Académique Documents
Professionnel Documents
Culture Documents
Hadoop:
The goal is to design and build a data warehouse / business intelligence (BI) architecture that
provides a flexible, multi-faceted analytical ecosystem for each unique organization.
A traditional BI architecture has analytical processing first pass through a data warehouse.
In the new, modern BI architecture, data reaches users through a multiplicity of organization data
structures, each tailored to the type of content it contains and the type of user who wants to
consume it.
The data revolution (big and small data sets) provides significant improvements. New tools like
Hadoop allow organizations to cost-effectively consume and analyze large volumes of semi-
structured data. In addition, it complements traditional top-down data delivery methods with more
flexible, bottom-up approaches that promote predictive or exploration analytics and rapid
application development.
In the above diagram, the objects in blue represent traditional data architecture. Objects in pink
represent the new modern BI architecture, which includes Hadoop, NoSQL databases, high-
performance analytical engines (e.g. analytical appliances, MPP databases, in-memory databases),
and interactive, in-memory visualization tools.
Most source data now flows through Hadoop, which primarily acts as a staging area and online
archive. This is especially true for semi-structured data, such as log files and machine-generated
data, but also for some structured data that cannot be cost-effectively stored and processed in SQL
engines (e.g. call center records).
From Hadoop, data is fed into a data warehousing hub, which often distributes data to downstream
systems, such as data marts, operational data stores, and analytical sandboxes of various types,
where users can query the data using familiar SQL-based reporting and analysis tools.
Today, data scientists analyze raw data inside Hadoop by writing MapReduce programs in Java and
other languages. In the future, users will be able to query and process Hadoop data using familiar
SQL-based data integration and query tools.
The modern BI architecture can analyze large volumes and new sources of data and is a significantly
better platform for data alignment, consistency and flexible predictive analytics.
Thus, the new BI architecture provides a modern analytical ecosystem featuring both top-down and
bottom-up data flows that meet all requirements for reporting and analysis.
In the top-down world, source data is processed, refined, and stamped with a predefined data
structure--typically a dimensional model--and then consumed by casual users using SQL-based
reporting and analysis tools. In this domain, IT developers create data and semantic models so
business users can get answers to known questions and executives can track performance of
predefined metrics. Here, design precedes access. The top-down world also takes great pains to
align data along conformed dimensions and deliver clean, accurate data. The goal is to deliver a
consistent view of the business entities so users can spend their time making decisions instead of
arguing about the origins and validity of data artifacts.
Creating a uniform view of the business from heterogeneous sets of data is not easy. It takes time,
money, and patience, often more than most departmental heads and business analysts are willing to
tolerate. They often abandon the top-down world for the underworld of spreadmarts and data
shadow systems. Using whatever tools are readily available and cheap, these data hungry users
create their own views of the business. Eventually, they spend more time collecting and integrating
data than analyzing it, undermining their productivity and a consistent view of business information.
The bottom up world is a different process. Modern BI architecture creates an analytical ecosystem
that brings prodigal data users back into the fold. It allows an organization to perform true ad hoc
exploration (predictive or exploratory analytics) and promotes the rapid development of analytical
applications using in-memory departmental tools. In a bottom-up environment, users can't
anticipate the questions they will ask on a daily or weekly basis or the data they'll need to answer
those questions. Often, the data they need doesn't yet exist in the data warehouse.
The modern BI architecture creates analytical sandboxes that let power users explore corporate and
local data on their own terms. These sandboxes include Hadoop, virtual partitions inside a data
warehouse, and specialized analytical databases that offload data or analytical processing from the
data warehouse or handle new untapped sources of data, such as Web logs or machine data. The
new environment also gives department heads the ability to create and consume dashboards built
with in-memory visualization tools that point both to a corporate data warehouse and other
independent sources.
Combining top-down and bottom-up worlds is challenging but doable with determined commitment.
The prior article in this series discussed the human side of analytics. It explained
how companies need to have the right culture, people, and organization to succeed
with analytics. The flip side is the "hard stuff"- the architecture, platforms, tools, and
data--that makes analytics possible. Although analytical technology gets the
lionshare of attention in the trade press--perhaps more than it deserves for the value
it delivers--it nonetheless forms the bedrock of all analytical initiatives. This article
examines the architecture, platforms, tools, and data needed to deliver robust
analytical solutions.
Architecture
Staging Sandbox. This is a staging area for a data warehouse that contains raw,
non-integrated data from multiple source systems. Analysts generally prefer to query
a staging area that contains all the raw data than each source system individually.
Hadoop is a staging area for large volumes of unstructured data that a growing
number of companies are adding to their BI ecosystems.
In-memory BI sandbox. Some desktop BI tools maintain a local data store, either in
memory or on disk, to support interactive dashboards and queries. Analysts love
these types of sandboxes because they connect to virtually any data source and
enable analysts to model data, apply filters, and visually interact with the data
without IT intervention.
Analytical Platforms
Since the beginning of the data warehousing movement in the early 1990s,
organizations have used general-purpose data management systems to implement
data warehouses and, occasionally, multidimensional databases (i.e., "cubes") to
support subject-specific data marts, especially for financial analytics. General-
purpose data management systems were designed for transaction processing (i.e.,
rapid, secure, synchronized updates against small data sets) and only later modified
to handle analytical processing (i.e., complex queries against large data sets.) In
contrast, analytical platforms focus entirely on analytical processing at the expense
of transaction processing.
Moreover, many of these analytical platforms contain built-in analytical functions that
make life easier for business analysts. These functions range from fuzzy matching
algorithms and text analytics to data preparation and data mining functions. By
putting functions in the database, analysts no longer have to craft complex, custom
SQL or offboard data to analytical workstations, which limits the amount of data they
can analyze and model.
In 2010, vendors turned their attention to meeting the needs of power users after
ten years of enhancing reporting and dashboard solutions for casual users. As a
result, the number of analytical tools on the market has exploded.
Analytical tools come in all shapes and sizes. Analysts generally need one of every
type of tool. Just as you wouldn't hire a carpenter to build an addition to your house
with just one tool, you don't want to restrict an analyst to just one analytical tool. Like
a carpenter, an analyst needs a different tool for every type of job they do. For
instance, a typical analyst might need the following tools:
Excel to extract data from various sources, including local files, create reports, and
share them with others via a corporate portal or server (managed Excel).
BI Search tools to issue ad hoc queries against a BI tool's metadata.
Planning tools (including Excel) to create strategic and tactical plans, each
containing multiple scenarios.
Mashboards and ad hoc reporting tools to create ad hoc dashboards and reports
on behalf of departmental colleagues
Visual discovery tools to explore data in one or more sources of data and create
interactive dashboards on behalf of departmental colleagues
Multidimensional OLAP (MOLAP) tools to explore small and medium sets of data
dimensionally at the speed of thought and run complex dimensional calculations.
Relational OLAP tools to explore large sets of data dimensionally and run complex
calculations
Text analytics tools to parse text data and put it in a relational structure for analysis.
Data mining tools to create descriptive and predictive models.
Hadoop and MapReduce to process large volumes of unstructured and semi-
structured data in a parallel environment.
Figure 2 plots these tools on a graph where the x axis represents calculation
complexity and the y axis represents data volumes. Ad hoc analytical tools for casual
users (or more realistically super users) are clustered in the bottom left corner of the
graph, while ad hoc tools for power users are clustered slightly above and to the
right. Planning and scenario modeling tools cluster further to the right, offering
slightly more calculation complexity against small volumes of data. High-powered
analytical tools, which generally rely on machine learning algorithms and specialized
analytical databases, cluster in the upper right quadrant.
Data
Business analysts function like one-man IT shops. They must access, integrate,
clean and analyze data, and then present it to other users. Figure 2 depicts the
typical workflow of a business analyst. If an organization doesn't have a mature data
warehouse that contains cross-functional data at a granular level, they often spend
an inordinate amount of time sourcing, cleaning, and integrating data. (Steps 1 and 2
in the analyst workflow.) They then create a multiplicity of analytical silos (step 5)
when they publish data, much to the chagrin of the IT department.
In the absence of a data warehouse that contains all the data they need, business
analysts must function as one-man IT shops where they spend an inordinate amount
of time iterating between collecting, integrating, and analyzing data. They run into
trouble when they distribute their hand-crafted data sets broadly.
Data Warehouse. The most important way that organizations can improve the
productivity and effectiveness of business analysts is to maintain a robust data
warehousing environment that contains most of the data that analysts need to
perform their work . This can take many years. In a fast-moving market where the
company adds new products and features continuously, the data warehouse may
never catch up. But, nonetheless, it's important for organizations to continuously add
new subject areas to the data warehouse, otherwise business analysts have to
spend hours or days gathering and integrating this data themselves.
Atomic Data. The data warehouse also needs to house atomic data, or data at the
lowest level of transactional detail, not summary data. Analysts generally want the
raw data because they can repurpose in many different ways depending on the
nature of the business questions they're addressing. This is the reason that highly
skilled analysts like to access data directly from source systems or a data warehouse
staging area. At the same time, less skilled analysts appreciate the heavy lifting done
by the IT group to clean and integrate disparate data sets using common metrics,
dimensions, and attributes. This base level of data standardization expedites their
work .
Summary
The technical side of analytics is daunting. There are many moving parts that all
have to work synergistically together. However, the most important part of the
technical equation is the data. The old adage holds true: "garbage in, garbage out."
Analysts can't deliver accurate insights if they don't have access to good quality
data. And it's a waste of their time to spend days trying to prepare the data for
analysis. A good analytics program is built on a solid data warehousing foundation
that embeds analytical sandboxes tailored to the requirements of individual analysts.
Culture refers to the rules--both written and unwritten--for how things get done in an
organization. These rules emanate from two places: 1) the words and actions of top
executives and 2) organizational inertia and behavioral norms of middle
management and their subordinates (i.e., "the way we've always done it.") Analytics,
like any new information technology, requires executives and middle managers to
make conscious choices about how work gets done.
Executives. For advanced analytics to succeed, top executives must first establish a
fact-based decision making culture and then adhere to it themselves. Executives
must consciously change the way they make decisions. Rather than rely on gut feel
alone, executives must make decisions based on facts or intuition validated by data.
They must designate authorized data sources for decision making and establish
common metrics for measuring performance. They must also hold individuals
accountable for outcomes at all levels of the organization.
Executives also need to evangelize the value and importance of fact-based decision
making and the need for a performance-driven culture. They need to recruit like-
minded executives and continuously reinforce the message that the organization
"runs on data." Most importantly, they not only must "talk the talk," they must "walk
the walk." They need to hold themselves accountable for performance outcomes and
use certifiable information sources, not resort to their trusted analyst to deliver the
data view they desire. Executives who don't follow their own rules send a cultural
signal that this analytics fad will pass and so it's "business as usual."
Managers and Organizational Inertia. Mid-level managers often pose the biggest
obstacles to implementing new information technologies because their authority and
influence stems from their ability to control the flow of information, both up and down
organizational ladders. Mid-level managers have to buy into new ways of capturing
and using information for the program to succeed. If they don't, they, too, will send
the wrong signals to lower level workers. To overcome organizational inertia,
executives need to establish new incentives for mid-level managers and hold them
accountable for performance metrics aligned with strategic goals around the decision
making and the use of information.
It's impossible to do advanced analytics without analysts. That's obvious. But hiring
the right analysts and creating an environment for them to thrive is not easy.
Analysts are a rare breed. They are critical thinkers who need to understand a
business process inside and out and the data that supports it. They also must be
computer-literate and know how to use various data access, analysis, and
presentation tools to do their jobs . Compared to other employees, they are
generally more passionate about what they do, more committed to the success of
the organization, more curious about how things work, and more eager to tackle new
challenges.
But not all analysts do the same kind of work , and it's important to know the
differences. There are four major types of analysts:
Super Users. These are tech-savvy business users who gravitate to reporting and
analysis tools deployed by the business intelligence (BI) team. These analysts
quickly become the "go to" people in each department to get an ad hoc report or
dashboard, if you don't want to wait for the BI team. While super users don't normally
do advanced analytics, they play an important role because they offload ad hoc
reporting requirements from more skilled analysts.
Business Analysts. These are Excel jockeys that executives and managers answer
to create and evaluate plans, crunch numbers, and generally answer any question an
executive or manager might have that can't be addressed by a standard report or
dashboard. With training, they can also create analytical models.
Analytical Modelers. These analysts have formal training in statistics and a data
mining workbench, such as those from IBM (i.e., SPSS) or SAS. They build
descriptive and predictive models that are the heart and soul of advanced analytics.
Analytical modelers and data scientists require more training and skills. These
analysts generally have a background in statistics or number crunching. Statisticians
with business knowledge or social scientists with computer skills tend to excel in
these roles. Given advances in data mining workbenches, it's not critical that
analytical modelers know how to write SQL or code in C, as in the past. However,
data scientists aren't so lucky. Since Hadoop is an early stage technology, data
scientists need to know the basics of parallel processing and how to write Java and
other programs in MapReduce. As such, they are in high demand right now.
Business analysts play a key role in any advanced analytics initiative. Given the
skills required to build predictive models, analysts are not cheap to hire or easy to
retain. Thus, building the right analytical organization is key to attracting and
retaining skilled analysts.
Today, most analysts are hired by department heads (e.g., finance, marketing, sales,
or operations) and labor away in isolation at the departmental level. Unless given
enough new challenges and opportunities for advancement, analysts are easy
targets for recruiters.
Analytics Center of Excellence. The best way to attract and retain analysts is to
create an Analytics Center of Excellence. This is a corporate group that oversees
and manages all business analysts in an organization. The Center of Excellence
provides a sense of community among analysts and enables them to regularly
exchange ideas and knowledge. The Center also provides a career path for analysts
so they are less tempted to look elsewhere to advance their careers. Finally, the
Center pairs new analysts with veterans who can give them the mentoring and
training they need to excel in their new position.
The best way to accommodate these twin needs is by creating a matrixed analytical
team. Analysts should report directly to department heads and indirectly to a
corporate director of analytics or vice versa. In either case, the analyst should
physically reside in his assigned department most or all days of the week, while
participating in daily "stand up" meetings with other analysts so they can share ideas
and issues as well as regular off-site meetings to build camaraderie and develop
plans. The corporate director of analytics needs to work closely with department
heads to balance local and enterprise analytical requirements.
Summary
Advanced analytics is a technical discipline. Yet, some of the keys to its success
involve non-technical facets, such as culture, people, and organization. For an
analytics initiative to thrive in an organization, executives must create a fact-based
decision making culture, hire the right people, and create an analytics center of
excellence that attracts, trains, and retains skilled analysts.
Business intelligence is changing. I've argued in several reports that there is no longer just one
intelligence--i.e., business intelligence--but multiple intelligences, each supporting a unique
architecture, design framework, end-users, and tools. But all these intelligences are still designed to
help business users leverage information to make smarter decisions and support the creation of either
reporting or analysis applications.
2. Analytics Intelligence. Addresses the needs of "power users," providing ad hoc access to
any data inside or outside the enterprise to answer business questions that can't be identified
in advance using spreadsheets, desktop databases, OLAP tools, data mining tools and visual
analysis tools.
4. Content Intelligence. Gives business users the ability to analyze information contained in
documents, Web pages, email messages, social media sites and other unstructured content
using NoSQL and semantic technology.
You may wonder how all these intelligences fit together architecturally. They do, but it's not the clean,
neat architecture that you may have seen in data warehousing books of yore. Figure 1 below depicts
a generalized architecture that supports the four intelligences.
The top half of the diagram represents the classic top-down, data warehousing architecture that
primarily delivers interactive reports and dashboards to casual users (although the streaming/complex
event processing (CEP) engine is new.) The bottom half of the diagram adds new architectural
elements and data sources that better accommodate the needs of business analysts and data
scientists and make them full-fledged members of the corporate data environment.
A recent report I wrote describes the components of this architecture in some detail and provides
market research on the adoption of analytic platforms (e.g. DW appliances and columnar and MPP
databases), among other things. The report is titled: "Big Data Analytics: Profiling the Use of
Analytical Platforms in User Organizations." You can download it for free at Bitpipe by clicking on the
hyperlink in the previous sentence.
Since "Multiple Intelligences" framework and BI ecosystem that supports it represent what I think the
future holds for BI, I'd love to get your feedback.
The Next Wave in Big Data Analytics: Exploiting Multi-core Chips and SMP
Machines
As companies grapple with the gargantuan task of processing and analyzing "big data," certain technologies
have captured the industry limelight, namely massively parallel processing (MPP) databases, such as those from
Aster Data and Greenplum; data warehousing appliances, such as those from Teradata, Netezza, and Oracle;
and, most recently, Hadoop, an open source distributed file system that uses the MapReduce programming
model to process key-value data in parallel across large numbers of commodity servers.
SMP Machines. Missing in action from this list is the venerable symmetric multiprocessing (SMP) machine that
parallelizes operations across multiple CPUs (or cores) . The industry today seems to favor "scale out" parallel
processing approaches (where processes run across commodity servers) rather than "scale up" approaches
(where processes run on a single server.) However, with the advent of multi-core servers that today can pack
upwards of 48 cores in a single CPU, the traditional SMP approach is worth a second look for processing big
data analytics jobs .
The benefits of applying parallel processing within a single server versus multiple servers are obvious: reduced
processing complexity and a smaller server footprint. Why buy 40 servers when one will do? MPP systems
require more boxes, which require more space, cooling, and electricity. Also, distributing data across multiple
nodes chews up valuable processing time and overcoming node failures, which are more common when you
string together dozens, hundreds, or even thousands of servers into a single, coordinated system, adds to
overhead, reducing performance.
Multi-Core CPUs. Moreover, since chipmakers maxed out the processing frequency of individual CPUs in 2004,
the only way they can deliver improved performance is by packing more cores into a single chip. Chipmakers
started with two-core chips, then quad-cores, and now eight- and 16-core chips are becoming commonplace.
Unfortunately, few software programs that can benefit from parallelizing operations have been redesigned to
exploit the tremendous amount of power and memory available within multi-core servers. Big data analytics
applications are especially good candidates for thread-level parallel processing. As developers recognize the
untold power lurking within their commodity servers, I suspect next year that SMP processing will gain an
equivalent share of attention among big data analytic proselytizers.
Pervasive DataRush
One company that is on the forefront of exploiting multi-core chips for analytics is Pervasive Software, a $50
million software company that is best known for its Pervasive Integration ETL software (which it acquired from
Data Junction) and Pervasive PSQL, its embedded database (a.k.a. Btrieve.)
In 2009, Pervasive released a new product, called Pervasive DataRush, a parallel dataflow platform designed to
accelerate performance for data preparation and analytics tasks. It fully leverages the parallel processing
capabilities of multi-core processors and SMP machines, making it unnecessary to implement clusters (or MPP
grids) to achieve suitable performance when processing and analyzing moderate to heavy volumes of data.
Sweet Spot. As a parallel data flow engine, Pervasive DataRush is often used today to power batch processing
jobs , and is particularly well suited to running data preparation tasks (e.g. sorting, deduplicating, aggregating,
cleansing, joining, loading, validating) and machine learning programs, such as fuzzy matching algorithms.
Today, DataRush will outperform Hadoop on complex processing jobs that address data volumes ranging from
500GB to tens of terabytes. Today, it is not geared to handling hundreds of terabytes to petabytes of data, which
is the territory for MPP systems and Hadoop. However, as chipmakers continue to add more cores to chips and
when Pervasive releases DataRush 5.0 later this year which supports small clusters, DataRush's high-end
scalability will continue to increase.
Architecture. DataRush is not a database; it's a development environment and execution engine that runs in a
Java Virtual Machine. Its Eclipse-based development environment provides a library of parallel operators for
developers to create parallel dataflow programs. Although developers need to understand the basics of parallel
operations--such as when it makes sense to partition data and/or processes based on the nature of their
application-- DataRush handles all the underlying details of managing threads and processes across one or more
cores to maximize utilization and performance. As you add cores, DataRush automatically readjusts the
underlying parallelism without forcing the developer to recompile the application.
Versus Hadoop. To run DataRush, you feed the execution engine formatted flat files or database records and it
executes the various steps in the dataflow and spits out a data set. As such, it's more flexible than Hadoop, which
requires data to be structured as key-value pairs and partitioned across servers, and MapReduce, which forces
developers to use one type of programming model for executing programs. DataRush also doesn't have the
overhead of Hadoop, which requires each data element to be duplicated in multiple nodes for failover purposes
and requires lots of processing to support data movement and exchange across nodes. But like Hadoop, it's
focused on running predefined programs in batch jobs , not ad hoc queries.
Competitors. Perhaps the closest competitors to Pervasive DataRush are Ab Initio, a parallelizable ETL tool,
and Syncsort, a high-speed sorting engine. But these tools were developed before the advent of multi-core
processing and don't exploit it to the same degree as DataRush. Plus, DataRush is not focused just on back-end
processing, but can handle front-end analytic processing as well. Its data flow development environment and
engine are generic. DataRush actually makes a good complement to MPP databases, which often suffer from a
data loading bottleneck. When used as a transformation and loading engine, DataRush can achieve 2TB/ hour
throughput, according to company officials.
Despite all the current hype about MPP and scale-out architectures, it could be that scale-out architectures that
fully exploit multi-core chips and SMP machines will win the race for mainstream analytics computing. Although
you can't apply DataRush to existing analytic applications (you have to rewrite them), it will make a lot of sense to
employ it for most new big data analytics applications.
I recently spoke with James Phillips, co-founder and senior vice president of products, at Membase, an emerging
NoSQL provider that powers many highly visible Web applications, such as Zynga's Farmville and AOL's ad
targeting applications. James helped clarify for me the role of NoSQL in today's big data architectures.
Membase, like many of its NoSQL brethren, is an open source, key-value database. Membase was designed to
run on clusters of commodity servers so it could "solve transaction problems at scale," says Philips. Because of
its transactional focus, Membase is not technology that I would normally talk about in the business intelligence
(BI) sphere.
However, today the transaction community is grappling with many of the same technical challenges as the BI
community--namely, accessing and crunching large volumes of data in a fast, affordable way. Not coincidentally,
the transactional community is coming up with many of the same solutions--namely, distributing data and
processing across multiple nodes of commodity servers linked via high-speed interconnects. In other words, low-
cost parallel processing.
Key-Value Pairs. But the NoSQL community differs in one major way from a majority of analytics vendors
chasing large-scale parallel processing architectures: it relinquishes the relational framework in favor of key-value
pair data structures. For data-intensive, Web-based applications that must dish up data to millions of concurrent
online users in the blink of an eye, key-value pairs are a fast, flexible, and inexpensive approach . For example,
you just pair a cookie with its ID, slam it into a file with millions of other key-value pairs, and distribute the files
across multiple nodes in a cluster. A read works in reverse: the database finds the node with the right key-value
pair to fulfill an application request and sends it along.
The beauty of NoSQL, according to Philips, is that you don't have to put data into a table structure or use SQL
to manipulate it. "With NoSQL, you put the data in first and then figure out how to manipulate it," Phillips says.
"You can continue to change the kinds of data you store without having to change schemas or rebuild indexes
and aggregates." Thus, the NoSQL mantra is "store first, design later." This makes NoSQL systems highly
flexible but programmatically intensive since you have to build programs to access the data. But since most
NoSQL advocates are application developers (i.e. programmers), this model aligns with their strengths.
In contrast, most analytics-oriented database vendors and SQL-oriented BI professionals haven't given up on the
relational model, although they are pushing it to new heights to ensure adequate scalability and performance
when processing large volumes of data. Relational database vendors are embracing techniques, such as
columnar storage, storage-level intelligence, built-in analytics, hardware-software appliances, and, of course,
parallel processing across clusters of commodity servers. BI professionals are purchasing these purpose-built
analytical platforms to address performance and availability problems first and foremost and data scalability
issues secondarily. And that's where Hadoop comes in.
Hadoop. Hadoop is an open source analytics architecture for processing massively large volumes of structured
and unstructured data in a cost-effective manner. Like its NoSQL brethren, Hadoop abandons the relational
model in favor of a file-based, programmatic approach based on Java. And like Membase, Hadoop uses a
scale-out architecture that runs on commodity servers and requires no predefined schema or query language.
Many Internet companies today use Hadoop to ingest and pre-process large volumes of clickstream data which
are then fed to a data warehouse for reporting and analysis. (However, many companies are also starting to run
reports and queries directly against Hadoop.)
Membase has a strong partnership with Cloudera, one of the leading distributors of open source Hadoop
software. Membase wants to create bidirectional interfaces with Hadoop to easily move data between the two
systems.
Membase Technology
Membase's secret sauce--the thing that differentiates it from its NoSQL competitors, such as Cassandra,
MongoDB, CouchDB, and Redis--is that it incorporates Memcache, an open source, caching technology.
Memcache is used by many companies to provide reliable, ultra-fast performance for data-intensive Web
applications that dish out data to millions of current customers. Today, many customers manually integrate
Memcache with a relational database that stores cached data on disk to store transactions or activity for future
use.
Membase, on the other hand, does that integration upfront. It ties Memcache to a MySQL database which stores
transactions to disk in a secure, reliable, and highly performant way. Membase then keeps the cache populated
with working data that it pulls rapidly from disk in response to application requests. Because Membase distributes
data across a cluster of commodity servers, it offers blazingly fast and reliable read/write performance required
by the largest and most demanding Web applications.
Document Store. Membase will soon transform itself from a pure key-value database to a document store (a la
MongoDB.) This will give developers the ability to write functions that manipulate data inside data objects stored
in predefined formats (e.g. JSON, Avro, or Protocol Buffers.) Today, Membase can't "look inside" data objects to
query, insert, or append information that the objects contain; it largely just dumps object values into an
application.
Phillips said the purpose of the new document architecture is support predefined queries within transactional
applications. He made it clear that the goal isn't to support ad hoc queries or compete with analytics vendors:
"Our customers aren't asking for ad hoc queries or analytics; they just want super-fast performance for pre-
defined application queries."
The most optimal BI organization adopts a federated structure which blends a mix of
centralized and decentralized features. I discussed the basic features of a federated BI
organization in my last blog entry and the journey most organizations take to get there. (See
"Organizing the BICC Part I: Move to the Middle"). This blog discusses the division of
responsibilities between a corporate BI team and embedded BI teams in a federated BI
organization.
Organizational federation is not a new or unique concept. For instance, most countries have
a federated government. In the United States, federal, state, and municipal governments
share civic responsibility. The federal government funds and manages issues of national
concern, such as defense, immigration , and interstate commerce, while state and
municipal governments fund and oversee local functions, such as policing, transportation,
and trash disposal. However, there is a fuzzy boundary between jurisdictions, requiring
federal, state, and local officials to collaborate to meet the needs of citizens.
The same is true in the world of BI. Corporate and departmental BI teams focus on different
tasks but ultimately share responsibility for the delivery of BI solutions and must work in
concert to meet business needs. However, there is a reasonably clear division of
responsibility in two areas: data management and report development.
Data Management
The corporate BI team is chiefly responsible for managing data that is shared among two or
more business units. In other words, the corporate BI team builds and maintains an
enterprise data warehouse (EDW) and dependent data marts geared to individual
departments. The corporate BI team also facilitates data governance processes to create
standard definitions for commonly used metrics, dimensions, and hierarchies. It publishes
these definitions into a data dictionary shared by all business units and embeds them into a
business model within the enterprise BI tool.
Report Management
On the reporting side, departments assume the lionshare of responsibility for creating
departmental dashboards and scorecards. The BI Reporting Framework (see figure 2)
depicts this division of responsibility. The left half of the circle shows the reporting
responsibilities of the corporate BI team and right half shows the reporting responsibilities of
business units. These responsibilities are divided by the other axis, which represents "top
down BI" and "bottom up" BI.
Figure 2. BI Reporting Framework
Top Down and Bottom Up. In top-down BI, the corporate BI team delivers standard reports
and dashboards to casual users. The team gathers requirements, models and sources the
data, and then loads it into the data warehouse. Developers then build reports and
dashboards that dynamically query data warehouse data. In contrast, in bottom-up BI, power
users query the data warehouse directly, in an ad hoc fashion. They explore, analyze, and
integrate data from various systems and then create reports based on their findings, which
they often publish to departmental colleagues and executives. Basically, top-down BI
delivers standard reports to casual users, and bottom-up BI world enables power users to
explore and analyze data and create ad hoc reports.
Enterprise BI. Thus, the upper left quadrant in figure 2 represents the intersection of
enterprise BI and top-down BI. This is where corporate BI developers (i.e. enterprise) create
production reports and dashboards (i.e. top-down) that would be difficult for any single
division to create on their own. In the bottom left quadrant, corporate statisticians or data
scientists (i.e. enterprise) who are aligned with individual divisions but not collocated, explore
and query data in an ad hoc fashion to create predictive learning models.
Divisional BI. In the bottom-right quadrant, business unit analysts in each division use self-
service BI tools to analyze data and create ad hoc reports to display their insights. If
executives and managers want to continue seeing these reports on a regular basis, the
business unit analysts turn them over to the collocated BI professionals who convert them
into production reports (top-right quadrant.) This handoff between analysts and embedded BI
staff is critically important, but rarely happens in more organizations. Too often, business unit
analysts publish reports and then end up perpetually maintaining them. This is a job they
aren't trained or paid to do and keeps them from spending time on more value-added tasks,
like analyzing the business.
To succeed, there needs to be a bidirectional flow of information between each of the
sectors as depicted in figure 2.
Figure 2. Handoffs between BI Sectors
Going around the diagram, we can describe the handoffs that occur between each group in
a BI Center of Excellence:
Bridging the disparate worlds of top-down and bottom-up BI and enterprise and divisional BI
is not easy. It requires a lot of communication and collaboration. The structure that glues
together these BI sectors requires both matrixed reporting relationships and a multi-
layered governance body. This is the topic of my next blog.
POSTED OCTOBER 17, 2013 10:54 AM
PERMALINK | NO COMMENTS |
1. Customer Knowledge. BI teams can understand the types of power users in their
organization, better anticipate their needs, and tailor access to their requirements.
2. Departmental Deficits. Executives know which departments or business units lack
the power users required to support various types of analytical initiatives.
3. Self Knowledge. Business analysts understand where they stand in the spectrum of
analytical capabilities and it gives them a concrete set of objectives to pursue to
move up the ranks.
4. Training. Executives will be motivated to establish and fund training programs and
career paths to improve the analytical capabilities of the organization.
Self-Service BI Tips
One of the biggest paradoxes in business intelligence (BI) is that self-service BI requires a
lot of hand holding to succeed. That was the predominant sentiment voiced in a recent
Webcast panel I conducted with Laura Madsen, a healthcare BI consultant at Lancet,
Russell Lobban, director of BI and customer analytics at Build.com, and Brad Peters, CEO
and co-founder of the multi-faceted BI vendor, Birst. (The Webcast will soon air on
SearchBusinessAnalytics (date TBD.)
Training required. The panelists iterated the need for significant training and support to
ensure the success of a self-service BI initiative. Madsen said companies should implement
multi-modal training (i.e., Web, classroom, self-pace) on a continuous basis. Peters said
many of his customers are effectively using social media to grease self-service BI wheels;
specifically, discussion forums that enable users to share experiences and answer each
other's questions. Lobban said it's critical to offer an integrated data dictionary that defines
data elements used in the BI tool.
Know your audience. Another critical success factor is knowing the audience for self-
service BI. Peters, for example, said there are two types of self-service: "data self service"
for power users and "business self service" for casual users. Power users require ad hoc
access to data using visualization tools that enable them to explore application data and
local files. Casual users, on the other hand, need structured access to data via a semantic
layer that adds business context. Although a semantic layer requires time and effort to build,
all three panelists said it is critical to the success of any self-service BI program and helps
ensure that users make accurate decisions.
Lobban emphasized the importance of starting small and iterating quickly. He said users
often get discouraged when they don't find the data they need. Thus, it's critical for the BI
team to add new data quickly to keep up with user requirements. The other major challenge
Lobban sees with self-service BI is that business users often misinterpret the data in reports
and dashboards. This is especially true for new hires or transfers from other departments.
Thus, it's critical new hires and transfers get mentored by experienced BI users, either in
person or virtually via help desks, training classes, or online forums.
Which tools? The panel also spent a lot of time discussing the types of tools that are best
suited to self-service BI. Most thought the new generation of in-memory visualization tools
are great for power users who can navigate their way through existing databases and
applications, but inadequate for casual users who need more structured access to data. The
panel also discussed the benefits of traditional OLAP tools versus the new visualization
tools. The consensus is that OLAP tools still play an important role in BI tool portfolios
because they provide robust dimensional views and calculations that new visualization tools
don't support.
Finally, the panel discussed the tradeoffs between best of breed versus all-in-one BI suites.
Best of breed tools provide the best functionality available or satisfy the parochial needs of
individual workgroups or departments, while BI suites provide an integrated experience and
architecture that addresses the entire spectrum of BI needs in an organization and is thus
easier to administer. Ultimately, the panel agreed that each organization needs to decide
which approach best fits their individual requirements and culture.
Summary. Self-service BI is the holy grail for BI professionals, but it has been difficult to
achieve. BI practitioners and business users expect self-service BI to be easy when it's not.
It requires clean, comprehensive data, integrated metadata, and continuous training and
support. Ultimately, self-service BI is the true test of a BI program's overall maturity
Cognos