Odi best-practice-data-warehouse-168255

Oracle Data Integrator
Best Practices for a Data Warehouse

An Oracle White Paper
August 2010



Preface ............................................................................................................ 4
Purpose ....................................................................................................... 4
Audience ..................................................................................................... 4
Additional Information .................................................................................. 4
Introduction to Oracle Data Integrator (ODI) .................................................... 5
Objectives.................................................................................................... 5
Business-Rules Driven Approach ................................................................ 5
Traditional ETL versus E-LT Approach ........................................................ 6
Understanding Oracle Data Integrator Interfaces ......................................... 7
A Business Problem Case Study ................................................................. 8
Implementation using Manual Coding ........................................................ 10
Implementation using Traditional ETL tools ............................................... 12
Implementation using ODI’s E-LT and the Business-rule Driven Approach 14
Benefits of E-LT Combined with a Business-rule Driven Approach ............ 17
Using ODI in your Data Warehouse Project ................................................... 19
ODI and the Data Warehouse Project........................................................ 19
Organizing the Teams ............................................................................... 19
Reverse-engineering, Auditing and Profiling Source Applications .............. 21
Designing and Implementing the Data Warehouse’s Schema.................... 23
Specifying and Designing Business Rules ................................................. 24
Building a Data Quality Framework............................................................ 27
Developing Additional Components ........................................................... 28
Packaging and Releasing Development .................................................... 29
Versioning Development............................................................................ 29
Scheduling and Operating Scenarios......................................................... 30
Monitoring the Data Quality of the Data Warehouse .................................. 30
Publishing Metadata to Business Users..................................................... 30
Planning for Next Releases ....................................................................... 31
Oracle Data Integrator for Oracle Best Practices ........................................... 32
Architecture of ODI Repositories ............................................................... 32
Reverse-engineering an Oracle Schema ................................................... 32
Oracle Loading Strategies ......................................................................... 32
Using Changed Data Capture .................................................................... 34
Oracle Integration Strategies ..................................................................... 35
Defining a Data Quality Strategy................................................................ 36
Setting up Agents in an Oracle environment .............................................. 37
Architecture Case Studies ............................................................................. 38

2


Setting up Repositories.............................................................................. 38
Using ODI Version Management ............................................................... 41
Going to Production ................................................................................... 44
Setting up Agents ...................................................................................... 46
Backing up Repositories ............................................................................ 48
Appendices ................................................................................................... 49
Appendix I. Oracle Data Integrator for Teradata Best Practices ..................... 49
Architecture of ODI Repositories ............................................................... 49
Reverse-engineering a Teradata Schema ................................................. 49
Teradata Loading Strategies ..................................................................... 50
Teradata Integration Strategies ................................................................. 51
Setting up Agents in a Teradata environment ............................................ 52
Appendix II: Additional Information ................................................................ 53
Acronyms used in this document ............................................................... 53

3


Preface
Purpose
This document describes the best practices for implementing Oracle Data Integrator (ODI)
for a data warehouse solution. It is designed to help setup a successful environment for data
integration with Enterprise Data Warehouse projects and Active Data Warehouse projects.
This document applies to Oracle Data Integrator 11g.

Audience
This document is intended for Data Integration Professional Services, System Integrators and
IT teams that plan to use Oracle Data Integrator (ODI) as the Extract, Load and Transform
tool in their Enterprise or Active Data Warehouse projects.

Additional Information
The following resources contain additional information:
• Oracle website: http://www.oracle.com
• Oracle Data Integrator 11g on-line documentation:
http://download.oracle.com/docs/cd/E14571_01/odi.htm
• Java reference : http://www.oracle.com/technetwork/java/index.html
• Jython reference http://www.jython.org

4


Introduction to Oracle Data Integrator (ODI)
Objectives
The objective of this chapter is to
• Introduce the key concepts of a business-rule driven architecture
• Introduce the key concepts of E-LT
• Understand what an Oracle Data Integrator (ODI) interface is
• Through a business problem case study, understand and evaluate some different
development approaches including:
o Manual coding
o Traditional ETL
o ODI’s business-rule driven approach combined with E-LT

Business-Rules Driven Approach

Introduction to Business rules

Business rules specify mappings, filters, joins and constraints. They often apply to metadata to
transform data and are usually described in natural language by business users. In a typical
data integration project (such as a Data Warehouse project), these rules are defined during the
specification phase in documents written by business analysts in conjunction with project
managers.
Business Rules usually define “What” to do rather than “How” to do it.
They can very often be implemented using SQL expressions, provided that the metadata they
refer to is known and qualified in a metadata repository.
Examples of business rules are given in the table below:
BUSINESS RULE TYPE SQL EXPRESSION

Sum of all amounts of items SUM(
Mapping
sold during May 2010 multiplied CASE WHEN SALES.YEARMONTH=201005 THEN
SALES.AMOUNT * PRODUCT.ITEM_PRICE
by the item price
ELSE
0
END
)
Upper(PRODUCT.PRODUCT_NAME) like ‘CPU%’
Products that start with ‘CPU’ Filter
And PRODUCT.CATEGORY = ‘HARDWARE’
and that belong to the

hardware category

CUSTOMER.CUSTOMER_ID = ORDER.ORDER_ID
Customers with their orders Join
And ORDER.ORDER_ID = ORDER_LINE.ORDER_ID

5


BUSINESS RULE TYPE SQL EXPRESSION

and order lines

CONSTRAINT CUST_NAME_PK
Reject duplicate customer Unique Key
PRIMARY KEY (CUSTOMER_NAME)
names Constraint

CONSTRAINT CUSTOMER_FK
Reject orders with a link to an Reference Constraint
FOREIGN KEY (CUSTOMER_ID)
non-existent customer REFERENCES CUSTOMER(CUSTOMER_ID)

Mappings

A mapping is a business rule implemented as a SQL expression. It is a transformation rule
that maps source columns (or fields) onto one of the target columns. It can be executed by a
relational database server at run-time. This server can be the source server (when possible), a
middle tier server or the target server.

Joins

A join operation links records in several data sets, such as tables or files. Joins are used to link
multiple sources. A join is implemented as a SQL expression linking the columns (fields) of
two or more data sets.
Joins can be defined regardless of the physical location of the source data sets involved. For
example, a JMS queue can be joined to a relational table.
Depending on the technology performing the join, it can be expressed as an inner join, right
outer join, left outer join and full outer join.

Filters

A filter is an expression applied to source data sets columns. Only the records matching this
filter are processed by the data flow.

Constraints

A constraint is an object that defines the rules enforced on data sets’ data. A constraint
ensures the validity of the data in a given data set and the integrity of the data of a model.
Constraints on the target are used to check the validity of the data before integration in the
target.

Traditional ETL versus E-LT Approach
Traditional ETL tools operate by first Extracting the data from various sources, Transforming
the data on a proprietary, middle-tier ETL engine, and then Loading the transformed data onto
the target data warehouse or integration server. Hence the term “ETL” represents both the
names and the order of the operations performed, as shown in Figure 1 below.

6


Figure 1: Traditional ETL approach compared to E-LT approach

In response to the issues raised by ETL architectures, a new architecture has emerged, which
in many ways incorporates the best aspects of manual coding and automated code-generation
approaches. Known as “E-LT”, this new approach changes where and how data
transformation takes place, and leverages existing developer skills, RDBMS engines and server
hardware to the greatest extent possible.
In essence, E-LT moves the data transformation step to the target RDBMS, changing the
order of operations to: Extract the data from the source tables, Load the tables into the
destination server, and then Transform the data on the target RDBMS using native SQL
operators. Note, with E-LT there is no need for a middle-tier engine or server as shown in
Figure 1 above.

Understanding Oracle Data Integrator Interfaces
An interface is an ODI object stored in the ODI Repository that enables the loading of one
target datastore with data transformed from one or more source datastores, based on
business rules implemented as mappings, joins, filters and constraints.
A datastore can be:
• a table stored in a relational database
• an ASCII or EBCDIC file (delimited, or fixed length)
• a node from a XML file
• a JMS topic or queue from a Message Oriented
• a node from a LDAP directory
• an API that returns data in the form of an array of records

Figure 2 shows a screenshot of an ODI interface that loads data into the FACT_SALES
target table. Source Data is defined as a heterogeneous query on the CORRECTIONS file,
the ORDERS and LINES tables.

7


Mappings, joins, filters and constraints are defined within this window.

Figure 2: Example of an ODI Interface

Wherever possible, ODI interfaces generate E-LT operations that relegate transformations to
the target RDBMS servers.

A Business Problem Case Study
Figure 3 describes an example of a business problem to extract, transform and load data from
a Microsoft SQL Server database and a file into a target Oracle table.
Data is coming from two Microsoft SQL Server tables (ORDERS joined to LINES) and is
combined with data from the CORRECTIONS file. The target SALES Oracle table must
match some constraints such as the uniqueness of the ID column and valid reference to the
SALES_REP table.
Data must be transformed and aggregated according to some mappings as shown in Figure 3.

8


Figure 3: Example of a business problem

Translating these business rules from natural language to SQL expressions is usually
straightforward. In our example, the rules that appear in the figure could be translated as
follows:
TYPE RULE SQL EXPRESSION / CONSTRAINT

Filter Only ORDERS marked as closed ORDERS.STATUS = ‘CLOSED’

Join A row from LINES has a matching ORDER_ID in ORDERS.ORDER_ID = LINES.ORDER_ID
ORDERS

Mapping Target’s SALE is the sum of the order lines’ SUM(LINES.AMOUNT + CORRECTIONS.VALUE)
AMOUNT grouped by sales rep., with the
corrections applied.

Mapping Sales Rep = Sales Rep ID from ORDERS ORDERS.SALES_REP_ID

Constraint ID must not be null ID is set to “not null” in the data model

Constraint ID must be unique A Primary Key is added to the data model with (ID) as
set of columns

Constraint The Sales Rep. ID should exist in the Target sales A Reference (Foreign Key) is added in the data model
Rep table on SALES.SALES_REP =
SALES_REP.SALES_REP_ID

9


Implementation using Manual Coding
When implementing such a data flow using, manual coding, one would probably use several
steps, several languages, and several scripting tools or utilities.
Figure 4 gives an overview of the different steps needed to achieve such an extract, transform
and load process.

Figure 4: Sequence of Operations for the Process

There are, of course, several technical solutions for implementing such a process. One of
them (probably the most efficient, as it uses an Oracle data warehouse as a transformation
engine) is detailed in the following table:
STEP DESCRIPTION EXAMPLE OF CODE

1 Execute the join between ORDERS and LINES as well Create view C$_SALES
as the filters on the source Microsoft SQL Server As select ... from ORDERS, LINES
database using a database view.
where ORDERS.STATUS = ‘CLOSED’

and ORDERS.ORDER_ID = LINES.ORDER_ID

Extract the content of the view into a flat file using the bcp C$_SALES out c$_sales_extract.bcp –c -S... -U... -P... –
BCP utility. tb

Use the SQL*Loader utility to load the temporary BCP sqlldr control=TEMP_1.ctl log=logfile.log userid=.../...
file into the TEMP_1 Oracle table.

2 Use the SQL*Loader utility to load the CORRECTIONS sqlldr control=TEMP_2.ctl log=logfile.log userid=.../...
ASCII file into the TEMP_2 Oracle table.

3 Join, transform and aggregate the 2 temporary tables insert into TEMP_SALES (...)
rd
TEMP_1 and TEMP_2 and load the results into a 3 select
table (TEMP_SALES) using SQL
SUM(TEMP_1.AMOUNT+TEMP_2.VALUE),

TEMP1.SALES_REP_ID,

...

from

TEMP_1, TEMP_2

where TEMP_1.LINEID = TEMP_2.CORR_ID)

10


STEP DESCRIPTION EXAMPLE OF CODE

...

4 Check Unique constraints using SQL and insert the insert into Errors(...)
errors into the Errors table select ... from TEMP_SALES

where ID in (select ID

from TEMP_SALES

group by ID

having count(*) > 1)

Check Reference constraints using SQL and insert the insert into Errors(...)
errors into the Error table select ... from TEMP_SALES

where SALES_REP not in

(select SALES_REP_ID from SALES_REP)

5 Finally, use SQL logic to insert / update into the target update SALES set ...
SALES table using a query on TEMP_SALES from ...

where ID in

(select ID

from TEMP_SALES

where IND_UPDATE=’U’)

insert into SALES (...)

select ...

from TEMP_SALES

where IND_UPDATE=’I’

...

The benefits of this approach are:
• High performance:
o Uses pure set-oriented SQL to avoid row-by-row operations
o Uses Oracle as a transformation engine to leverage the power of the
RDBMS
o Uses in-place utilities such as External Tables
• Code flexibility:
o Leverages the latest features of Oracle such as the built-in transformation
functions

11


However, this approach raises several issues that become painful as the Enterprise Data
Warehouse projects grow, and more developers get involved. These issues are:
• Poor productivity
o Every load process needs to be developed as a set of scripts and programs,
within different environments and with several languages.
o Business rules (“what happens to the data” – SUM(AMOUNT + VALUE))
are mixed with technical logic (“how to move / load the data” –
SQL*Loader, External Table, insert etc.)
o Moving to production is often difficult when developers haven’t designed
environment variables, or variable qualified names for their objects.
• High cost of data quality implementation
o Data cleansing / data quality according to predefined constraints is usually
avoided due to the cost of its implementation
o Every project has its own definition of the data quality without any
centralized framework (default structure of the error tables, error recycling
etc.)
• Hard maintenance
o Even when project managers have set up a framework, every script may
“reinvent the wheel” and contain specific routines that make it hard to
understand
o Developments are spread in several machines and folders, without a central
repository
o Impact analysis is impossible as there is no metadata management and no
cross references mechanism
• No project flexibility
o The cost of a change to the data models or to the business rules becomes
such a constraint that IT teams refuse it, leading to frustration amongst the
business users.

Implementation using Traditional ETL tools
Traditional ETL tools perform all the transformations in a proprietary engine. They often
require additional hardware to stage the data for the transformations. None of them really
leverages the power of the RDBMS.
A typical ETL architecture is shown in Figure 5.

12


Figure 5: Implementation Using an ETL

Every transformation step requires a specific connector or transformer.
ETL tools are often known for the following advantages:
• Centralized development and administration
o Single graphical user interface
o Centralized repository
• Easier maintenance
o Impact analysis (for certain tools)
Unfortunately this ETL approach presents several drawbacks:
• Poor performance
o As the data needs to be processed in the engine, it is often processed row by
row
o When data from the target database is referenced - table lookups for
example - it needs to be extracted from the database, into the engine and
then moved back again to the target database.
o Very few mappings, joins, aggregations and filters are given to the powerful
engine of the RDBMS
• Bad productivity
o Every load process needs to be developed as a set of steps that mix business
rules (“what happens to the data” – SUM(AMOUNT + VALUE)) with
technical logic (“how to load the data” – connector 1, connector 2 etc.)
o Moving to production is often difficult when developers haven’t designed
environment variables, or variable qualified names within their queries
o Some of them still require the use of heavy manual coding to achieve certain
particular tasks, or to leverage the RDBMS’ powerful transformation
functions

13


• High Cost
o ETL tools require additional hardware
o ETL tools require specific skills

Implementation using ODI’s E-LT and the Business-rule Driven Approach
Implementing a business problem using ODI is a very easy and straightforward exercise. It is
done by simply translating the business rules into an interface. Every business rule remains
accessible from the Diagram panel of the interface’s window.

Specifying the Business Rules in the Interface

Figure 6 gives an overview of how the business problem is translated into an ODI interface:
• The ORDERS, LINES and CORRECTION datastores are dragged and dropped into
the “Source” panel of the interface
• The Target SALES datastore is dropped in the “Target Datastore” panel
• Joins and filters are defined by dragging and dropping columns in the “Source” panel
• Mappings are defined by selecting every target column and by dragging and dropping
columns or by using the advanced expression editor.
• Constraints are defined in the “Control” tab of the interface. They define how flow data
is going to be checked and rejected into the Errors table.

Figure 6: Implementation using Oracle Data Integrator

14


Business Rules are Converted into a Process

Business rules defined in the interface need to be split into a process that will carry out the
joins, filters, mappings and constraints from source data to target tables. Figure 7 defines the
problem to be solved.

Figure 7: How to convert business rules into a process?

By default, Oracle Data Integrator will use the RDBMS as a staging area for loading source
data into temporary tables and applying all the required mappings, staging filters, joins and
constraints.
The staging area is a separate area in the RDBMS (a user/database) where ODI creates its
temporary objects and executes some of the rules (mapping, joins, final filters, aggregations
etc). When performing the operations this way, ODI leverages the E-LT architecture as it first
extracts and loads the temporary tables and then finishes the transformations in the target
RDBMS.
In some particular cases, when source volumes are small (less than 500,000 records), this
staging area can be located in memory in ODI’S in-memory relational database – ODI
Memory Engine. ODI would then behave like a traditional ETL tool.
Figure 8 shows the process automatically generated by Oracle Data Integrator to load the
final SALES table. The business rules, as defined in Figure 7 will be transformed into code by
the Knowledge Modules (KM). The code produced will generate several steps. Some of
these steps will extract and load the data from the sources to the staging area (Loading
Knowledge Modules - LKM). Others will transform and integrate the data from the staging area
to the target table (Integration Knowledge Module - IKM). To ensure data quality, the Check
Knowledge Module (CKM) will apply the user defined constraints to the staging data to
isolate erroneous records in the Errors table.

15


Figure 8: ODI Knowledge Modules in action

ODI Knowledge Modules contain the actual code that will be executed by the various servers
of the infrastructure. Some of the code contained in the Knowledge Modules is generic. It
makes calls to the ODI Substitution API that will be bound at run-time to the business-rules
and generates the final code that will be executed. Figure 9 illustrates this mechanism.
During design time, rules are defined in the interfaces and Knowledge Modules are selected.
During run-time, code is generated and every API call in the Knowledge Modules (enclosed
by <% and %>) is replaced with its corresponding object name or expression, with respect to
the metadata provided in the Repository.
For example, a call to <%=odiRef.getTable(“TARG_NAME”)%> will return the name of the
target table of the interface with the appropriate qualifier according to context information,
topology setup etc.
A typical SQL INSERT statement would be coded in a Knowledge Module as follows:
INSERT INTO <%=odiRef.getTable(“TARG_NAME”)%> ...

This template of code will of course generate different SQL statements depending on the
target table (“INSERT INTO MyDB1.SALES...” when the target is the SALES table, “INSERT
INTO DWH_DB.PRODUCT” when the target is the PRODUCT table etc.). This is also particularly
useful when migrating ODI processes from an environment to another (promoting processes
from Development to QA for example) as ODI will automatically substitute the correct

16


schema information based on the specified environment without requiring any code
modifications.

Figure 9: How Knowledge Modules generate native code

Once the code is generated, it is submitted to an ODI Agent, which will either redirect it to
the appropriate database engines and operating systems, or will execute it when needed
(memory engine transformation, java or jython code etc.)
In most cases, the agent is simply a conductor that does not touch the data.

Benefits of E-LT Combined with a Business-rule Driven Approach
Compared to other architectures (manual coding and traditional ETL), ODI mixes the best of
both worlds:
• Productivity / Maintenance
o The business-rules driven approach delivers greater productivity as
developers simply need to concentrate on the “What” without caring about
the “How”. They define SQL expressions for the business rules, and ODI
Knowledge Modules generate the entire set of SQL operations needed to
achieve these rules.
o When a change needs to be made in operational logic (such as “creating a
backup copy of every target table before loading the new records”), it is

17


simply applied in the appropriate Knowledge Module and it automatically
impacts the hundreds of interfaces already developed. With a traditional
ETL approach, such a change would have necessitated opening every job
and manually adding the new steps, increasing the risk of mistakes and
inconsistency.
o Flexibility and a shallow learning curve are ensured by leveraging the
RDBMS’ latest features.
o With a centralized repository that describes all the metadata of the sources
and targets and a single unified and comprehensive graphical interface,
maintenance is greatly optimized as cross-references between objects can be
queried at any time. This gives the developers and the business users a single
entry point for impact analysis and data lineage (“What is used where?”,
“Which sources populate which targets?” etc.)
o In the ODI repository, the topology of the infrastructure is defined in detail,
and moving objects between different execution contexts (Development,
Testing, QA, Production, etc.) is straightforward. With a powerful version
control repository, several teams can work on the same project within
different release stages, with guaranteed consistency of deliverables.
o With a centralized framework for Data Quality, developers spend less time
on defining technical steps, and more time on the specification of data
quality rules. This helps to build a consistent and standardized Data
Warehouse.
• High Performance:
o The E-LT architecture leverages the power of all the features of in-place
databases engines. ODI generates pure set-oriented SQL optimized for each
RDBMS which can take advantage of advanced features such as parallel
processing or other advanced features.
o Native database utilities can be invoked by the ODI Knowledge Modules
provided.
o When data from the target database is referenced - table lookups for
example, it doesn’t need to be extracted from the database, into an engine.
It remains where it is, and it is processed by database engine.
• Low Cost:
o Oracle Data Integrator doesn’t require a dedicated server. The loads and
transformations are carried out by the RDBMS.
In conclusion, with its business-rule driven E-LT architecture, Oracle Data Integrator is the
best solution for taking advantage of both manual coding and traditional ETL worlds.

18


Using ODI in your Data Warehouse Project
ODI and the Data Warehouse Project
The main goal of a Data Warehouse is to consolidate and deliver accurate indicators to
business users to help them make decisions regarding their everyday business. A typical
project is composed of several steps and milestones. Some of these are:
• Defining business needs (Key Indicators)
• Identifying source data that concerns key indicators; specifying business rules to
transform source information into key indicators
• Modeling the data structure of the target warehouse to store the key indicators
• Populating the indicators by implementing business rules
• Measuring the overall accuracy of the data by setting up data quality rules
• Developing reports on key indicators
• Making key indicators and metadata available to business users through ad-hoc query
tools or predefined reports
• Measuring business users’ satisfaction and adding/modifying key indicators
Oracle Data Integrator will help you cover most of these steps, from source data investigation
to metadata lineage and through loading and data quality audit. With its repository, ODI will
centralize the specification and development efforts and provide a unique architecture on
which the project can rely to succeed.

Organizing the Teams
As Oracle Data Integrator relies on a centralized repository, different types of users may need
to access it. The list below describes how ODI may be used by your teams.
PROFILE DESCRIPTION ODI MODULES USED

Business User Business users have access to the final calculated key indicators through reports ODI Console
or ad-hoc queries. In some cases, they need to understand what the definition of
the indicators is, how they are calculated and when they were updated.
Alternatively, they need to be aware of any data quality issue regarding the
accuracy of their indicators.

Business Analyst Business Analysts define key indicators. Designer Navigator

They know the source applications and specify business rules to transform (limited access)

source data into meaningful target indicators. ODI Console

They are in charge of maintaining translation data from operational semantics to
the unified data warehouse semantic.

Developer Developers are in charge of implementing the business rules in respect to the Topology Navigator
specifications described by the Business Analysts. They release their work by (read only access)
providing executable scenarios to the production team. Developers must have Designer Navigator:

19


PROFILE DESCRIPTION ODI MODULES USED

both technical skills regarding the infrastructure and business knowledge of the Limited access to
source applications. Models

Full access to Projects

Operator Navigator

ODI Console

Metadata Metadata Administrators are in charge of reverse-engineering source and target Topology Navigator
Administrator applications. They guarantee the overall consistency of Metadata in the ODI (limited access)
Repository. They have an excellent knowledge of the structure of the sources and Designer Navigator:
targets and they have participated in the data modeling of key indicators. In
Full access to Models
conjunction with Business Analysts, they enrich the metadata by adding
comments, descriptions and even integrity rules (such as constraints). Metadata Restore access to

Administrators are responsible for version management. Projects

ODI Console

Database Database Administrators are in charge of defining the technical database Topology Navigator
Administrator infrastructure that supports ODI. They create the database profiles to let ODI (full access)
access the data. They create separate schemas and databases to store the Designer Navigator
Staging Areas. They make the environments accessible by describing them in the (full access)
Topology
Operator Navigator
(full access)

ODI Console

System Administrator System Administrators are in charge of maintaining technical resources and Agents
infrastructure for the project. For example, they may Topology Navigator
install and monitor Scheduler Agents (limited access)

backup / restore Repositories ODI Console

install and monitor ODI Console

Setup environments (development, test, maintenance etc.)

Security Administrator The Security Administrator is in charge of defining the security policy for the ODI Security Navigator
Repository. He or she creates ODI users and grants them rights on models, (full access)
projects and contexts. Designer Navigator
(read access)

Topology Navigator
(read access)

ODI Console

Operator Operators are in charge of importing released and tested scenarios into the Operator Navigator
production environment. They schedule the execution of these scenarios. They ODI Console
monitor execution logs and restart failed sessions when needed.
Oracle Enterprise
Manager Plug-in For
ODI

20


The ODI Master Repository contains built-in default profiles that can be assigned to users.
The following table suggests how to use these built-in profiles:

PROFILE BUILT-IN PROFILE IN ORACLE DATA INTEGRATOR

Business User CONNECT, NG REPOSITORY EXPLORER

Business Analyst CONNECT, NG REPOSITORY EXPLORER, NG DESIGNER

Developer CONNECT, DESIGNER

Metadata Administrator CONNECT, METDATA ADMIN, VERSION ADMIN

Database Administrator CONNECT, DESIGNER, METADATA ADMIN, TOPOLOGY ADMIN

System Administrator CONNECT, OPERATOR

Security Administrator CONNECT, SECURITY ADMIN

Operator CONNECT, OPERATOR

Reverse-engineering, Auditing and Profiling Source Applications
A good starting point for projects is to understand the contents and the structure of source
applications. Rather than having paper-based documentation on source applications, a good
practice is to connect to these source applications using ODI and to capture their metadata.
Once this is achieved, it is usually helpful to define some data quality business rules in ODI
Repository. This helps check the data consistency in the source and validate your
understanding of the data models. At this phase of the project it is important to start having
an answer to the following questions:
• How many different sources do we have to take in account to calculate our indicators?
• Is the data needed for our indicators present in the source systems?
• What data quality challenges will we have to address to ensure the accuracy of the target
warehouse?
• What source system(s) will provide the master data (dimensions) that will be referenced
by our indicators?
• What data volumes are we going to manipulate?
• And so forth.
In some cases, source applications are not accessible directly. Only ASCII or binary files
extracts from these applications are provided. Starting to work with these source files before
implementing the data warehouse model is recommended, as they represent the “vision of the
truth” that the production source systems want to give you. The metadata of these files
should typically be described in the repository. Samples of the files can also be initially loaded
into temporary target tables to validate their structure and content.

21


All this can be implemented in Oracle Data Integrator as follows:
• Connect source applications or files in Topology Navigator
• Define a logical architecture in Topology Navigator
• Create one model per logical schema in Designer Navigator
• Reverse Engineer models when possible or manually describe datastores
o Use standard JDBC Reverse-engineering to get database metadata
o Use Customized Reverse-engineering strategies (Reverse Knowledge
Modules) when standard JDBC reverse is not applicable (or not accurate)
o Use Cobol Copy Book import for ASCII or binary files if available
o Use Delimited File Reverse for ASCII delimited files
• Enrich the metadata by adding information when it is not available:
o Datastores and Columns descriptions and titles
o Unique keys information (Primary Keys, Alternate Keys)
o Referential Integrity assumptions between datastores (Foreign Keys)
o Check constraints (for example to check the domain of values for a column)
• When the source data is stored in files, develop simple interfaces to load these files into
a temporary area in the target database to better evaluate their level of data quality.
• Locate the important entities in the source data and use the following features to profile
the contents:
o View Data on a datastore
o View distribution of values for a column of a datastore
o Count the number of records
o Run ad-hoc SQL queries from Designer Navigator
• Validate the constraints that you have defined in the models by performing data quality
control:
o Schedule or interactively run data quality “static” control on the datastores
and models
o Access and understand the contents of the error tables
o Propose alternatives for handling data discrepancies in the future
o Define the “acceptable” level of data quality for the target warehouse
Of course, these action items should be done in conjunction with business analysts. Normally,
in parallel to this phase, you will be designing the data model of the target Data Warehouse.

22


Data quality audit errors and data profiling will help you refine your understanding of the
truth, and consequently, will lead you to a better modeling of the target warehouse.

Designing and Implementing the Data Warehouse’s Schema
This section could be a book on its own. The goal here is not to provide you with guidelines
in Data Warehouse modeling, but rather to give some general hints and tips that may affect
your ODI development.
It is always a good practice to have an Operational Data Store (ODS) to store raw operational
data. Usually, the data model of the ODS is very close to the OLTP source applications data
model. Any source data should be accepted in the ODS and almost no data quality rule
should be implemented. This ensures to have a store representing all the data-of-the-day from
the operational systems.
When designing the Data Warehouse schema, the following general tips may be helpful:
• Where possible, describe columns inside the data dictionary of the database (for
example use “COMMENT ON TABLE” and “COMMENT ON COLUMN” SQL
statements). By doing so, you allow ODI to retrieve these comments and descriptions
into its metadata repository.
• Design a storage space for the Staging Area where ODI will create its necessary
temporary tables and data quality error tables.
• Do not use primary keys of the source systems as primary key for your target tables.
Use counters or identity columns whenever it is possible. This makes a flexible data
model that will easily evolve over time.
• Design referential integrity (RI) and reverse engineer foreign keys in ODI models. Do
not implement these foreign keys in the target database as they may lead to
performance issues. With the automatic data quality checks, ODI will guarantee the
consistency of data according to these RI rules.
• Standardize your object naming conventions. For example, use three letters to prefix
table names by subject area. Avoid using very long names as ODI may add a prefix to
your table names when creating temporary objects (for example E$_Customer will be
the error table for table Customer).
• Use of either 3rd Normal Form modeling (3NF) or dimensional modeling (“Snow
Flakes” or “Star Schemas”) doesn’t have any impact on ODI, but rather on the way you
will design your future business rules development and business users’ reports.
Once the ODS and DW are designed, you will have to create the appropriate models in
Designer and perform reverse-engineering.
ODI comes with modeling capabilities in the form of a feature called Common Format
Designer. It can help you design the structure of the ODS and the DW from the existing
structures of the source systems by simply drag and dropping tables columns and constraints.

23


The generated Data Description Language statements (DDL) automatically include specific
features of the target database. It also keeps track of the origin of source columns to
automatically generate the interfaces for loading the target, thus resulting in significant time
savings. Best practices on the usage of Common Format Designer are out of the scope of this
document.

Specifying and Designing Business Rules
Usually this step starts almost at the same time as the design of the target schema.
Understanding the business needs is clearly a key factor in the success of the overall project.
This influences the key indicators selected for the warehouse as well as the rules used to
transform source to target data. The more accurate the specification of a business rule, the
easier the design of this rule in ODI. Several successful projects have used ODI Designer to
specify the business rules, thus avoiding having discrepancies between a Word document and
the final developed rules.
The following table summarizes what needs to be specified for every target table of the ODS
or the DW before designing business rules.
TARGET DATASTORE: Give the name of the target datastore, possibly preceded by a qualification prefix (such
as the database or schema name)

DESCRIPTION OF THE Give a short description of the purpose of the transformation
TRANSFORMATION:

INTEGRATION STRATEGY: Define how data should be written to the target. Examples:

Replace content with new data from sources

Append incoming data onto the target

Update existing records and insert new ones according to an update key (specify the key
here)

Use a Slowly Changing Dimension strategy. Specify the surrogate key, the slowly
changing attributes, the updatable ones, the columns for start and end date etc.

A home-made strategy (for example: BASEL II compliance, SOX historical audit trail
etc.)

Every strategy specified here will correspond to an ODI Integration Knowledge Module.

REFRESH FREQUENCY: Specify when this datastore should be loaded (every night, every month, every 2 hours
etc)

DEPENDENCIES: Give all dependencies in the load process (e.g. what datastores have to be loaded prior
to the current one; what specific jobs need to be executed prior to this one etc.)

SOURCE DATASTORES: Give the list of source datastores involved for loading the target table. This list should
include all the lookup tables as well. For each source datastore, include the following
information.

SOURCE ODI MODEL Give the name of the source model as it appears in Designer

DATASTORE NAME Give the name of the datastore

24


PURPOSE/DESCRIPTION Indicate whether it is the main source set or a lookup one. If Change Data Capture is to
be used, it should be indicated here.

FIELD MAPPINGS AND For every target field (or column), indicate what transformations need to be applied to
TRANSFORMATIONS: the source fields. These transformations should be written as expressions or formulas as
often as possible.

TARGET COLUMN Give the target column name

MAPPING DESCRIPTION Describe the purpose of the mapping

MAPPING EXPRESSION Give the expression using the source column names. Try to express it in pseudo code.

LINKS OR JOIN CRITERIA For every pair of source datastores, specify the criteria to lookup matching records. This
is often known as an SQL join.

DATASTORE 1 Name of the first datastore

DATASTORE 2 Name of the second datastore

LINK EXPRESSION Give the expression used to link the 2 datastores in pseudo code

DESCRIPTION Describe this link by specifying if it is a left, right or full outer join.

FILTERS: Give the list of filters that apply to source data. Filters should be expressed in natural
language and in pseudo code whenever possible.

FILTER DESCRIPTION Describe this Filter.

FILTER DESCRIPTION Give the expression used to implement the filter in pseudo code.

DATA QUALITY REQUIREMENTS: List here all the data quality requirements including error recycling if appropriate. The
requirements should be expressed as constraints whenever possible.

CONSTRAINT NAME Name or short description of the constraint to validate.

DESCRIPTION Purpose of the data quality check

CONSTRAINT EXPRESSION Expression in pseudo code required to control the data

To illustrate the use of such a form, the following table gives an example for the business
problem case study defined in the chapter “Introduction to Oracle Data Integrator (ODI)”.
TARGET DATASTORE: Oracle Warehouse.SALES

DESCRIPTION OF THE Aggregate orders and order lines from the Microsoft SQL Server source production system.
TRANSFORMATION:

INTEGRATION STRATEGY: Append new records of the day

REFRESH FREQUENCY: Every night

DEPENDENCIES: The following tables need to be loaded prior to this one for referential integrity:

PRODUCT

SALES_REP

SOURCE DATASTORES:

25


SOURCE ODI MODEL DATASTORE NAME PURPOSE/DESCRIPTION

Microsoft SQL Server Source ORDERS Orders from the production system

Microsoft SQL Server Source LINES Order lines. Main source for aggregation

User Data Files CORRECTIONS File containing the value to add to the
amount sold if any manual correction has
occurred for a particular order line.

FIELD MAPPINGS AND

TRANSFORMATIONS:

TARGET COLUMN MAPPING DESCRIPTION MAPPING EXPRESSION

ID Order Id ORDERS.ORDER_ID

PRODUCT_PK Product Id as it appears in the order lines LINES.PRODUCT_ID

SALES_REP Sales Rep Id as it appears in the order ORDERS.SALES_REP_ID

SALE Amount sold. If a correction for this SUM ( LINES.AMOUNT +
amount exists in the “corrections” file, the (CORRECTIONS.VALUE when it exists) )
value should be added. The total value
needs to be summed.

QUANTITY Total quantity of product sold. SUM ( LINES.QUANTITY )

LINKS OR JOIN CRITERIA

DATASTORE 1 DATASTORE 2 LINK EXPRESSION DESCRIPTION

ORDERS LINES Link orders and order lines ORDERS.ORDER_ID =
on the order ID. Every order LINES.ORDER_ID
line must match an existing
order

LINES CORRECTIONS Lookup a correction value LINES.LINE_ITEM_ID =
from the correction file if it CORRECTIONS.CORR_ID
exists given the line item ID
(left join)

Filters:
FILTER DESCRIPTION FILTER EXPRESSION

Orders of the day ORDER.ORDER_DATE between yesterday and now

Orders validated ORDER.STATUS = Closed

DATA QUALITY REQUIREMENTS:

CONSTRAINT DESCRIPTION CONSTRAINT EXPRESSION

Reference to SALES_REP Ensure that every SALES_REP ID exists in SALES_REP references SALES_REP(ID)
the SALES_REP reference table

Reference to PRODUCT Ensure that every PRODUCT_PK exists in PRODUCT_PK references PRODUCT(ID)
the PRODUCT reference table

ID Not null ID Column is mandatory Check ID not null

Quantity greater than 0 The quantity sold must always be positive QUANTITY > 0

Uniqueness of sale A sales rep can’t sell the same product twice (PRODUCT_PK, SALES_REP) is unique
inside the same order

26


The next step is to design these business rules in the Designer Navigator. The translation
from the specification to the design is straightforward. Every specification of the loading of a
target table will be converted into an interface. This process is even faster if the specification
is made directly using the Designer Navigator.
The steps to implement the interface are typically as follows.
• Drag and drop the target datastore
• Drag and drop source datastores
• For every target field, translate the pseudo code of the transformation into SQL
expressions. When possible, decide where to execute the expression (source, staging
area, target)
• For every join, define the join in the source panel of the interface. Depending on the
specification, define the join as inner, left, right, natural or full outer join. When
possible, try to execute joins on the sources to minimize network traffic.
• For every filter, translate the pseudo code into an SQL expression. When the source
database accepts SQL filters, set it as “execute on source” to minimize network traffic.
• In the flow diagram of the interface, define the loading strategies (LKM) and specify the
integration strategy (IKM) to match the specification. Choose the appropriate options
for the Knowledge Module and activate the data quality control when appropriate.
• In the control tab of the interface, select the constraints to control. If the constraints
are not defined on the target datastore, define them in the Models view of the Designer
Navigator.
After the design of every interface, you should review the code generated using the
Simulation mode and then test it by executing it. The Operator Navigator allows you to easily
follow the execution of your interface to get the number of records processed and many
other useful indicators (such as the generated code, the elapsed time, the number of inserts,
updates, deletes etc.) After the interface completes, the resulting data and errors can be
directly selected from Designer to validate the accuracy of the business rules.
Thanks to ODI Contexts, the execution of the interface will happen on a “Development”
environment and will not affect production data.

Building a Data Quality Framework
With an approach based on business rules, ODI is certainly the most appropriate tool to help
you build a data quality framework to track data inconsistencies. Thanks to the Check
Knowledge Modules, you simply have to define the control business rules, and inconsistent
data is automatically isolated for you in error tables. However, isolating erroneous data is not
the only issue in Data Quality. Even if ODI automatically detects duplicated keys, mandatory
fields, missing references, and more complex constraints, you will need to involve business
users in the process of qualifying data discrepancies and make the appropriate decisions. You

27


should have an answer to each of the following questions before defining your data quality
strategy:
• What level of data quality is required for the Data Warehouse?
• Who are the business owners of source data?
• What should we do with rejected records?
• Do we need to define an error recycling strategy?
• Do we need to involve business owners of the data and report rejected records?
• How would business users modify erroneous source data?
• Do we provide a GUI to have them modify erroneous records in the error tables?
The document “Comprehensive Data Quality with Oracle Data Integrator” further discusses
Data Quality issues and the recommended best practices.

Developing Additional Components
Not all the typical tasks of the loading of the Data Warehouse can be achieved using ODI
Interfaces. It is usual to develop additional components or jobs that would carry on tasks
such as, for example:
• Receiving and sending e-mails
• Copying, moving, concatenating, renaming files in the file system
• Compressing, decompressing files
• Executing web services
• Writing and executing shell scripts for a specific operating system
• Writing and executing java programs
• And so on.
These components can be developed and tested within the Designer Navigator as Procedures,
Variables, User Functions, or Steps in a package. ODI Procedures offer a broad range of
possibilities for developing these components. They can include, for example:
• An ad-hoc SQL statement written for any database
• An Operating System call
• ODI built-in tools and APIs (Send Mail, Read Mail, Wait For File etc.)
• Code written in any of the scripting languages supported by the Jakarta Bean Scripting
Framework. This includes Java, Java Script, Python, Perl, NetRexx, Groovy, etc.
Of course, the risk here is to start developing transformation processes as Procedures by hand
coding shell and SQL scripts rather than using the powerful mechanism of ODI Interfaces

28


combined with Knowledge Modules. To avoid that, try as much as possible to specify your
transformations as business rules and not as a technical processes. ODI Procedures should
always be considered as technical steps that have to be achieved in the overall process, but
they shouldn’t have in-depth business logic that applies to the data. Typical Data Warehouse
projects would have less than 10% of development in the form of Procedures.

Packaging and Releasing Development
Now that the Interfaces, Procedures and Variables are developed and tested, they need to be
ordered as steps within Packages. Start thinking about what should happen in case of error in
any of these steps. By default, the ODI Agent will stop the execution of a package at the step
that has failed and it will rollback any open transaction on all the connected databases. Even
though errors are reported in ODI Logs, it is always a good practice to have an “on-error”
procedure that is triggered if some of the steps fail. This procedure can for example send an
email to an operator to warn them that the package has failed, reporting the session ID.
Particular attention has to be paid to the following step attributes:
• Next step on failure
• Number of attempts in case of failure
• Interval between attempts
Try to avoid heavy loops inside packages (more than 50 iterations). In most cases, a loop can
be avoided by simply adding a table as source to an interface!
When the package is successfully tested, you will have to release it as a Scenario. Think of a
scenario as a compiled version of your source objects such as interfaces, procedures,
packages, etc. This Scenario will then be imported in the Test environment before validation
for Production.

Versioning Development
Before going to production, you should record a stable version of all the objects involved in
the release of the Scenarios. Creating versions of objects will allow you to restore previous
released items for further maintenance in case of problem.
Starting from ODI 10g, versions of objects are stored in the Master Repository and the
dependencies between objects are maintained in objects called Solutions. To make sure all the
objects involved in the release are consistently versioned, it is thus recommended that you
create a Solution object and add the Project to this Solution. Designer calculates all the
dependencies between objects and automatically creates a new version for Projects, Models,
Scenarios and global objects used. Restoring these objects from a previous version can be
done when needed from the Designer Navigator by simply selecting the Restore menu.
Section “Using ODI Version Management” further discusses the best practices for version
control.

29


Scheduling and Operating Scenarios
Scheduling and operating scenarios is usually done in the Test and Production environments
in separate Work Repositories. The Operator Navigator provides a graphical interface to
perform these tasks. Any scenario can be scheduled by an ODI Agent or by any external
scheduler, as scenarios can be invoked by an operating system command.
When scenarios are running in production, agents generate execution logs in an ODI Work
Repository. These logs can be monitored either through the Operator Navigator, through any
web browser using the ODI Console or the Oracle Enterprise Manager Plug-in for ODI.
Operating teams need to be trained on the use of the Operator Navigator, ODI Console or
Oracle Enterprise Manager. They will have to monitor logs, restart failing jobs and submit ad-
hoc tasks for execution.

Monitoring the Data Quality of the Data Warehouse
Business users rely on the accuracy of key indicators of the Data Warehouse to make
decisions. If these indicators are wrong, the decisions are worthless.
Depending on the data quality strategy you have defined, business users will actively
participate in the monitoring of data discrepancies. They will have to help the IT team to
better refine the calculations of the indicators as well as the qualification of the erroneous
data. This generally leads to modifications of the business rules. These updates are done in the
development environment and follow the normal release cycle defined for the project.
Typical projects report data quality errors to business users through reports generated from
ODI’s error tables. These reports are pushed to the data owner in their email system or to any
workflow application used internally.

Publishing Metadata to Business Users
Business users usually need to have access to other data quality indicators based on metadata
such as:
• When was the last time my table was updated?
• How many records were added, removed or updated in my table?
• What are the rules that calculate a particular indicator?
• Where does the data come from, and how is it transformed?
• Where does the data go to, and how is it transformed?
• Etc.
All these questions can be answered if you give access to the ODI Console to business users.
This is done simply by assigning them a user ID in the Security Navigator. The web-based
interface allows them to see all the data models, and the interactions between them. This
includes:

30


• Flow Maps
• Data Lineage, which is usefully for understanding the path taken by data and the
transformations between applications
• Execution logs with accurate statistics on the number of records processed (number of
inserts, updates, delete and errors).

Planning for Next Releases
A Data Warehouse evolves as the business need of the enterprise change. This leads to
frequent updates of the data models. This constant changing has an impact on the
development of new components and the maintenance of existing ones. However it is
important to understand that it should always be a cycle driven by business requirements. The
general steps listed below give you a brief overview of how this could be planned:
1. Define or refine Business Rules specifications
2. Make sure previous releases of the objects are properly versioned and can be restored
safely in a new empty work repository before modifying anything in your current work
repository.
3. Define new sources or targets in Topology Navigator.
4. Reverse-engineer new and/or existing models. Use the ODI cross references to
evaluate the impact of the changes to the structure of source and target fields. Pay
special attention to fields that were removed from tables as they may still be used in
some transformations.
5. Develop new interfaces and update existing ones. Create or update additional
components. Test every item separately in the development environment.
6. Update existing packages or create new ones. Test every package, even those that
haven’t changed since the last release, as the underlying schema may have changed.
7. Regenerate the new scenarios for your next release.
8. Release the scenarios into each Test environment. Correct any bug that would be
reported from the Test process. Involve business users in the test process.
9. Create new Solutions and version every project and model. Use the change indicators
that appear on every object to detect which objects should be versioned.
10. When acceptance tests are successful, release your new scenarios into production.

31


Oracle Data Integrator for Oracle Best Practices
Architecture of ODI Repositories
It is recommended that you install the ODI Master and Work repositories in an OLTP
database distinct from the source or target applications.

Reverse-engineering an Oracle Schema
The Oracle JDBC driver implements most metadata APIs. You can therefore use the
standard JDBC reverse-engineering with ODI. Using this method, you can capture metadata
for:
• Tables, Views etc. including table comments,
• Columns including data types, length and scale, and comments
• Primary Keys
• Foreign Keys
• Check Constraints
In some cases a custom reverse-engineering approach using the RKM Oracle may be
necessary. For example ODI 11g natively supports database partitions which can only be
reverse-engineered using the RKM Oracle.
RKM DESCRIPTION

RKM Oracle Reverse-engineering Knowledge Module for Oracle

This RKM uses the Oracle system tables known as the Oracle Data Dictionary to extract the
metadata definition of a table and its related artifacts (columns, keys, partitions etc.).

Oracle Loading Strategies
When loading an Oracle data warehouse the goal is always to load the data into the database
in the most efficient manner and Oracle Data Integrator provides several Loading Knowledge
Modules optimized for Oracle that will help you achieve this. Choosing the right Knowledge
Module for each use case will greatly improve the efficiency of your data integration
processes.

32


You can refer to the following table for a list of the Loading Knowledge Modules optimized
for Oracle:
LKM DESCRIPTION

LKM File to Oracle (EXTERNAL TABLE) Loads data from a file to an Oracle staging area using the EXTERNAL
TABLE SQL Command.

LKM File to Oracle (SQLLDR) Loads data from a file to an Oracle staging area using the SQL*Loader
command line utility. The Oracle client must be installed on the machine
hosting the ODI Agent.

LKM SQL to Oracle Loads data through the agent from any SQL RDBMS to an Oracle
staging area.

LKM MSSQL to Oracle (BCP SQLLDR) Loads data from a Microsoft SQL Server database to an Oracle staging
area using the BCP and SQL*Loader utilities. The utilities must be
installed on the machine hosting the ODI Agent.

LKM Oracle to Oracle (DBLINK) Loads data from an Oracle source database to an Oracle staging area
database using database links.

LKM Oracle to Oracle (datapump) Loads data from an Oracle source database to an Oracle staging area
database using external tables in the datapump format.

There are also several additional Knowledge Modules which can be used to extract data from
SAP ERP, SAP BW or Oracle BI and load the records into an Oracle database.

Using Loaders for Flat Files

When your interface contains a flat file as a source, you may want to use a strategy that
leverages the most efficient loading utility available for your staging area technology rather
than the standard “LKM File to SQL” which uses the ODI Agent to load the flat file data.
The agent uses JDBC to write to the target, using this method would lead to significant
performance issues compared to loading into Oracle using the External Table SQL command
or SQL*Loader.
We advise against using the ODI Agent when loading large files and recommend the “LKM
File to Oracle (EXTERNAL TABLE)” to transfer large files into an Oracle database.

Using Unload/Load for Remote Servers

When the source result set is located on a remote database server, an alternative to using the
agent to transfer the data would be to unload the date into a file and then load it into the
staging area.
The “LKM MSSQL to Oracle (BCP SQLLDR)” is an example of a Knowledge Module that
follows these steps, it uses the BCP utility to bulk unload data from a Microsoft SQL Server
database into a file and SQL*Loader to load the data back into an Oracle staging area. This
Loading Knowledge Module provides the most optimal approach to extracting data out of
Microsoft SQL Server and loading it into an Oracle database.

33


Several Knowledge Modules use a similar approach including the SAP ERP and BW LKMs.

Oracle to Oracle

There are use cases in which there is a need to move data between Oracle systems and ODI
provides several ways to achieve this.
One of the best methods when it comes to loading data in such a scenario is to use Oracle
Data Pump and Oracle Data Integrator provides the “LKM Oracle to Oracle (datapump)”
which leverages Oracle Data Pump to export and load the data efficiently through the
creation of External Tables using Oracle Data Pump on both the source and the target
databases.
ODI also provides the “LKM Oracle to Oracle (DBLINK)” which has the ability to create a
database link between the source and the target Oracle database to load the data.

Using Changed Data Capture
Applying a Change Data Capture (CDC) approach can significantly improve the performance
of your ODI processes.
For example if you are loading your data warehouse nightly you only need to extract the data
that has changed in the past 24 hours, loading the entire source data sets will result in a
significant degradation of the efficiency of your ODI jobs.
Using CDC ensures that the extract from your various source systems is done incrementally.
This reduces the amount of data transferred from your source systems into your target
database as well as the amount of data transformed by ODI. The creation of the CDC
process can be challenging and ODI simplifies it by providing Knowledge Modules which
leverage many CDC mechanisms including the native database CDC capabilities.

Change Data Capture on Oracle

When it comes to detecting changes occurring in an Oracle database ODI offers several
options:
• CDC using Oracle GoldenGate
• CDC using Oracle Streams
• CDC using database triggers
The Oracle Journalization Knowledge Modules (JKMs) automatically create the underlying
infrastructure required by the ODI CDC framework so that end users do not have to worry
about it. For example, the Oracle JKMs have the ability to configure Oracle GoldenGate or
the Oracle Streams infrastructure when required.

34


You can refer to the following table for a list of the JKMs for Oracle provided with ODI:
JKM DESCRIPTION

JKM Oracle to Oracle Consistent (OGG) Creates and manages the ODI CDC framework infrastructure when using
Oracle GoldenGate for CDC.

JKM Oracle 10g Consistent (Streams) Creates the journalizing infrastructure for consistent set journalizing on Oracle
10g tables, using Oracle Streams.

JKM Oracle 11g Consistent (Streams) Creates the journalizing infrastructure for consistent set journalizing on Oracle
11g tables, using Oracle Streams.

JKM Oracle Consistent Creates the journalizing infrastructure for consistent set journalizing on Oracle
tables using triggers

JKM Oracle Consistent (Update Date) Creates the journalizing infrastructure for consistent set journalizing on Oracle
tables using triggers based on a Last Update Date column on the source
tables

JKM Oracle Simple Creates the journalizing infrastructure for simple journalizing on Oracle tables
using triggers

It is also possible to leverage other mechanism in ODI to detect changes like timestamps,
dates or sequence numbers.

There are many more JKMs provided with ODI, you can refer to the ODI Documentation
for a complete list of the JKMs.

Oracle Integration Strategies
The Integration Knowledge Modules optimized for the Oracle database are implementing
many of the best practices to insert data into an Oracle data warehouse:
- Direct Path loading is applied whenever possible using the /*+ APPEND */ hint
- Additional optimizer hints such as PARALLEL which specifies the degree of
parallelism for a given database operation can be specified via the
OPTIMIZER_HINT option of the IKMs
- The ODI Staging Area tables are created by default using NOLOGGING to ensure
best performance
- The MERGE SQL statement can be leveraged using the “IKM Oracle Incremental
Update (MERGE)”. MERGE combines an INSERT and an UPDATE in one single
optimized SQL command
- ODI gathers tables statistics to ensure that the Oracle database Cost Based
Optimizer (CBO) selects the most optimal execution plan over time
- Indexes are created to speed up the joins and filters execution. End users can choose
from several options including the creation of bitmap indexes which can greatly

35


improve the efficiency of joins between large fact tables and smaller dimension tables
in a star schema.
You can refer to the following table for a list of the IKMs that can be used with an Oracle
database:
IKM DESCRIPTION

IKM SQL Control Append Integrates data in any ISO-92 compliant database target table in
TRUNCATE/INSERT (append) mode.) Data quality can be checked. Invalid
data is rejected in the “E$” error table and can be recycled.

IKM Oracle Incremental Update Set-based incremental update for Oracle

IKM Oracle Slowly Changing Dimension Slowly Changing Dimension Type II for Oracle

IKM Oracle Incremental Update Integrates data in an Oracle target table in incremental update mode, using a
(MERGE) MERGE statement

IKM Oracle Multi Table Insert Integrates data from one source into one or many Oracle target tables in
append mode, using a multi-table insert statement (MTI).

IKM Oracle AW Incremental Update Integrates data in an Oracle target table in incremental update mode and is
able to refresh a Cube in an Analytical Workspace.

It is also possible to extend the ODI Knowledge Modules with your own best practices when
needed. Thanks to the reusability of the KMs it is simple to ensure that best practices are
enforced and utilized by each developer.

Defining a Data Quality Strategy
Loading data that is not clean in your data warehouse can have a major impact on the overall
performance of your ETL processes. Primary key or foreign key violations, business rules that
are not met, null values that are not expected can waste time and resources if the appropriate
Data Quality framework is not in place within your data integration process.
Oracle Data Integrator provides an out of the box framework to handle Data Quality errors
occurring while data is being loaded into a database. Business rules and constraints are
defined at the metadata level and those validation rules are checked against the data prior to
being integrated into the target tables. The constraints can be enabled or disabled when
required. The errors that are captured are routed into error tables managed by ODI that can
be reviewed through the Designer Navigator as well as the ODI Console.
“CKM Oracle” is the recommended Check Knowledge Module for checking data integrity
constraints defined on a given Oracle table.

36


Setting up Agents in an Oracle environment

Where to Install the Agent(s)?

A typical data warehouse implementation can have one or more ODI agent in production. In
an Oracle environment, the agent is usually installed on the host machine that is used to load
data in the data warehouse. The agent requires a connection to the source databases or files,
and triggers the appropriate load utilities.
Starting from ODI 11g it is also possible to install the ODI agents in WebLogic Server to
achieve High-Availability (HA) through a clustered deployment.

37


Architecture Case Studies
Setting up Repositories

General Architecture for ODI Repositories

In a typical environment for a data warehouse project, you would create the following
repositories:
• A single master repository that holds all the topology and security information. All the
work repositories are registered in this master repository. This single master repository
contains all the versions of objects that are committed by the designers.
• A “Development” work repository shared by all ODI designers. This repository holds
all the projects and models under development.
• A “Testing” work repository shared by the IT testing team. This repository contains
all the projects and models being tested for future release.
• A “User Acceptance Tests” work repository shared by the IT testing team and the
business analysts. This repository contains all the projects and models about to be
released. Business analysts will use the ODI Console on top of this repository to
validate the scenarios and transformations before releasing them to production.
• A “Production” work repository shared by the production team, the operators and the
business analysts. This repository contains all the projects and models in read-only
mode for metadata lineage, as well as all the released scenarios.
• A “Hot fix” work repository shared by the maintenance team and the development
team. This work repository is usually empty. Whenever a critical error happens in
production, the maintenance team restores the corresponding projects and models in
this repository and performs the corrections with the help of the development team.
Once the problems are solved, the scenarios are released directly to the production
repository and the new models and projects are versioned in the master repository.

38


This recommended architecture is described in the figure below:
Development Team
Maintenance Team
IT Testing Team
Business Analysts
Production Team

Development Hot Fix Testing User Acceptance Production
Tests
Work Repositories Projects Projects Projects Projects Projects

Models Models Models Models Models

Scenarios and Logs Scenarios and Logs Scenarios and Logs Scenarios and Logs Scenarios and Logs

Create and archive versions Import versions for Hot Fix Import released versions of models, projects Import released and tested
of models, projects and and release scenario patches and scenarios for testing versions of projects, models
scenarios and scenarios for production
Master Repository
Master Repository
Topology

Security

Versions

Figure 10: ODI Repositories and Teams Organization

The master repository and all work repositories are usually created in the same OLTP
database instance in separate schemas or catalogs.
When developers have finished working on a project and decide to release it, they create a
version for their projects and models and store it in the master repository. This version is
then restored by the IT testing team in the testing repository. After the technical tests have
completed, the testing team initializes the “user acceptance tests” repository for business
analysts. They restore the same version they were working on to have it tested by business
users. Once this version is functionally accepted, it is restored by the production team in the
production repository.
When a critical bug is discovered in production, the developers are usually already working on
the next release. Therefore they are usually not able to stop their development and restore the
previous version for corrections. The maintenance team is then in charge of restoring the
version used in production into a separate empty work repository called “Hot Fix” and
applying the appropriate fixes. Once done, the maintenance team releases its modified
projects, models and scenarios into the master repository so that the production team can
restore them as a patch in the production repository.

Creating a Separate Master Repository for Production

For some particular security requirements, you may not want to share the same master
repository between development and production. In this case, the solution is to duplicate the
master repository and to isolate the production environment from other environments as
shown below:

39


Development Production
Environment Environment
Development Team
Maintenance Team
IT Testing Team
Business Analysts
Production Team

Development Hot Fix Testing User Acceptance Production
Tests
Work Repositories Projects Projects Projects Projects Projects

Models Models Models Models Models

Scenarios and Logs Scenarios and Logs Scenarios and Logs Scenarios and Logs Scenarios and Logs

Create and archive versions Import versions for Hot Fix Import released versions of models, projects
of models, projects and and release scenario patches and scenarios for testing
scenarios
Development Master XML Production Master
Master Repository XML
Topology Export/Import released versions for
production using XML export/import Topology
Security
Security
Versions Manually replicate the Topology

Figure 11: Several Master Repositories

To create a new master repository for your production environment, it is recommended that
you use an export of your master repository from your Topology Navigator and then use the
Master Repository Import Wizard. You can find the Master Repository Import Wizard in the
ODI Studio. When creating the new master repository, you should assign a new ID to it,
different from the ID used by the development master repository.
Once created, do the following to set up the production environment:
• Create the production context
• Create all the production data servers and physical schemas in the physical architecture
of the Topology
• Link your production physical schemas to the existing logical schemas defined by the
designers in the production context.
• Do not change or remove existing contexts and logical schemas.
• Update the Security so that only production users and business analysts can access the
repository.
• Create the production work repository and give it an ID different from any of the IDs
used by the other work repositories (Development, Hot Fix, Testing and User
Acceptance Tests). See Understanding the Impact of Work Repository IDs.
• Every time the development master repository is updated, manually replicate the
changes in the production master repository.

40


• Export projects, models and scenarios that are ready for production from the
development master repository into XML files. You can use the “Version Browser” to
do so. Alternatively, if using Solutions, export your solutions into compressed files
• Connect to the production work repository with Designer and import the XML files or
the solution’s compressed files.

Understanding the Impact of Work Repository IDs

When creating a master repository or a work repository, ODI asks you for a 3 digit ID for
this repository. You should select this ID by following the rules below:
• Every master repository created in your enterprise should have a unique ID.
• Every work repository created in your enterprise should have a unique ID even if it
belongs to a different master repository.
Every type of object in an ODI repository has a unique ID calculated according to the
following rule:
<auto number> concatenated with the 3 digits of the repository ID.
For example, if the internal ID of an interface is 1256023, you can automatically guess that it
was first created in the work repository ID 023.
The main goal of this rule is to enable export and import in “Synonym mode” of objects
across work repositories without any risk of ID collision.
If 2 work repositories have the same ID, there is a chance that 2 different objects within these
2 repositories have the same ID. Therefore, importing the first object from the first
repository to the second may overwrite the second object! The only way to avoid that is, of
course, to have 2 different work repository IDs.

Using ODI Version Management

How Version Management Works

Version management in ODI is designed to allow you to work on different versions of
objects across work repositories as shown below:

41


Development Master
Solution V1.0
Topology
Several versions Solution V2.0
Archived
Security Solution V2.1

Versions Solution V2.0.5

Created Imported validated
versions: version 2.0 for critical
Imported validated
1.0, 2.0 and 2.1. maintenance. Is about
Imported version 2.0 for
Is about to to release version
released version production.
release version 2.0.5 that fixes the
2.1 for testing. Discovered a critical
2.2 critical bug.
bug in this version.

Development Testing Production Hot Fix

Projects Projects Projects Projects

Models Models Models Models

Scenarios and Logs Scenarios and Logs Scenarios and Logs Scenarios and Logs

Open versions Versions being Versions in Versions being
Currently under tested production fixed for a critical
development patch

Solution V2.2 Solution V2.1 Solution V2.0 Solution V2.0.5
Figure 12: Version Management with ODI

Note: The user acceptance tests repository does not appear in this diagram as the mechanism
for populating it with new versions is similar to the one used for the testing repository.
The developers have already released versions 1.0, 2.0 and 2.1 of their development in the
master repository. Every time they have released a version they have used a Solution. See
Using Solutions for Configuration Management for details. They are now working on the next
release as they are about to release solution version 2.2. Meanwhile, the IT testing team is
testing version 2.1 in its work repository. The production team is still working on version 2.0
and it has discovered a critical bug in this version. Therefore, the maintenance team has
restored version 2.0 in the Hot Fix work repository and is working on a patch release, v2.0.5.
As soon as this patch release is committed to the master repository, the production team will
have to restore it into the production repository. Developers will have to manually update
versions 2.1 and 2.2 so they reflect the changes made by the maintenance team to version
2.0.5.
Note: Every work repository can only have a single version of an object.

Creating and Restoring Versions of Objects

To create a version of an object, users simply have to right-click on the object and select
“Version->Create Version”.

42


When the version of the object is created, it becomes available in the “Version Browser” and
in the “Version” tab of the object. ODI also updates all the “I” and “U” flags relating to this
object to indicate that it is up to date with the version in the master repository. If you change
your object in the work repository after committing it in the master, you will notice that its
icon is changed and a small “U” appears besides it to indicate that its status is now
“Updated”. This feature is very helpful as it shows you all the differences between the current
version and the last version committed in the master repository.
When you create a version of an object, the object is exported to XML in memory,
compressed and stored in the master repository as a binary large object. Therefore, you
should consider creating a version of an object only when you are about to commit it
definitely for a release.
To restore an object to one of its previous versions, users simply have to right-click on the
object and select “Version->Restore ”. Then select from the list of available versions which
one to restore. You should use caution for the following reasons:
• All updates that you have done since the last time you have created a version of your
current object will be lost
• The object that you are about to restore may reference other objects that were removed
from your work repository. The typical situation is restoring an interface that references
a column that doesn’t exist anymore in your models. If this happens, the restored object
will be marked as “invalid” with a red exclamation mark and you will have to edit it to
correct all missing references.

Using Solutions for Configuration Management

During design time you will probably create versions of your objects for backup. For
example, you will create versions of your models, folders, interfaces, variables, Knowledge
Modules, etc. Most of these objects are designed to work together as they may depend on one
another. For example, interface version 2.1.0 requires tables from model 3.1.2 and Knowledge
Modules version 1.2 and 4.5. Maintaining the list of dependencies can be very tedious and
time consuming. Therefore, when you are about to release what has been developed for the
testing team, you would prefer to have a single object that manages these dependencies.
ODI solution objects are designed to manage these dependencies between a project and all
models, global objects and scenarios it references. When you are about to release your
development, you simply need to create a new solution and drag and drop your project into
this solution. ODI will then create a new version of your project (if required) and a new
version of every model, scenario, other project or global object referenced by this project.
Therefore, instead of releasing several objects, you simply release the solution.
Notes:
For performance reasons, you should consider having small projects to improve the efficiency
of the version creation mechanism. It is recommended that you split your development into
several small projects containing less than 300 objects.

43


When calculating the dependencies for a project in order to build the solution, ODI creates a
version of every model referenced by this project. Then it looks at every model to see what
Knowledge Modules it references (RKM, JKM and CKM). And finally it creates a new
version of every project to which these Knowledge Modules belong. So if your models
reference Knowledge Modules from 3 different projects, the solution will reference these 3
projects as well. A best practice to avoid this is to create a project called “Metadata
Knowledge Modules” and to have in this project all the RKMs, CKMs and JKMs that you
will reference in your models. Therefore every time you create a solution, this project will be
added as a dependency, instead of 3 or 4 different projects.

Going to Production

Releasing Scenarios

Section Using ODI Version Management gives an overview on how to handle versions of objects
to release them to the production environment. However, sometimes you will only want your
scenarios in the production work repository without their corresponding projects and models.
For that, you would create an “Execution” work repository rather than a “Development”
work repository. When your scenarios are released into the master repository or as simple
XML files, you would either restore them to your work repository or import them in
“synonym” or “duplication” mode from the Operator Navigator.

Executing Scenarios

Executing a Scenario Manually

You can execute your scenario manually from the Operator Navigator. In this case if your
scenario requires variables to be set, you would edit every variable used by the scenario and
update its default value. To execute the scenario, you also need to indicate the context for the
execution as well as the logical agent in charge of this execution.
An alternative way to execute a scenario is to use the ODI Console to start the scenario from
your web browser. This approach can be useful if you are considering having scenarios that
will be executed on-demand by your business users and analysts. Should your scenario require
variables, the user would be asked to fill in the variable values accordingly.

Executing a Scenario Using an Operating System Command

You can also start a scenario from an operating system command using the “startscen” shell
script. You would do that if you plan to use an external scheduler to schedule your jobs. Refer
to the ODI Documentation for details on this operating system command
When starting a scenario from the operating system, you do not need a running agent. The
“startscen” command has its own built-in agent that executes the scenario and stops at the
end of the session.

44


Assigning Session Keywords for a Scenario

You may occasionally find it difficult to browse the Operator Navigator log as it may contain
a large number of sessions. To overcome this limitation, ODI allows keywords to be assigned
to a session. This session will be stored automatically in the appropriate session folder created
in the Operator Navigator.
In the figure below, the folder “Product Dimensions” was created for the “PDIM” keyword.
When starting the LOAD_FAMILY_HIERARCHY and LOAD_PRODUCTS scenarios, the
“-KEWORDS” parameter was set to “-KEYWORDS=PDIM”. Therefore when using the
operator to access the executed sessions, they are automatically available under the “Product
Dimensions” folder.

Figure 13: Session Folders and Keywords

You can use this feature to distribute your sessions across your business areas.

Using the Operator Navigator in Production

The Operator Navigator in the ODI Studio is the module that is used the most in a
production environment. From this module, operators have the ability to:
• Import scenarios into the production work repository
• Execute scenarios manually and assign default values to scenario variables
• Follow execution logs with advanced statistics including the number of row processed
and the execution elapsed time
• Abort a currently-running session.
• Restart a session that has failed and possibly redefine the restart point by updating the

45


task statuses.
• Create session folders to hold sessions with pre-defined keywords
• Access and define scheduling information

Using the ODI Console in Production

The ODI Console can be used in production for 2 purposes:
• To give business users, analysts and operators access to metadata lineage, metadata
searching, scenario execution, and log monitoring. In this situation, the production
work repository should be a “development” repository, and all the released projects,
models and scenarios should be restored in this repository.
• To give operators access only to scenarios and logs. In this case, the production work
repository would only be an “execution” work repository, and would only contain the
released scenarios.

Setting up Agents


A typical data warehouse implementation usually requires one or more ODI agents in
production. In a data warehousing environment an ODI agent is usually installed on the host
machine that is used to load data in the data warehouse. The agent requires a connection to
source databases or files, and triggers the appropriate load utilities.
ODI 11g introduces Java EE agents which are deployed in a Java EE application server and
can leverage its features, such agents are recommended when one of your requirements is to
centralize the deployment and management of your agents or when there is a need for high
availability.
The network bandwidth between the machine hosting the agents and the data warehouse
should be large enough to handle the volume of data that will be loaded by the utilities in the
warehouse database. On the other hand, as the agents may need to access other source
servers, close attention should be paid to the network bandwidth to and from these source
servers.
If your Knowledge Modules generate operating-system-specific commands, these must match
the operating system on which the agent is installed.
In a normal environment, you will set up:
• 1 physical agent for the development team, preferably on the same operating system as
the production agent
• 1 physical agent shared by the testing team, the maintenance team and the user
acceptance team. Again, try to use the same operating system as the agent in
production.

46


• 1 physical agent for the production

Using Load Balancing

There are cases when a single agent can become a bottleneck, especially when it has to deal
with large numbers of sessions started in parallel. For example, suppose you want to retrieve
source data from 300 stores. If you attempt to do this in parallel on a single agent, it may lead
to excessive overhead as it will require opening 300 threads (1 for every session). A way to
avoid that is to set up load balancing across several agents.
To set up load balancing in ODI, you can follow these steps in the Topology Navigator:
• Define the agent that will be in charge of distributing the workload. In the following,
this agent will be referred to as the “master agent”. The master agent usually has only 2
concurrent sessions. All the scenarios will be executed by this single agent by using the
–AGENT parameter.
• Define several child agents that will be in charge of actually executing the scenarios. Set
the maximum number of sessions of each agent to a value between 2 and 20.
• Edit the master agent in the Topology Navigator, and in the “Load Balancing” tab,
select all your child agents.
The diagram below gives you an example of how you can setup this architecture:

Machine 2

AGENT1 3 Running Sessions
4 sessions
Port: 20900
Sessions: 3 1 Pending Session Queued

Machine 1 4 sessions AGENT2 3 Running Sessions
Port: 20910
14 incoming Sessions: 3 1 Pending Session Queued
sessions Master Agent
Port: 20910
Sessions: 2
Machine 3

3 sessions AGENT3
Port: 20900 3 Running Sessions
Sessions: 3

3 sessions AGENT4
Port: 20910 3 Running Sessions
Sessions: 3

Figure 15: Example of Load Balancing

In this example, 4 agents are installed on 2 different machines. Each of them accepts a
maximum of 3 sessions in parallel. The master agent is installed on another machine.

47


The master agent receives 14 requests to start several scenarios in parallel. It distributes these
14 requests to the available agents. Each agent will start 3 scenarios in parallel, and the first 2
agents will simply queue the remaining sessions for future execution. As soon as one of the
agents becomes available, it will pick one of the queued sessions and begin executing it.
The master agent initially calculates the workload per agent according to the ratio given by the
number of sessions currently being executed versus the maximum number of sessions
allowed for this agent. It distributes incoming sessions one by one to every agent and then
recalculates the workload’s ratio. As the queued sessions for an agent can be picked up by
another agent, the entire system becomes balanced.
This flexible architecture allows you to add as many agents as needed to reach the maximum
scalability for your system. If you plan to start several scenarios in parallel, you can refer to
section Using ODI Version Management for a detailed example on how to achieve that
elegantly.
Notes: The number of agents required and the maximum number of sessions that they may
support will depend on your environment settings. It should be tested carefully before going
to production. The main bottleneck in this architecture is obviously the RDBMS hosting the
ODI repositories as all agents access the same database tables. It is recommended that you
dedicate a database server to the ODI repository if you plan to setup such architecture. You
would therefore have more flexibility to tune this server in the case of locks or performance
issues.

Backing up Repositories
ODI recommends that you backup your master and work repositories on a daily basis to
avoid any loss of your work.
To backup your repositories, you simply need to backup every schema on which you have
installed a master or work repository, using your standard database backup procedure.
Refer to your specific database documentation for information on how to backup a schema
or catalog.

48


Appendices

Appendix I. Oracle Data Integrator for Teradata Best
Practices
Architecture of ODI Repositories
It is recommended that you install the master and work repositories in an OLTP database
distinct from the source or target applications. The Teradata database is not recommended
for hosting ODI repositories.

Reverse-engineering a Teradata Schema
The Teradata JDBC driver implements most metadata APIs. You can therefore use the
standard JDBC reverse-engineering with ODI. Using this method, you can capture metadata
for:
• Tables and Views including table comments,
• Columns including data types, length and scale, and comments
• Primary Keys when they are defined using a PRIMARY KEY constraint statement in
the databases
However, you will probably face the following limitations:
• Foreign Key metadata is not implemented in the JDBC driver. Therefore ODI will not
retrieve foreign keys.
• Unique Primary Indexes (UPI) and Non Unique Primary Indexes (NUPI) are not
imported. To have a UPI imported as a primary key in ODI, you need to define it as a
PRIMARY KEY constraint at the database level.
• Other indexes are not imported
• Check constraints are not imported
You can bypass some of these limits by using the Reverse Knowledge Module for Teradata
provided with Oracle Data Integrator. This RKM is based on the DBC catalogue tables
(DBC.Tables, DBC.Columns etc). You may also enhance this Knowledge Module and adapt
it to your needs.
RKM DESCRIPTION

RKM Teradata Retrieves metadata from the Teradata database using the DBC system views.
This RKM supports UNICODE columns.

49


Teradata Loading Strategies
You can refer to the following table for Loading Knowledge Modules optimized for Teradata:
LKM DESCRIPTION

LKM SQL to SQL Loads data through the agent from any SQL RDBMS to any SQL
RDBMS staging area

LKM File to Teradata (TTU) Loads data from a File to a Teradata staging area database using the
Teradata bulk utilities. The utilities must be installed on the machine
hosting the ODI Agent.

LKM SQL to Teradata (TTU) Loads data from a SQL compliant source database to a Teradata
staging area database using a native Teradata bulk utility. The utilities
must be installed on the Unix machine hosting the ODI Agent.

Using Loaders for Flat Files

When your interface contains a flat file as a source, you may want to use a strategy that
leverages the most efficient loading utility available for the staging area technology, rather
than the standard “LKM File to SQL”.
The ODI Agent accepts binary files and supports several native binary data types such as
Binary Big and Little Endian, Ebcdic, Ebcdic zoned decimals, Packed decimals, etc. However,
we recommend against using ODI Agent to load large binary files into the Teradata staging
area.
As the agent uses JDBC to write to the target, using this method would lead to significant
performance issues compared to loading using the native Teradata Tools and Utilities (TTU)
which include FastLoad, MulitLoad, TPump, Teradata Parallel Transporter or FastExport.
You can refer to the TTU documentation for a way to load binary files in a Teradata table.
The “LKM File to Teradata (TTU)” should meet your needs for generating and executing the
appropriate TTU scripts.

Using Unload/Load for Remote Servers

When the source result set is on a remote database server, an alternative to using the agent to
transfer the data would be to unload it to a file and then load that into the staging area.
The “LKM SQL to Teradata (TTU)” follows these steps and uses the OdiSqlUnload tool to
unload data from any remote RDBMS. Of course, this KM can be optimized if the source
RDBMS is known to have a fast unload utility.

Using Piped Unload/Load

When using an unload/load strategy, data needs to be staged twice: once in the temporary file
and a second time in the “C$” table, resulting in extra disk space usage and potential
efficiency issues. A more efficient alternative would be to use pipelines between the “unload”
and the “load” utility. Unfortunately, not all the operating systems support file-based pipelines
(FIFOs).

50


ODI provides the “LKM SQL to Teradata (TTU)” that uses this strategy. To have a better
control on the behavior of every detached process (or thread), this KM was written using
Jython. The OdiSqlUnload tool is also available as a callable object in Jython.

Teradata Integration Strategies
The following table lists some of the Integration Knowledge Modules optimized for Teradata:
IKM DESCRIPTION

IKM Teradata Control Append Integrates data in a Teradata target table in replace/append mode.

IKM Teradata Incremental Update Integrates data in a Teradata target table in incremental update mode.

IKM Teradata Slowly Changing Integrates data in a Teradata target table used as a Type II Slowly Changing
Dimension Dimension in your Data Warehouse.

IKM File to Teradata (TTU) This IKM is designed to leverage the power of the Teradata utilities for loading
files directly to the target.

IKM SQL to SQL Append Integrates data into an ANSI-SQL92 target database from any remote ANSI-
SQL92 compliant staging area in replace or append mode.

IKM SQL to Teradata (TTU) Integrates data from a SQL compliant database to a Teradata database target
table using Teradata Utilities TPUMP, FASTLOAD OR MULTILOAD.

IKM Teradata Multi Statement Integrates data in Teradata database target table using multi statement
requests, managed in one SQL transaction.

IKM Teradata to File (TTU) Integrates data in a target file from a Teradata staging area in replace mode.

IKMs with Staging Area Different from Target

File to Server Append

There are some cases when your source is composed of a single file that you want to load
directly into the target table using the most efficient method. By default, ODI will suggest
putting the staging area on the target server and performing such a job using an LKM to stage
the file in a “C$” table and an IKM to apply the source data of the “C$” table to the target
table.
The recommended Knowledge Module when a file needs to loaded into Teradata without
staging the data is the “IKM File to Teradata (TTU)”, which allows the generation of
appropriate scripts for each of these Teradata utilities depending on your integration strategy.

Server to Server Append

When using a staging area different from the target and when setting this staging area to an
RDBMS, you can use an IKM that will move the transformed data from the staging area to
the remote target. This kind of IKM is very close to an LKM and follows almost the same
rules.
The IKM SQL to Teradata (TTU) uses this strategy.

51


Server to File or JMS Append

When the target datastore is a file or JMS queue or topic you will obviously need to set the
staging area to a different place than the target. Therefore, if you want to target a file or a
queue datastore you will have to use a specific “Multi-Connection” IKM that will export the
transformed data from your staging area to this target. The way that the data is exported to
the file or queue will depend on the IKM. For example, you can choose to use the agent to
have it select records from the staging area and write them to the file or queue using standard
ODI features. Or you can use specific unload utilities such as Teradata FastExport if the
target is not JMS based.
You can refer to the following IKMs for further details:
IKM DESCRIPTION

IKM Teradata to File (TTU) Integrates data in a target file from a Teradata staging area in
replace mode.

IKM SQL to File Append Exports data from any SQL compliant staging area to an ASCII or
BINARY file using the ODI File driver

Setting up Agents in a Teradata environment


A typical data warehouse implementation usually requires one or more ODI agents in
production. In a Teradata environment, the agent is usually installed on the host machine that
is used to load data in the data warehouse. The agent requires a connection to source
databases or files, and triggers the appropriate load utilities (fastload, multiload, tpump).
Starting from ODI 11g it is also possible to install the ODI agents in WebLogic Server to
achieve High-Availability (HA) through a clustered deployment.

52


Appendix II: Additional Information
Acronyms used in this document
3NF 3rd Normal Form

API Application Programming Interface

ASCII American Standard Code for Information Interchange

CKM Check Knowledge Module

DDL Data Description Language

DML Data Manipulation Language

DW Data Warehouse

EBCDIC Extended Binary-Coded Decimal Interchange Code

E-LT Extract, Load and Transform

ETL Extract, Transform and Load

GUI Graphical User Interface

HTTP Hypertext Transport Protocol

IKM Integration Knowledge Module

IT Information Technology

JEE Java Enterprise Edition

JDBC Java Database Connectivity

JKM Journalizing Knowledge Module

JMS Java Message Services

JNDI Java Naming Directory Interface

JVM Java Virtual Machine

KM Knowledge Module

LDAP Lightweight Directory Access Protocol

LKM Loading Knowledge Module

ODS Operational Data Store

OLTP Online Transactional Processing

53


RDBMS Relation Database Management System

RI Referential Integrity

RKM Reverse-engineering Knowledge Module

SOX Sarbanes-Oxley

SQL Simple Query Language

URL Unique Resource Locator

XML Extended Markup Language

54

Oracle Data Integrator Best Practices for a Data Copyright © 2010, Oracle and/or its affiliates. All rights reserved. This document is provided for information purposes only and the
Warehouse contents hereof are subject to change without notice. This document is not warranted to be error-free, nor subject to any other
August 2010 warranties or conditions, whether expressed orally or implied in law, including implied warranties and conditions of merchantability or
Author: ODI Product Management fitness for a particular purpose. We specifically disclaim any liability with respect to this document and no contractual obligations are
Contributing Authors: ODI Product Development formed either directly or indirectly by this document. This document may not be reproduced or transmitted in any form or by any
means, electronic or mechanical, for any purpose, without our prior written permission.
Oracle Corporation
World Headquarters Oracle and Java are registered trademarks of Oracle and/or its affiliates. Other names may be trademarks of their respective
500 Oracle Parkway owners.
Redwood Shores, CA 94065
U.S.A. AMD, Opteron, the AMD logo, and the AMD Opteron logo are trademarks or registered trademarks of Advanced Micro Devices. Intel
and Intel Xeon are trademarks or registered trademarks of Intel Corporation. All SPARC trademarks are used under license and are
Worldwide Inquiries: trademarks or registered trademarks of SPARC International, Inc. UNIX is a registered trademark licensed through X/Open
Phone: +1.650.506.7000 Company, Ltd. 0410
Fax: +1.650.506.7200
oracle.com

Odi best-practice-data-warehouse-168255

More Related Content

What's hot (20)

Viewers also liked (20)

Similar to Odi best-practice-data-warehouse-168255 (20)

Recently uploaded (20)

Odi best-practice-data-warehouse-168255