mondrian cubes on top of a RDBMS to further tune and optimize analytical queries.3

Page 4: Forrester Dawning of Age of Bi Dbms

© 2011, Forrester research, Inc. reproduction Prohibited May 27, 2011

It’s The Dawning Of The Age Of BI DBMS For Business Process Professionals

3

Unfortunately, there’s one basic problem with these approaches: It’s impossible to build denormalized data models, indexes, and aggregates for every possible query that users will execute during the lifetime of a database. So BI pros must pick their battles and optimize the RDBMS based on current or near-term expected usage. But in today’s fast-paced business environment, “near-term” may mean days or even hours, requiring BI and data warehouse (DW) pros to spend a significant amount of their time doing nothing more than constantly optimizing and reoptimizing these databases.4

Specialized BI Databases Will Come To The Rescue

Mainstream row-oriented RDBMS will not be the lead database technology for BI going forward. First, there’s the need to constantly redesign entity and attribute relationships in the logical data model and reoptimize indexes and aggregates in the physical data model. And these RDBMSes offer minimal support for:

· Unstructured content. Databases crave structure. Regardless of what many RDBMS vendors say, there’s no easy way to organize unstructured content from emails, documents, and web traffic unless you first extract and parse structured entities, like customers and products, and attributes like customer name and address. Data management pros must first organize unstructured content into well-defined data structures before adding it to a RDBMS — even one that has XML extensions — to handle hierarchical and semistructured data.

· Diverse data structures. RDBMS vendors and users have significant experience supporting financial information, with its relatively constant and easily defined data structures. For example, when organizing data for income statements and balance sheets, one can typically structure everything by account, time period, geographic region, and business or legal unit. But it’s much more difficult to organize data structures representing lower levels of detail for operational and product analysis — say, by product type — into a neat RDBMS model. Any manufacturer, distributor, or retailer who deals with thousands of product types — where each product has a unique set of attributes — can articulate these limitations. For example, a car manufacturer may use attributes like diameter, width, and seasonality to describe tires and attributes like manual, automatic, 4-speed, and 5-speed to describe transmissions. Attempting to fit diverse descriptions into the same attribute list results in ragged, sparse, and unbalanced product hierarchies — a nightmare for data architects to model and users to query.5

To address these challenges, both mainstream BI vendors and startups introduced databases designed specifically for BI reporting and analysis use cases. Some of these specialized technologies have been available for years, but Forrester still sees relatively low levels of adoption and buyer interest. 2009 was the first year in which we witnessed early adoption, but adoption levels actually dipped slightly in 2010. However, the number of IT execs and decision-makers saying that they plan to implement these new technologies soon grew slightly (see Figure 1).

Page 5: Forrester Dawning of Age of Bi Dbms

© 2011, Forrester research, Inc. reproduction ProhibitedMay 27, 2011

It’s The Dawning Of The Age Of BI DBMS For Business Process Professionals

4

Figure 1 Adoption Of And Interest In BI-Specific DBMS

Source: Forrester Research, Inc.58853

“What are your firm’s plans to adopt specialized database engines like OLAP,columnar databases, or warehouse appliances?”

Expanding/upgrading implementation

Implemented, not expanding

Planning to implement in the next 12 months

Planning to implement in a year or more

Interested but no plans

Not interested

Don’t know

20102009*

14%15%

16%18%

10%9%

11%8%

18%21%

25%22%

6%7%

Base: 784 IT executives and decision-makers in North America and Europe*Base: 853 IT executives and decision-makers in North America and Europe

Source: Forrsights Software Survey, Q4 2010*Source: Enterprise And SMB Software Survey, North America And Europe, Q4 2009

LEARN TO NAvIgATE THE CROWDED LANDSCAPE OF SPECIALIzED BI DBMS

Relatively low adoption of BI-specific DBMSes reflects the fact that these databases are very specialized and not meant to support a wide variety of use cases — unlike their bigger, older, more versatile, jack-of-all-trades RDBMS cousins. Additionally, not all BI-specialized DBMSes are created equal, and inconsistent or conflicting vendor marketing messages only add to customer confusion. Business process pros should consider four types of BI-specific DBMS: columnar RDBMS and in-memory, inverted, and associative index DBMS. The applicability of and use cases for each specialized DBMS vary greatly depending on multiple factors, including but not limited to database size, data structure complexity, and a requirement to organize and report on unstructured content (see Figure 2). Four other non-relational DBMSes are worth mentioning, but they’re too unique and differentiated for comparative review. The first two are NoSQL key value stores and RDF, semantic, or triplet stores — although Forrester still questions whether they’re really applicable to BI. The other two are graph DBMS and “big data” technologies like Hadoop (HDFS and HBase) and MapReduce.6

Page 6: Forrester Dawning of Age of Bi Dbms

© 2011, Forrester research, Inc. reproduction Prohibited May 27, 2011

It’s The Dawning Of The Age Of BI DBMS For Business Process Professionals

5

Figure 2 Four Types Of Specialized BI DBMS

Source: Forrester Research, Inc.58853

Quicklychanging

requirements

Slowlychanging

requirements

Similar data Disparate data

Relative scalability

High

Medium

Low

Agile BI

Traditional BI

ColumnarRDBMS

In-memory

Inverted/associativeindex

Best for:Columnar

RDBMSIn-memory

index*Inverted

index* Associative

Large volumes

Reusing existing BI infrastructure

Nonrelational

Polystructured multiple star schemasand/or sparse, ragged, unbalanced hierarchies

Unstructured content

Data-driven BI (versus schema-driven)

*These technologies often overlap signi�cantly. For example, all of the in-memory vendors use columnar, butnot relational, technology. Some of the inverted index vendors use a combination of in-memory (for caching)and columnar (for persistence) technologies.

Disparatedata

Full functionality Limited/partialfunctionality

Very limitedfunctionality

Nearly fullfunctionality

No capabilitiesin this area

Page 7: Forrester Dawning of Age of Bi Dbms

© 2011, Forrester research, Inc. reproduction ProhibitedMay 27, 2011

It’s The Dawning Of The Age Of BI DBMS For Business Process Professionals

6

use Columnar RDBMS To Leverage And Turbocharge Your Existing BI Applications

Columnar DBMS is the closest relative of traditional row-based RDBMS; it’s also relational and thus has most of the benefits of row-based RDBMS, such as the ability to work with all mainstream BI tools. Unlike other RDBMSes, however, it stores data in columns rather than rows (see Figure 3). Querying requires selecting, sorting, filtering and aggregating data by columns, so consider this specialized database type as innately tuned for BI. But just like all other specialized BI DBMSes, this comes at the expense of being less optimized for transaction processing.

A major healthcare company accelerated report generation from five days to a matter of seconds and increased the number of users from 200 to 2,000 by replacing its row-based RDBMS database with Sybase IQ columnar RDBMS. Row-based RDBMSes can also be tuned to improve performance, but this retuning needs to occur for every new application and new query — not a very agile approach. In a similar example, a leading business-to-business (B2B) and retail office products company uses ParAccel to augment its existing row-based DW, where it needs the extra BI performance and agility. In addition to its ability to leverage existing BI tools and apps, Forrester recommends using columnar RDBMS technology when:

· Data volumes are exceptionally large. Data stored in columns compresses better than data stored in rows because data values in a row, such as combinations of numbers, strings, and dates, often differ — whereas data values in most columns, like all dates, are similar. Additionally, loading raw data into a row-based RDBMS and tuning it for BI can increase the size of the raw data by a factor of three to seven after building indexes and aggregates. In a columnar RDBMS, a column already represents its own index; aggregates are calculated on the fly, which can keep the database size roughly equal to that of the raw data set — or sometimes even cut it in half.

· Business requirements call for agility and flexibility. Columnar DBMSes greatly reduce the need for physical modeling and query tuning, as there are fewer or no indexes and aggregates, and often eliminate the need to build multidimensional data sets like cubes. This means that BI developers can implement new business requirements much more quickly and with less effort.

In recent years, columnar RDBMS has greatly improved upon its inherent weakness: the ability to insert, change, and delete detailed transactions, usually in so-called microbatches. But even with these improvements, it still can’t match the innate transaction-processing strength of its row-based cousins. Hence, Forrester does not recommend basing a BI implementation on columnar RDBMS for very low-latency, real-time operational reporting like real-time algorithmic trading analytics.

Pure-play columnar RDBMS vendors on Forrester’s watchlist include Calpont InfiniDB, HP Vertica, Infobright, Kx Systems kdb+, ParAccel, PivotLink, Quiterian, SAND Technology, SAP Sybase IQ, SAP BW Accelerator and the new HANA appliances, SenSage, SiSense, and 1010data. Mainstream DBMS vendors and vendors with hybrid row-based and columnar DMBSes include EMC Greenplum, IBM Netezza, Ingres VectorWise, Oracle, and Teradata Aster Data. Microsoft’s PowerPivot product is also based on a columnar VertiPaq engine.7

Page 8: Forrester Dawning of Age of Bi Dbms

© 2011, Forrester research, Inc. reproduction Prohibited May 27, 2011

It’s The Dawning Of The Age Of BI DBMS For Business Process Professionals

7

Figure 3 Major Differences Between row-Based And columnar rDBMSes

Source: Forrester Research, Inc.58853

Need a specialindex to speed

up query bycolumn

Aggregation:scan all rows;

extract columnvalues or buildand maintain

an index;aggregate

Customer No. Name Address Balance

Customer No. Name Address Balance

Easy-to-manipulate rows

Values in row are di�erentand harder to compress

Row-based

RDBMS

Valu

es in

col

umns

are

sim

ilar

and

easi

er to

com

pres

s

Each

col

umn

is it

s ow

n in

dex

Diffi

cult-

to-m

anip

ulat

eco

lum

ns

Easy

-to-

man

ipul

ate

colu

mns

Aggr

egat

ion:

just

add

all

valu

es in

a c

olum

n

Difficult-to-manipulaterowsColumnar

RDBMS

1

2

3

John

Andre

David

Newton

Washington

San Diego

123

JohnAndreDavid

NewtonWashingtonSan Diego

$100

$200

$300

$100$200$300

BenefitChallenge

use In-Memory Index When Nothing Else Is Fast Or Agile Enough

An in-memory index DBMS can be described as a spreadsheet on steroids. It does precisely what spreadsheets were designed to do: flexible reporting and analysis.8 Spreadsheets do not require complex data models — just load rows and columns of raw data into them and analyze it by sorting, filtering, and pivoting. BI applications based on in-memory indexes can impose fixed models, but the data model, data, and resulting report are one and the same. So creating a report is the same process as modeling data and is known as inline modeling.

For example, a large global transportation company uses Microsoft PowerPivot to address the ongoing challenges the logistics industry faces, as unstable geopolitics and unpredictable atmospheric and geological events result in constantly changing and often unpredictable requirements. And

Page 9: Forrester Dawning of Age of Bi Dbms

© 2011, Forrester research, Inc. reproduction ProhibitedMay 27, 2011

It’s The Dawning Of The Age Of BI DBMS For Business Process Professionals

8

several major financial services institutions use Tibco Spotfire to make quick modifications to any dashboards that are not already preconfigured so they can see changes as they happen without requesting IT support. Finally, a global manufacturing company likes its QlikView product because of the agility with which it can access varied views of information — agility difficult to replicate in RDBMS-based BI applications. Additionally, when evaluating in-memory technologies, consider that:

· The best use cases support advanced agility and flexibility requirements. In-memory BI offers the most flexibility and agility of the specialized DBMS technologies reviewed here. Consider in-memory when RDBMS-based BI is too inflexible and spreadsheets don’t offer enough depth or breadth of BI functionality, like formatting, advanced analytics, and interactive data visualization.

· The functionality in-memory vendors offer varies greatly. The leading in-memory BI vendors differ significantly — for example, whether the product can manage data models too large to fit into a single memory space and whether it can update in-memory models row by row for low-latency, operational BI or whether it needs to reload or rebuild the entire model (see Figure 4). Business process pros have to ask questions like: Are the in-memory DBMS and the vendor GUI inseparable, or can other BI tools access the in-memory DBMS? Does the vendor offer integrated advanced analytics like statistical analysis and predictive modeling? Is the in-memory DBMS a true index that does not require a fixed OLAP model, or is it more of a traditional OLAP cube — with a star schema, facts, and dimensions — that just happens to be all loaded into memory, like with IBM Cognos TM1 and MicroStrategy? Finally, if business users are to use these tools, how much IT involvement is required to set up and maintain the applications?

· You may not be able to fit your entire data model into a single memory space. All in-memory vendors offer pretty good data compression, so most small to medium-size business and departmental BI applications should fit into a single memory space without issue. Even when deploying these applications on 64-bit servers, Forrester is not aware of any production applications that are larger than 200 GB to 300 GB. When sizing these applications, consider the size of the raw data set, compression ratios — look for at least 10:1 — and the number of concurrent users, as each active user may take up anywhere from a few dozen to a few hundred megabytes of RAM. There are also other vendor-specific factors, like data heterogeneity and sparseness, to take into account. If the total number is larger than a few hundred gigabytes, consider Tibco Spotfire, which can dynamically swap chunks of your model in and out of RAM, or in-memory/disk hybrid offerings from Endeca, SiSense, and Quiterian, because you’re probably not a good candidate for other in-memory vendor technologies.9

· Operational risk may increase. All of the power, flexibility, and agility that in-memory technology offers is not free of risk: If you don’t know what you’re doing, you may come up with the wrong answers. While existing RDBMS data models are indeed rigid, in most cases they do protect business users from creating queries and reports that may produce wrong results and lead to incorrect decisions. In-memory-based BI could be a free-for-all. Yes, IT can restrict how you

Page 10: Forrester Dawning of Age of Bi Dbms

© 2011, Forrester research, Inc. reproduction Prohibited May 27, 2011

It’s The Dawning Of The Age Of BI DBMS For Business Process Professionals

9

use or change in-memory models, but that would defeat the purpose and main advantages of in-memory BI. So dole out these applications carefully and create a set of guidelines and policies that spells out when to encourage the use of such agile but risky applications — and when not to. Additionally, follow Forrester’s recommendations on Agile BI; reduce the risk by making sure that no matter how business users use in-memory BI applications, IT is still in firm control of preparing data for those apps, including data integration, data cleansing, and data security.

Page 11: Forrester Dawning of Age of Bi Dbms

© 2011, Forrester research, Inc. reproduction ProhibitedMay 27, 2011

It’s The Dawning Of The Age Of BI DBMS For Business Process Professionals

10

Figure 4 Features Of Selected In-Memory Index vendors

Source: Forrester Research, Inc.58853

Advancedanalytics(statisticalanalysis, datamining,predictive)

Write-back(generatingnewtransactionsfor what-ifscenarios)

Row updatesand inserts(for lowlatency,operational BI)

Memoryswapping/paging (formodels thatare largerthan RAM)*

Feature

OLAPfunctionality(MDX queries)Associativeindex (analysiswith inlinemodeling)

Facetednavigation/search

Search SearchSearch

Other BItoolscan access it via MDX

viaEclipse

ODA

OEMKXEN via SPSS LimitedLimited Limited

In-memory index type

In-memory architecture

BI features

*While this is a nice-to-have option, using models that do not �t into a single memory space will signi�cantlyslow down analysis.

Actuate

BIRT Data

Objects

Advizor Solutio

ns

IBM Cognos TM1

Microsoft

PowerPivot

MicroStra

tegy

QlikView

SAP HANA

SAS JMP

TableauTibco

Spot�re

-

Page 12: Forrester Dawning of Age of Bi Dbms

© 2011, Forrester research, Inc. reproduction Prohibited May 27, 2011

It’s The Dawning Of The Age Of BI DBMS For Business Process Professionals

11

Figure 4 Features Of Selected In-Memory Index vendors (cont.)

Source: Forrester Research, Inc.58853

Other

Requiresspecializedhardware

O�ered asSaaS

SDK/APIs

Requires ITinvolvement

Feature Actuate

BIRT Data

Objects

Advizor Solutio

ns

IBM Cognos TM1

Microsoft

PowerPivot

MicroStra

tegy

QlikView

SAP HANA

SAS JMP

TableauTibco

Spot�re

use Inverted Index For Complex Data Structures And unstructured Content

Common RDBMS agility challenges involve the management of unstructured content. The inverted index BI DBMS approach questions the assumption that you must start with a database and then worry about tuning it by building indexes on top of it. This approach builds one big index, but instead of just pointing to data sources — as traditional search engines like Google or Yahoo do — it embeds data in the index itself. Typical use cases for inverted index DBMS include:

· Managing a variety of unharmonized data types and data sources. A Fortune 100 consumer packaged goods (CPG) company recently implemented Endeca to analyze heterogeneous data from multiple sources within a single view in order to determine the root causes of observed responsiveness to sales campaigns. The relevant data for this application came from different sources and spanned structured data, semistructured data, and unstructured content. An inverted index DBMS gave the CPG company the ability to see this complex span of data and content in a single interactive analytical view. This would have been much harder to achieve using row-oriented RDBMS, particularly because the firm included more challenging formats like unstructured content and unharmonized structured data and allowed the interactive discovery of relationships between these disparate collections.

· Combining SQL queries and keyword search. A global financial services firm recently implemented Attivio because the inflexibility of its traditional RDBMS would have required very costly and time-consuming data normalization processes to integrate numerous new data sources. The firm’s earlier-generation, RDBMS-based BI application could not handle the variety

Page 13: Forrester Dawning of Age of Bi Dbms

© 2011, Forrester research, Inc. reproduction ProhibitedMay 27, 2011

It’s The Dawning Of The Age Of BI DBMS For Business Process Professionals

12

of data types, including unstructured and semistructured content, that the firm needed to access, and did not support the desired range of query styles: SQL, keyword search, and the use of financial services-specific ontologies.

Forrester also recommends considering inverted index DBMS when you (see Figure 5):

· Build too many data marts or cubes. Earlier, we described examples from manufacturing and retail industries with complex data structures that often hold unbalanced, ragged hierarchies. To address the challenges that such complex data structures and hierarchies pose for traditional relational or even multidimensional DBMSes, IT often builds specialized data marts to flatten out hierarchies or programmatically fill in missing nodes. In some extreme cases, as many as one-third of all new reports require building specialized data marts or cubes. Such complex data structures are often more hierarchical than relational in nature. They’re easy to represent as XML and index in a single inverted index DBMS.

· Really need data-driven, not schema-driven, BI. Entirely schema-driven BI applications based on a relational data model hold the key to what one can analyze and report on. BI applications must be created specifically for the query at hand to navigate the data effectively. In contrast to the schema-bound data exploration and query refinement BI provides, an inverted index BI DBMS provides data-driven data exploration and query refinement. With a data-driven approach from ingest-through-the-index to presentation, a new source system column flows straight through to the application. In some vendor apps, such as Endeca’s, data-driven UI components even allow users to rearrange the data without requesting custom reports.

· Don’t know what you don’t know. All RDBMS- or OLAP-based BI applications assume that you need to know what you are looking for before you ask a question and construct a query. But BI end users often don’t! Faceted navigation — a key feature of all inverted DBMS BI applications — allows you to subtract what you know you don’t want. A good example is the faceted navigation on any eCommerce website, where you can eliminate all products but electronics, then all electronics but computers, and then all computers other than laptops in the $500 to $1,000 price range. Perhaps you didn’t know what you were looking for at the beginning of your search, but categories and prompts let you easily eliminate what you knew you didn’t want. In many business scenarios, it’s more obvious what you don’t want than what you do. In a similar example, one of Endeca’s retail customers makes its inside sales reps more productive by helping them eliminate all customers except for those who purchased products in the past six months, then all recent buyers except for those who bought part of a bundle, then all bundle buyers except for the ones for whom the rest of the bundle is now on sale.

Page 14: Forrester Dawning of Age of Bi Dbms

© 2011, Forrester research, Inc. reproduction Prohibited May 27, 2011

It’s The Dawning Of The Age Of BI DBMS For Business Process Professionals

13

Figure 5 Features Of Selected Inverted Index DBMS vendors

Source: Forrester Research, Inc.58853

Features*GUI

Access by third-party tools

*Exalead and Sinequa also claim similar capabilties.

AttivioCustom-built

SQL-based

Endeca

XQuery-basedPart of the package

use Associative When You Can’t Predict The Future But Have To Prepare For Anything

The future is truly unpredictable. Forrester often hears from DW planners that their business users want “everything” stored and tracked in their DW, just in case. That’s precisely what associative DBMS attempts to do. Imagine a DW that can store all-to-all relationships — associations or vectors — between every entity and every attribute in your domain, with counters, aggregates, and indexes for every intersection! A Global 100 manufacturing company uses Saffron Technology in a number of operational areas to leverage past engineering decisions and business experiences, such as component and part order optimization, failure root-cause analysis, and predictive maintenance. Other associative DBMS products include Ingres VectorWise, Illuminate Solutions iLuminate, Relavance Associate, and Lazysoft Sentences. A variation of an associative DBMS — an associative index, like Splunk — provides similar functionality by just pointing to data sources rather than physically pulling data into a DBMS.

The price of associative DBMS is its infamous X-factor: the multiplication factor used to calculate the size of a DBMS as a multiple of the raw data set. In many of the associative databases used in academia, this factor gets as high as 100, although Saffron claims to get it down to 30. Additionally, because associative DBMS structures do not resemble anything close to relational or multidimensional structures, they require a purpose-built GUI, and SQL- and MDX-based queries can’t easily access them. Associative DBMS does not think in terms of joins or “where” clauses, but rather in connections, analogies, classifications, and trends. Forrester is aware of one leading BI vendor releasing a version of its popular product to connect to Saffron via such REST APIs in the near future. More may follow.

Page 15: Forrester Dawning of Age of Bi Dbms

© 2011, Forrester research, Inc. reproduction ProhibitedMay 27, 2011

It’s The Dawning Of The Age Of BI DBMS For Business Process Professionals

14

r E c O M M E n D A T I O n S

CHOOSE WISELY BETWEEN MAINSTREAM AND SPECIALIzED DBMSES FOR BI

Forrester does not recommend deploying BI-specific DBMSes and specialized BI applications to replace traditional, broadly functional rDBMS and enterprise BI platforms. not just yet. rather, use them as complementary platforms and supplementary applications for all of the use cases highlighted in this report. In order to ensure the effectiveness of these complementary technologies and efforts, Forrester recommends that business process pros:

· Prioritize and emphasize data governance efforts. Data governance efforts are extremely important for BI. When and if you implement these additional complementary specialized BI applications, the data governance will be an order of magnitude more critical, as you will have to deal with multiple BI applications, multiple analytical databases, and less structured data usage.10

· Prioritize and emphasize common metadata efforts. until large-stack vendors like IBM, Microsoft, Oracle, and SAP acquire, build, and integrate all of the technologies we’ve described here, you are more than likely to end up using more than one BI tool. Therefore, business process pros should work with their enterprise architecture (EA) colleagues to create a single semantic layer using data federation technology — using infrastructure-as-a service (IaaS) products, for example — that can then serve as virtual data source to all, including specialized BI applications. Or create and maintain a common metadata repository.11 use these centralized metadata repositories to sync metadata between all BI platforms and applications. To ensure these metadata repositories integrate across competing BI applications, look for the vendors to OEM standard metadata adapters from Meta Integration Technology or adhere to metadata exchange standards like xMI.12

· Build a business case with tangible ROI. Forrester often finds that the most effective BI business cases — which are often also the most difficult to build and support — focus on top-line benefits.13 However, more pragmatic business cases usually involve calculating the cost of ownership, which mostly targets support and not just initial license costs. leverage Forrester’s BI DBMS Effort Estimation Model to help you understand the potential long-term cost-of-ownership savings from specialized BI DBMSes and tools.14

W H A T I T M E A n S

BI-SPECIFIC DBMS WILL OvERSHADOW TRADITIONAL RDBMS IN 10 YEARS

unlike claims by some of the vendors described in this report, we can’t predict the future and offer a crystal-ball vision into what percentage of BI applications will be based on these specialized DBMS technologies. But the tried and true 80/20 rule works pretty well most of the time. Therefore, we predict that within three years, 20% of all BI applications are going to be based on these specialized technologies. That fraction will then slowly but surely evolve to 80% over the next 10 years.

Page 16: Forrester Dawning of Age of Bi Dbms

© 2011, Forrester research, Inc. reproduction Prohibited May 27, 2011

It’s The Dawning Of The Age Of BI DBMS For Business Process Professionals

15

SuPPLEMENTAL MATERIAL

Companies Interviewed For This Document

Actuate

Advizor Solutions

Attivio

Endeca

IBM

Kanda Software

Microsoft

MicroStrategy

ParAccel

Pivotstream LLC

QlikView

Quiterian

Saffron Technology

SAP

SAS

SiSense

Splunk

Tableau Software

Tibco Software

You Get

ENDNOTES1 In Forrester’s 153-criteria evaluation of enterprise open source and closed source database management

systems (DBMSes), we rated the highest performers to determine which vendors offer the most mature, high-performance, scalable, secure, and flexible solutions. See the June 30, 2009, “The Forrester Wave™: Enterprise Database Management Systems, Q2 2009” report.

2 In Forrester’s 56-criteria evaluation of enterprise data warehousing (EDW) platform vendors, we found the EDW market increasingly competitive, as illustrated by tighter clustering of top vendors. See the February 10, 2011, “The Forrester Wave™: Enterprise Data Warehousing Platforms, Q1 2011” report.

3 Online analytical processing (OLAP) is a core component of a complex business intelligence (BI) architectural stack. Even as vendors begin to explore alternative technologies for slicing and dicing large data sets, traditional OLAP approaches will remain an essential component in most BI installations for at least the next few years. See the November 7, 2008, “OLAP: In Fashion Or Old-Fashioned?” report.

4 Forrester continues to see ever-increasing levels of interest in and adoption of business intelligence (BI) platforms, applications, and processes. There is one key über-trend and best practice: agility. BP pros should adopt Agile BI processes, technologies, and architectures to improve their chances of delivering successful BI initiatives. See the March 31, 2011, “Trends 2011 And Beyond: Business Intelligence” report.

5 In ragged or unbalanced hierarchies, the parent of at least one member of a dimension is not in the level immediately above the member and the branches of the hierarchies can descend to different levels. Additionally, in a ragged hierarchy, null values can appear in any of the level columns. Null column values between member names are skipped, so a parent can have a child member multiple levels below the parent level.

Page 17: Forrester Dawning of Age of Bi Dbms

© 2011, Forrester research, Inc. reproduction ProhibitedMay 27, 2011

It’s The Dawning Of The Age Of BI DBMS For Business Process Professionals

16

6 Growth in the volume and pace of business is driving new application and data requirements that are often well beyond the design limits of existing relational database technologies. Relational databases will ultimately innovate to deliver higher degrees of automation, improved performance, and scale as well as to support larger volumes of unstructured data. See the November 19, 2010, “Stay Alert To Database Technology Innovation” report.

7 Many of the in-memory index vendors mentioned in the in-memory index section are based on a variation of in-memory columnar DBMS/index. Microsoft recently announced that the Denali version of SQL Server will include VertiPaq as a new storage/index type.

8 In Forrester’s book Empowered, Forrester Research analysts describe ways that new technologies can empower business people and make them true HEROes through individual contributions to their respective company’s top and bottom lines. The book also points out how business intelligence (BI) is a key technology for HEROes, not only helping them to make sense of the mountains of data that they have to deal with, but actually allowing them to make better and faster decisions. See the October 26, 2010,

“Empower BI HEROes With Self Service Tools” report.

9 In the near future, in-memory technology will span multiple memory spaces and break through current limitation barriers. This is something that SAP claims its HANA in-memory BI appliance can already accomplish. Or perhaps other in-memory BI vendors will integrate their products with elastic cache technologies like Tibco ActiveSpaces to make random access of data across nodes feasible.

10 Business process and data management professionals recognize the business need and potential value effective data governance could deliver, yet data governance remains an immature and poorly understood competency within most organizations. Identify the most critical business processes and decisions that trusted data can optimize, define the policies, business rules and standards to ensure trusted data, and deliver business-value oriented metrics that will justify the resource investments. See the May 4, 2011,

“Data Governance Must Bridge Business Process, Policy, Architecture, And Value” report.

11 Some options include: standalone metadata repositories like ASG-Rochade; integrated data management metadata repositories from Ab Initio, IBM, and Informatica; and specialized BI metadata vendors like Kalido, WhereScape, BIReady, and BI-Insight.

12 Source: Meta Integration Technology (http://www.metaintegration.net).

13 As the demand for pervasive and comprehensive BI applications increases, the complexity and cost of large enterprise BI implementations are not easily reduced. Therefore it is now more critical than ever to demonstrate tangible value from BI by building bulletproof BI business cases. See the August 25, 2009, “The Business Case For BI: Now More Critical Than Ever” report.

14 See the May 27, 2011, “Forrester’s Business Intelligence DBMS Effort Estimation Model” report.

Page 18: Forrester Dawning of Age of Bi Dbms

Forrester Research, Inc. (Nasdaq: FORR)

is an independent research company

that provides pragmatic and forward-

thinking advice to global leaders in

business and technology. Forrester

works with professionals in 19 key roles

at major companies providing

proprietary research, customer insight,

consulting, events, and peer-to-peer

executive programs. For more than 27

years, Forrester has been making IT,

marketing, and technology industry

leaders successful every day. For more

information, visit www.forrester.com.

Headquarters

Forrester Research, Inc.

400 Technology Square

Cambridge, MA 02139 USA

Tel: +1 617.613.6000

Fax: +1 617.613.5000

Email: [email protected]

Nasdaq symbol: FORR

www.forrester.com

M a k i n g l e a d e r s S u c c e s s f u l E v e r y D a y

58853

For information on hard-copy or electronic reprints, please contact Client Support

at +1 866.367.7378, +1 617.613.5730, or [email protected].

We offer quantity discounts and special pricing for academic and nonprofit institutions.

For a complete list of worldwide locationsvisit www.forrester.com/about.

Research and Sales Offices

Forrester has research centers and sales offices in more than 27 cities

internationally, including Amsterdam; Cambridge, Mass.; Dallas; Dubai;

Foster City, Calif.; Frankfurt; London; Madrid; Sydney; Tel Aviv; and Toronto.