• Building the Data Warehouse:


  •   
  • FileName: ttbuild-1.pdf [read-online]
    • Abstract: Building the Data Warehouse:Getting StartedBYW. H. Inmon BUILDING THE DATA WAREHOUSE: GETTING STARTED“The journey of a thousand miles begins with a single step“old Chinese proverb

Download the ebook

Building the Data Warehouse:
Getting Started
BY
W. H. Inmon
BUILDING THE DATA WAREHOUSE: GETTING STARTED
"The journey of a thousand miles begins with a single step"
old Chinese proverb
"Perfection is spelled P-A-R-A-L-Y-S-I-S"
Winston Churchill
In any large undertaking, getting off the mark to a good start is the first step toward
success. And getting off to a good start is certainly at the heart of the success of the
data warehouse environment. Nothing delays success and destroys momentum in the
data warehouse environment more than undue deliberation at the outset. The hallmark
of success in the data warehouse environment is iterative analysis, development, and
processing, and it is impossible to do iterative processing without first starting.
The entire mindset of the classical operational system and application developer is
dominated by the need for perfection of design and the need to gather ALL requirements
before design and development ensues. Such an attitude may well be proper for the
operational application environment where processes are run repetitively, and where
requirements can be divined before a system is built. But the data warehouse
environment is one where many requirements CANNOT be discerned until the data
warehouse is built and the data in the warehouse is available for analysis. Simply stated,
only SOME of the informational requirements of a corporation can be known before the
data warehouse is built.
For this reason, the data warehouse is built iteratively, in small, fast bursts of
development. And the first small, fast burst of development is not achieved by massively
studying and gathering large amounts of requirements before the first of development
starts. It is imperative that the data warehouse developer have the attitude of moving
quickly through the steps of design and development, even though it is known that the
processing and data requirements are not complete at the outset of development.
The data warehouse environment is built in an entirely different way than the classical
operational environment has been built. The classical operational environment has been
built on the system development life cycle - the SDLC - that requires that requirements
be identified, that analysis proceed, followed by design and programming. Next testing
occurs, and is completed by implementation. The SDLC is fine where requirements can
be known in advance. But the DSS analyst - who is the ultimate beneficiary of the data
warehouse - does not and cannot know his/her requirements. DSS analysts'
requirements are discovered in a trial and error manner - in a mode of discovery. There
is an entirely different mode of development that is required for data warehouse
development.
PREREQUISITES - THE DATA MODEL
The starting point for the design and development of the data warehouse environment is
the data model. Without the data model, it is difficult to contemplate building the data
warehouse. The data model acts as the roadmap for development.
©Copyright 2000 by William H. Inmon, all rights reserved Page 1
BUILDING THE DATA WAREHOUSE: GETTING STARTED
The data warehouse data model is built from the corporate data model. The Tech Topic -
Creating The Data Warehouse Data Model From The Corporate Data Model - describes
this activity in depth and will not be repeated here. In addition to the corporate data
model being factored into the data warehouse data model, the known and obvious
informational requirements of the organization are also factored into the data warehouse
data model. Usually these requirements are gathered by means of a "time box". A time
box is a limitation of time - from one week to six months - predetermined such that all
known and obvious informational requirements are gathered in that period of time. Once
the deadline occurs, the informational requirements are entered into the data warehouse
data model.
The reason why time boxing is important is that informational requirements gathering
can continue for as much time as is allowed. If there is no finite limit to the amount of
time that is allowed, then the informational requirements gathering process continues
indefinitely and the first iteration of the data warehouse never gets built.
The data warehouse data model consists of at least two major components - a high level
model and a mid level model. Figure 1 graphically depicts the data warehouse data
model.
The high-level data model consists of the major subject areas (or "entities") of the
corporation. In addition the major relationships between the subject areas are also
included. The high-level data model is complete prior to the first iteration of data
warehouse development. The task of completing the high-level data model is not a
difficult or time consuming task. The mid level data model data model consists of keys
and attributes, and their grouping together. The mid level data model needs to be
complete only insofar as the mid level model must exist for those portions of the data
warehouse that will be built for the first iteration of design.
PREREQUISITES - TECHNOLOGY SELECTION
In order to build the first iteration of warehouse development it is necessary to have
selected the technology on which to build the data warehouse. Figure 2 shows that
technology must have been selected.
©Copyright 2000 by William H. Inmon, all rights reserved Page 2
BUILDING THE DATA WAREHOUSE: GETTING STARTED
The selection of data warehouse technology - both hardware and software - depends on
many factors, such as:
• the volume of data to be accommodated,
• the speed with which data is needed,
• the history of the organization,
• which level of data is being built,
• how many users there will be,
• what kind of analysis is to be performed,
• the cost of technology, etc.
Both hardware and software must be selected. The hardware is typically mainframe,
parallel, or client/server hardware. The software that must be selected is for the basic
data base manipulation of the data as it resides on the hardware. Typically the software
is either full function dbms or specialized data base software that has been optimized for
the data warehouse.
Other software that needs to be considered is the interface software that provides
transformation and metadata capability such as PRISM Solutions Warehouse Manager.
A final piece of software that is important is the software needed for changed data
capture, such as that provided by PRISM Solutions.
PREREQUISITES - SIZING THE DATA WAREHOUSE
Another prerequisite to the first iteration of data warehouse design and population being
performed is that of the sizing of the data warehouse, as depicted by Figure 3.
©Copyright 2000 by William H. Inmon, all rights reserved Page 3
BUILDING THE DATA WAREHOUSE: GETTING STARTED
Figure 3 shows that a rough sizing of data needs to be done to determine the fitness of
the hardware and software platforms. If the hardware and software platforms are either
much too big or are much too little for the amount of data that will reside in the data
warehouse, then no iterative development should occur until the fit is properly made.
If the hardware and dbms software are much too large for the data warehouse, the
costs of building and running the data warehouse will be exorbitant. Even though
performance will be no problem, development and operational costs and finances will be
a problem. Conversely, if the hardware and dbms software are much too small for the
size of the data warehouse, then performance of operations and the ultimate end user
satisfaction with the data warehouse will suffer. At the outset it is important that there
be a comfortable fit between the data warehouse and the hardware and dbms software
that will house and manipulate the warehouse.
In order to determine what the fit is like, the data needs to be sized. The sizing does not
need to be precise. If anything, the sizing needs to err on the size of being too large,
rather than too small. But a rough sizing of the data to be housed in the data warehouse
at this point can save much grief at a later point in time if in fact there is not a
comfortable fit between the data warehouse data and the environment it is built in. The
estimate of the data warehouse data should be in terms of order of magnitude.
Of course, one of the issues that relates to the volume of data in the data warehouse is
that of the level of granularity of data. If too much data is likely to be built into the data
warehouse, the level of granularity can always be adjusted so that the physical volume
of data is reduced.
PREREQUISITES - COLLECTING INFORMATIONAL REQUIREMENTS
The informational requirements of the organization need to be collected by means of a
time box, as previously discussed. Figure 4 shows the typical means by which those
informational requirements are identified and collected.
©Copyright 2000 by William H. Inmon, all rights reserved Page 4
BUILDING THE DATA WAREHOUSE: GETTING STARTED
Typically informational requirements are collected by looking at:
• reports. Existing reports can usually be gathered quickly and inexpensively. In most
cases the information displayed on these reports is easily discerned. But old reports
represent yesterday's requirements and the underlying calculation of information may
not be obvious ay all.
• spreadsheets. Spreadsheets are able to be easily gathered by asking the DSS analyst
community. Like standard reports, the information on spreadsheets is able to be
discerned easily. The problem with spreadsheets is that:
• they are very fluid. Important spreadsheets may have been created several months
ago that are not now available,
• they change with no documentation whatsoever,
• they may not be able to be easily gathered unless the analyst creating them wants
them to be gathered, and
• their structure and usage of data may be obtuse.
• other existing analysis. Through EIS and other channels there is usually quite a bit of
other useful information analysis that has been created by the organization. This
information is usually very unstructured and very informal (although in many cases it
is still very valuable information.)
• live interviews. Typically through interviews or JAD sessions, the end user can tell
what the informational needs of the organization are. Unfortunately JAD sessions
require an enormous amount of energy to conduct and assimilate. Furthermore, the
effectiveness of JAD sessions depends in no small part on the imagination and
spontaneity of the end user participating in the session.
In any case, gathering the obvious and easily accessed informational needs of the
organization should be done and should be factored into the data warehouse data model
prior to the development of the first iteration of the data warehouse.
THE FIRST ITERATION - HOW MUCH DATA TO LOAD?
The first issue of design and planning for the first iteration of the data warehouse to be
developed is exactly how much data is to be loaded and what variety of data is to be
loaded. It is almost never the case that huge amounts of data are loaded as a result of
the first iteration.
©Copyright 2000 by William H. Inmon, all rights reserved Page 5
BUILDING THE DATA WAREHOUSE: GETTING STARTED
A general guideline for the loading of the first iteration is that:
THE FIRST ITERATION SHOULD CONTAIN DATA THAT IS LARGE ENOUGH TO BE
MEANINGFUL AND SMALL ENOUGH TO BE QUICKLY DOABLE.
There are many different ways to cut down the size of data without losing its
effectiveness. Figure 5 illustrates a few of those ways.
There are practically speaking, an infinite number of ways to subset data. What is a
meaningful subset depends entirely on who the first user of the first iteration of the data
warehouse will be. The data architect needs to ask the question - who will be the first
user of the first iteration of data. Now, knowing whom the first user is, what data would
be meaningful to them?
There is a careful line to be walked here. On the one hand the data architect must take
care not to put too much data in the first iteration of development. On the other hand
the data architect must take care not to include so little data that the spontaneity of
discovery by the DSS analyst is limited.
THE FIRST ITERATION - FISHING IN THE RIGHT POND
There are many functional arenas that informational processing may bear fruit in. But
there are some classical functional areas that over the years have borne more fruit than
others. Figure 6 depicts those functional arenas.
©Copyright 2000 by William H. Inmon, all rights reserved Page 6
BUILDING THE DATA WAREHOUSE: GETTING STARTED
The classical functional arenas that data warehousing has proven to be effective in are:
• finance. Finance data tends to speak directly to the executive. It tends to get right to
the heart of the corporation. In addition, finance data is relatively smaller, in terms of
volume, than other types of data for most corporations. For these reasons, finance is
a popular arena in which to build the data warehouse,
• accounting. Accounting is similar in most regards to finance data,
• sales. Sales data is always important to the corporation. Sales data is like finance
data in that it typically is compact, compared to other data. (Of course, for some
industries this is not the case,)
• marketing data. Marketing and sales data are closely related and both are central to
the health and success of the corporation.
These then are the classical arenas where data warehouse has found fruitful ground.
But, depending on the industry or the circumstances of the company there are some
other fruitful grounds as well. Some of these other frequently fruitful grounds include:
• actuarial processing (for the insurance and insurance related industries),
• process control (for the manufacturing industry),
• human resources, and so forth.
THE FIRST ITERATION - SELECTING A SUBJECT AREA
Another important decision that needs to be made for the building of the first iteration of
the data warehouse environment is that of the selection of the first subject area to be
implemented. Figure 7 shows this selection.
©Copyright 2000 by William H. Inmon, all rights reserved Page 7
BUILDING THE DATA WAREHOUSE: GETTING STARTED
The high level data model is useful in the selection of the major subject area that will
entail the first iteration of data warehouse development. The first, best option is to
select an entire subject area for implementation. But on occasion the major subject area
that is to be implemented is too large. So the next best option is to select a subset of a
single subject area for implementation.
In some industries (such as banking and insurance) it often times is not possible to
select a single subject area for first iteration implementation. In these cases it is
necessary to select a series of subject for implementation. Figure 8 shows this option.
When it is necessary to select multiple subject areas for first iteration implementation,
only a subset of those subject areas should be implemented. The importance of the data
model becomes very apparent in this case as the data model shows what can be
implemented without overlap or conflict.
©Copyright 2000 by William H. Inmon, all rights reserved Page 8
BUILDING THE DATA WAREHOUSE: GETTING STARTED
The different options that apply to the building of the first iteration of the data
warehouse are:
• implement a single subject area,
• implement a subset of a single subject area, or
• implement a subset of multiple subject areas.
After this point the steps of design for the data warehouse will be described. It should be
recognized that these steps are done for very small amounts of data - there is nothing
sweeping or grand in design that is implied here. Typically there are from five to fifteen
physical tables that will result from the design efforts that are being described. If there
are more than that, then the whole point of doing iterative design and development has
been missed.
ADDING THE PHYSICAL ATTRIBUTES
After the subject area decision is made, the next task is to go from the data warehouse
data model into physical data base design. The first step in this direction is the
specification of physical attributes from the data model, as shown in Figure 9.
The specification of physical attributes includes the task of identifying keys and
attributes. In addition the physical specification for each key and each attribute is
created. At first glance the physical specification of the keys and attributes appears to
be an almost mechanical task; it is not.
©Copyright 2000 by William H. Inmon, all rights reserved Page 9
BUILDING THE DATA WAREHOUSE: GETTING STARTED
There is much consideration of:
• the end users understanding of the data,
• the format of the data as it resides in the legacy systems environment,
• the different legacy systems formats that must be consolidated,
• the volume of data that is to be created, and so forth.
After the physical characteristics of the data that will reside in the data warehouse is
determined, the next step is the identification of the unit of time that resides in each
unit of data warehouse data. In most cases the unit of time - year, quarter, month, day,
etc. - is appended to the key of the table. In a few cases, such as where full table
snapshots of data are taken monthly, the unit of time that identifies each occurrence of
data warehouse data is implicitly identified with the data warehouse.
Figure 10 shows that after the identification of the physical characteristics of data comes
the specification of the unit of time for each data warehouse table.
There is one consideration of design that is especially relevant and important here and
that is that the unit of time selected greatly influences the number of occurrences of
data. Selecting the unit of time of day implies that there will be many more records in
the data warehouse environment than if the unit of time of week were chosen, for
example. The design of the granularity of data is directly related to the specification of
the unit of time that resides in the data warehouse occurrences of data.
Once the unit of time has been specified, the next step in the creation of the first
iteration of the development of the data warehouse is to identify the system of record.
Figure 11 shows the identification of the system of record.
©Copyright 2000 by William H. Inmon, all rights reserved Page 10
BUILDING THE DATA WAREHOUSE: GETTING STARTED
Figure 11 shows that the system of record is the source data that resides in the
operational, legacy systems environment. The system of record is the data that is the:
• most complete,
• most accurate,
• most timely,
• has the best structural conformance to the data model, and
• is nearest to the point of operational entry.
The identification of the system of record is the most complex activity in the building of
the first iteration of data warehouse development. There are MANY circumstances that
must be accounted for in the specification of the system of record:
• multiple sources of data,
• no sources of data under certain cases,
• reformatting of data,
• summarization of data,
• conversion of data,
• recalculation of data,
• alteration of key structure,
• restructuring of data attributes,
• conditional merging of data, and so forth.
The data architect attempts to identify the system of record as accurately as possible.
But it is understood that if the system of record is incorrectly specified, that after the
first iteration of development is complete, that adjustments can (and will!) be made.
One of the aspects of specifying the system of record is that recognizing that there are
(at least!) two types of source of data - normal operational data bases and delta data
(i.e., "changed data capture"). Figure 12 illustrates these choices.
©Copyright 2000 by William H. Inmon, all rights reserved Page 11
BUILDING THE DATA WAREHOUSE: GETTING STARTED
Delta data is data that has been gleaned from the updates that have occurred to a file
since the last time a refreshment of the data warehouse was accomplished. Delta data is
used where there are very large files that undergo very little change from one
refreshment of the data warehouse to the next. Customer files are typically in this
category.
Once the transformations have been specified, they are transferred to the metadata
infrastructure that sits above the data warehouse. Figure 13 shows this transformation.
The metadata contains a record of what the source data is, what transformations and
summarizations have occurred, and what the description of the data warehouse looks
like. The metadata then becomes the DSS analyst guide to what is in the data
warehouse and how it got there.
Subsequent to the specification of the metadata from the transformations that occur is
the activity of specifying the frequency of transformation. Figure 14 shows this
specification.
©Copyright 2000 by William H. Inmon, all rights reserved Page 12
BUILDING THE DATA WAREHOUSE: GETTING STARTED
The frequency of execution of transformation depends on many things, such as:
• how much data will be transformed,
• whether a delta file is being used,
• the complexity of the transformation,
• the business need to see the updated data warehouse,
• the availability of the operational data,
• how frequently system checkpoints are being done,
• the amount of summarization and aggregation that needs to be done, and so forth.
At this point the DDL that does the actual definition of the data into the dbms is
executed, as seen in Figure 15.
The DDL of course specifies to the dbms the exact content and structure of data.
In some cases the DDL may have already been specified. If such is the case, then it is a
worthwhile activity to check that the DDL being generated by the first iteration of
processing is identical to the DDL that has already been produced. Figure 16 illustrates
this simple but very important check.
©Copyright 2000 by William H. Inmon, all rights reserved Page 13
BUILDING THE DATA WAREHOUSE: GETTING STARTED
If the DDL that the data architect has produced somehow does not match the DDL that
has already been defined to the dbms, then there MUST BE a reconciliation before any
other design and development ensues.
At this point in time, the transformations can be turned into code. Having an automatic
code generator such as PRISM Solutions Warehouse Manager turns this task into an
automatic and efficient exercise. Without PRISM Solutions Warehouse Manager this
activity is done manually and is anything but fast and efficient. Furthermore, given that
iterative development is the norm for data warehouse development, the automated
approach means that future maintenance will be done automatically.
One of the aspects of automated code generation is that of automated gathering of
legacy systems metadata. When legacy systems metadata can be gathered in an
automated manner, the process of defining the metadata to the transformation program
is streamlined, and the chance for error is greatly reduced than if the metadata were
captured in a manual fashion.
Figure 17 shows the turning of transformation into code.
©Copyright 2000 by William H. Inmon, all rights reserved Page 14
BUILDING THE DATA WAREHOUSE: GETTING STARTED
Concurrent with the production of code is the task of allocating space for the data
warehouse to reside on. Figure 18 depicts this task.
In some technologies the allocation of space is very straightforward. But in other cases
there are the considerations of secondary extents and the mapping of data across the
parallel environment. When secondary extents will be allocated, care should be taken so
that:
• the secondary allocation will not occur until well into the data warehouse
development and population life cycle, and
• when secondary allocation occurs that the system programmer is alerted.
When mapping data to the parallel environment, care should be taken so that as much
growth as can be anticipated can be accommodated. Trying to reorganize data after it
has been physically loaded onto the parallel environment and mapped there is not a
pleasant proposition and should be minimized as much as possible.
Once the allocation of space has been made, the programs that have been created can
be put into execution and the data populated into the data warehouse. Figure 19 shows
the population of data into the data warehouse.
©Copyright 2000 by William H. Inmon, all rights reserved Page 15
BUILDING THE DATA WAREHOUSE: GETTING STARTED
Once the population of the data warehouse occurs, the end user - the DSS analyst - is
exposed to the data. Figure 20 illustrates the DSS analyst being allowed to access the
data warehouse.
Of course the end user needs tools for the access of data. The access of data may be
direct or may be indirect. In any case, the end user should be encouraged to explore the
possibilities of the access and manipulation of data in the data warehouse.
At this point the first iteration of development of the data warehouse ends. While there
have been many activities to do, there has been relatively little data and few types of
data to do the activities on. Furthermore, the design and development activities have
been "time boxed" so that progress is made, even if the progress is not entirely
complete. And while every effort has been made to create a good and serviceable
design, there is no intention of completeness or perfection. It is understood that errors
in design will occur and that those errors will be corrected in the second iteration (and
beyond) of data warehouse design and development.
THE SECOND ITERATION
The seeds of the second iteration of data warehouse development are sowed in the
completion of the first phase of development. An important and essential aspect of the
©Copyright 2000 by William H. Inmon, all rights reserved Page 16
BUILDING THE DATA WAREHOUSE: GETTING STARTED
development life cycle is the monitoring of the end user for feedback. Figure 21 shows
this feedback loop.
The data architect listens very carefully to the DSS analyst to determine where
improvements to the data warehouse can be made. While the conversation between the
data architect and the DSS analyst is an ongoing one, the conversation that occurs as
the data warehouse is first populated is probably the most important. It is in the first
few uses of the data warehouse that the most poignant and the most valuable feedback
typically occurs.
SUMMARY
The data warehouse is built under a very different life cycle than the classical
operational environment. The data warehouse is built iteratively - quickly and in small
iterations.
The steps of development include such activities as:
• data modelling,
• selecting hardware and software platforms,
• sizing the data warehouse,
• collecting obvious and known informational requirements,
• determining meaningful subsets of data for initial loading into the data warehouse,
• selecting the most fruitful functional area for population,
• selecting a subject area for design,
• identifying physical characteristics of the subject area,
• identifying the unit of time for the data,
• identifying the system of record,
©Copyright 2000 by William H. Inmon, all rights reserved Page 17
BUILDING THE DATA WAREHOUSE: GETTING STARTED
• determining whether delta data should be used,
• loading transformation data to metadata,
• specifying the frequency of transformation,
• executing DDL,
• creating code that embodies the transformations,
• allocating space,
• population of data into the warehouse, and
• turning the end user loose on the data.
©Copyright 2000 by William H. Inmon, all rights reserved Page 18


Use: 0.3105