Oracle9i: SQL Tuning Workshop

Instructor Guide Vol. 1

D12395GC10
Production 1.0
January 2002
D34338
Authors Copyright © Oracle Corporation, 1998, 1999, 2000, 2001, 2002. All rights reserved.

Nancy Greenberg This documentation contains proprietary information of Oracle Corporation. It is provided
under a license agreement containing restrictions on use and disclosure and is also
Priya Vennapusa
protected by copyright law. Reverse engineering of the software is prohibited. If this
documentation is delivered to a U.S. Government Agency of the Department of Defense,
then it is delivered with Restricted Rights and the following legend is applicable:
Technical Contributors
Restricted Rights Legend
and Reviewers
Use, duplication or disclosure by the Government is subject to restrictions for commercial
Howard Bradley computer software and shall be deemed to be Restricted Rights software under Federal
Laszlo Czinkoczki law, as set forth in subparagraph (c)(1)(ii) of DFARS 252.227-7013, Rights in Technical
Dan Gabel Data and Computer Software (October 1988).
Connie Dialeris Green This material or any portion of it may not be copied in any form or by any means without
John Hibbard the express prior written permission of Oracle Corporation. Any other copying is a
Lilian Hobbs violation of copyright law and may result in civil and/or criminal penalties.
John Hoff
If this documentation is delivered to a U.S. Government Agency not within the
Alexander Hunold Department of Defense, then it is delivered with “Restricted Rights,” as defined in FAR
Tamas Kerepes 52.227-14, Rights in Data-General, including Alternate III (June 1987).
Susan Kotsovolos
Herve Lejeune The information in this document is subject to change without notice. If you find any
problems in the documentation, please report them in writing to Education Products,
Stefan Lindblad Oracle Corporation, 500 Oracle Parkway, Box SB-6, Redwood Shores, CA 94065. Oracle
Diana Lorentz Corporation does not warrant that this document is error-free.
Howard Ostrow
All references to Oracle and Oracle products are trademarks or registered trademarks of
Arjan Pellenkoft
Oracle Corporation.
Stacey Procter
Shankar Raman All other products or company names are used for identification purposes only, and may
Mariajesus Senise be trademarks of their respective owners.
Janet Stern
Don Sullivan
Ric Van Dyke
Lachlan Williams

Publisher
Shane Mattimoe
Contents

Instructor Preface

1 Following a Tuning Methodology

Objectives 1-2
Overview 1-3
Managing Performance 1-4
Factors to Be Managed 1-6
Performance Problems 1-8
Critical Resource 1-10
Excessive Demand 1-11
Understanding Scalability 1-13
Internet Scalability 1-15
Scalability with Application Design, Implementation, and Configuration 1-16
The Tuning Methodology 1-17
Tuning Roles 1-18
Tuning SQL Statements 1-19
Applying the Methodology 1-21
Summary 1-23
2 SQL Statement Processing
Objectives 2-2
Overview 2-3
System Global Area (SGA) 2-4
The Shared Pool 2-5
Shared SQL Areas 2-6
The Program Global Area (PGA) 2-7
SQL Statement Processing Phases 2-8
Sharing Cursors: Benefits 2-12
Sharing Cursors: Requirements 2-13
Writing SQL to Share Cursors 2-14
Bind Variables and Shared Cursors 2-15
Writing SQL to Share Cursors 2-16
Monitoring Shared Cursors 2-18
V$LIBRARYCACHE Columns 2-19
V$SQLAREA Columns 2-20
Monitoring Shared Cursor Use 2-21
Cursor Sharing 2-23
Automated SQL Execution Memory Management 2-24
Dynamic Performance Views 2-25
V$SYSSTAT and V$SESSTAT 2-26
Summary 2-27
Practice Overview 2-29

iii
3 EXPLAIN and AUTOTRACE
Objectives 3-2
Creating the Plan Table 3-3
The EXPLAIN PLAN Command 3-5
EXPLAIN PLAN Example 3-6
Displaying the Execution Plan 3-7
Interpreting the Execution Plan 3-9
Using V$SQL_PLAN 3-11
V$SQL_PLAN Columns 3-12
Querying V$SQL_PLAN 3-14
SQL*Plus AUTOTRACE 3-15
SQL*Plus AUTOTRACE Examples 3-16
SQL*Plus AUTOTRACE Statistics 3-17
Summary 3-18
Practice Overview 3-19
4 SQL Trace and TKPROF
Objectives 4-2
SQL Trace Facility 4-3
How to Use the SQL Trace Facility 4-4
Initialization Parameters 4-5
Switching On SQL Trace 4-7
Finding Your Trace Files 4-8
Formatting Your Trace Files 4-9
TKPROF Command Options 4-11
Output of the TKPROF Command 4-12
TKPROF Output Example: No Index 4-17
TKPROF Output Example: Unique Index 4-18
Some TKPROF Interpretation Pitfalls 4-19
Summary 4-20
Practice Overview 4-21
5 Rule-Based Optimization Versus Cost-Based Optimization
Objectives 5-2
Overview 5-3
Functions of the Oracle9i Optimizer 5-4
Rule-Based Optimization 5-5
Cost-Based Optimization 5-6
Choosing Between RBO and CBO 5-8
Setting the Optimizer Approach 5-9
First Rows Optimization 5-10
Rule-Based Optimization 5-11
RBO Ranking Scheme 5-12
Rule-Based Optimization Example 5-13
Influencing Rule-Based Optimization 5-14

iv
 Summary 5-15
Practice Overview 5-16
Guided Practice Page 5-17
6 Indexes and Basic Access Methods
Objectives 6-2
ROWIDs 6-3
Indexes 6-4
B*-Tree Indexes 6-5
B*-Tree Index Structure 6-6
B*-Tree Index Example 6-7
CREATE INDEX Syntax 6-8
Composite Index Guidelines 6-9
Index Statistics 6-10
Effect of DML Operations on Indexes 6-11
Indexes and Constraints 6-12
Indexes and Foreign Keys 6-13
Basic Access Methods 6-14
Skip Scanning of Indexes 6-15
Identifying Unused Indexes 6-18
Enabling and Disabling the Monitoring of Index Usage 6-19
Clusters 6-20
Cluster Example 6-21
Index Clusters 6-22
Index Clusters: Performance Characteristics 6-23
Index Clusters: Limitations and Guidelines 6-24
Hash Clusters 6-25
Hash Clusters: Limitations 6-26
When to Use Clusters 6-27
Summary 6-29
7 Collecting Statistics
Objectives 7-2
The ANALYZE Command 7-3
Table Statistics 7-5
Index Statistics 7-7
Column Statistics 7-10
The DBMS_STATS Package 7-12
DBMS_STATS: Generating Statistics 7-13
Copy Statistics Between Databases 7-14
Example: Copying Statistics 7-15
Example: Gathering Statistics 7-16
Predicate Selectivity 7-17
Bind Variables and Predicate Selectivity 7-18
Histograms 7-20

v
 Histograms and Selectivity 7-21
Histogram Statistics: Example 7-22
Histogram Tips 7-24
When to Use Histograms 7-25
Choosing a Sample Size 7-26
Choosing the Number of Buckets 7-27
Viewing Histogram Statistics 7-28
Summary 7-29
Practice Overview 7-30
8 Influencing the Optimizer
Objectives 8-2
Setting the Optimizer Mode 8-3
Some Additional Parameters 8-4
Optimizer Hint Syntax 8-6
Rules for Hints 8-7
Hint Recommendations 8-8
Optimizer Hint Example 8-9
Hint Categories 8-10
Basic Access Path Hints 8-11
Advanced Access Path Hints 8-13
Buffer Cache Hints 8-14
Hints and Views 8-15
View Processing Hints 8-17
Summary 8-18
Practice Overview 8-19
9 Sorting and Joining
Objectives 9-2
Tuning Sort Performance 9-3
Top-N SQL 9-4
Join Terminology 9-5
Join Operations 9-7
Nested Loops Joins 9-8
Nested Loops Join Plan 9-9
Sort/Merge Joins 9-10
Sort/Merge Join Plan 9-11
Hash Joins 9-12
Hash Join Plan 9-13
Joining Multiple Tables 9-14
Outer Joins 9-15
SQL: 1999 Outer Joins 9-16
Full Outer Joins 9-17
Execution of Outer Joins 9-18
The Optimizer and Joins 9-19
Join Order Rules 9-20

vi
 RBO Join Optimization 9-21
CBO Join Optimization 9-23
Estimating Join Costs 9-24
Star Joins 9-25
Hints for Join Orders 9-27
Hints for Join Operations 9-28
Other Join Hints 9-30
Subqueries and Joins 9-31
Initialization Parameters that Influence Joins 9-33
Throwaway of Rows 9-34
Minimize Throwaway of Rows 9-35
Minimize Processing 9-36
Summary 9-37
10 Optimizer Plan Stability
Objectives 10-2
Optimizer Plan Stability 10-3
Plan Equivalence 10-4
Creating Stored Outlines 10-5
Using Stored Outlines 10-6
Data Dictionary Information 10-7
Execution Plan Logic 10-8
Maintaining Stored Outlines 10-9
Outline Editing Overview 10-11
Editable Attributes 10-13
Outline Cloning 10-14
Outline: Administration and Security 10-15
Configuration Parameters 10-17
Create Outline Syntax 10-18
Outline Cloning Examples 10-19
Summary 10-22
Practice Overview 10-23
11 Advanced Indexes
Objectives 11-2
Bitmapped Indexes 11-3
Bitmapped Index Structure 11-4
Creating Bitmapped Indexes 11-5
Using Bitmapped Indexes for Queries 11-7
Combining Bitmapped Indexes 11-8
When to Use Bitmapped Indexes 11-9
Advantages of Bitmapped Indexes 11-10
Bitmapped Index Guidelines 11-11

vii
 What Is a Bitmap Join Index? 11-12
Bitmap Join Index: Advantages and Disadvantages 11-14
Indexes and Row-Access Methods 11-16
Index Hints 11-17
INDEX_COMBINE Hint Example 11-18
Star Transformation 11-20
Star Transformation Example 11-22
Function-Based Indexes 11-24
Function-Based Indexes: Usage 11-25
Data Dictionary Information 11-26
Summary 11-27
12 Materialized Views and Temporary Tables
Objectives 12-2
Materialized Views 12-3
Create Materialized Views 12-4
Refresh Materialized Views 12-5
Materialized Views: Manual Refresh 12-7
Query Rewrites 12-8
Create Materialized Views: Syntax Options 12-11
Enabling and Controlling Query Rewrites 12-12
Query Rewrite Example 12-13
Dimensions: Overview 12-15
Dimensions and Hierarchies 12-16
Dimensions: Example Table 12-17
Dimensions and Hierarchies 12-18
Create Dimensions and Hierarchies 12-19
Dimensions Based on Multiple Tables 12-20
Dimensions with Multiple Hierarchies 12-21
Temporary Tables 12-22
Creating Temporary Tables 12-24
Summary 12-26
13 Alternative Storage Techniques
Objectives 13-2
Storing User Data 13-3
Index-Organized Tables 13-4
IOT Performance Characteristics 13-5
IOT Limitations 13-6
When to Use Index-Organized Tables 13-7
Creating Index-Organized Tables 13-8
IOT Row Overflow 13-9
Retrieving IOT Information 13-10
External Tables 13-11
External Tables Performance Characteristics 13-12

viii
 External Tables Limitations 13-13
Why Use External Tables? 13-14
Creating External Tables 13-15
Retrieving External Tables Information 13-16
Summary 13-17

14 Data Warehousing Considerations

Objectives 14-2
The WITH Clause: Overview 14-3
The WITH Clause 14-4
The WITH Clause: Example 14-5
The WITH Clause: Implementation 14-6
The WITH Clause: Usage Notes 14-7
The WITH Clause: Benefits 14-8
Multitable INSERT Statements 14-9
Multiple INSERT Statements 14-11
Advantages of Multitable INSERTs 14-12
MERGE Statements 14-13
Merge Syntax 14-14
Example of Using the MERGE Statement in Data Warehousing 14-15
Summary 14-16

Appendix A: Workshops

Appendix B: Diagnostic Tools Reference

Appendix C: Table Descriptions

Appendix D: Workshop Scripts

Appendix E: Practice Solutions

Index

ix
x
Instructor
Preface
Instructor Preface - 2
How This Book Is Organized

Unit 1: Introduction

Lesson Aim
Lesson 1: Following a Tuning This lesson covers the process of SQL statement tuning
Methodology including defining a tuning methodology. By applying a
tuning methodology, students learn to optimize
performance, improve response time and scalability.
Tuning is an ongoing process in the life cycle of a project
that needs to start in the development phase and
continue through the maintenance phase.
Lesson 2: SQL Statement Processing Students learn the four main stages in processing a SQL
statement: Parse, Bind, Execute, and Fetch. The
advantages and disadvantages of using bind variables are
discussed.

Unit 2: Diagnostic Tools

Lesson Aim
Lesson 3: EXPLAIN and AUTOTRACE This unit shows students the tools which are available to
help test and diagnose performance problems with a
SQL statement; this lesson shows the Explain command
and the SQL*Plus AUTOTRACE feature.
Lesson 4: SQL Trace and TKPROF This lesson shows the SQL Trace facility. By using SQL
Trace, students can evaluate the performance of SQL
statements. Trace files can be formatted with TKPROF.

Unit 3: The Oracle Optimizer

Lesson Aim
Lesson 5: Rule-Based Optimization Students learn to use both rule-based and cost-based
Versus Cost-Based Optimization optimization. Rule-based optimization is still supported
for backward compatibility and uses a ranking scheme.
Cost-based optimization was introduced in Oracle7. The
optimizer uses statistics to calculate the cost of access
paths, taking into account the number of logical I/Os,
CPU usage, and network traffic.
Lesson 6: Indexes and Basic Access This lesson introduces students to B*-tree indexes. An
Methods index is a database object that is logically and physically
independent of the table data. The Oracle9i Server can
use an index to access data required by a SQL statement,
or use indexes to enforce integrity constraints.

Instructor Preface - 3
Unit 3: The Oracle Optimizer (continued)

Lesson Aim
Lesson 7: Collecting Statistics This lesson covers table statistics used with the cost-
based optimizer. By using the ANALYZE command or
DBMS_STATS package, students learn that they can
collect statistics on tables which are used by the cost-
based optimizer to determine optimal execution plans
for SQL statements. Histograms are covered too.
Lesson 8: Influencing the Optimizer This lesson covers how to influence the Oracle
optimizer at three levels: the instance level, the session
level, and the statement level (using hints).
Lesson 9: Sorting and Joining This lesson covers Top-N SQL, performance issues with
sorting, and how to optimize join statements.
Lesson 10: Optimizer Plan Stability This lesson covers the use of stored outlines and its
influence on optimizer plan stability, and the use of the
OUTLN_PKG package on stored outlines.

Unit 4: Advanced Data Access and Storage

Lesson Aim
Lesson 11: Advanced Indexes Students learn about bitmapped indexes, star
transformations, and function-based indexes.
Lesson 12: Materialized Views and This lesson shows students materialized views, query
Temporary Tables rewrites, dimensions and hierarchies, and temporary
tables.
Lesson 13: Alternative Storage This lesson shows some alternative storage techniques
Techniques including index-organized tables, index clusters, and
hash clusters.
Lesson 14: Data Warehousing This lesson reviews the efficiencies introduced with the
Considerations Oracle9i Server that enhance performance in data
warehousing environments. The commands WITH,
multitable INSERT, and MERGE statements are
reviewed from a performance perspective.

Instructor Preface - 4
 Following a Tuning Methodology

Copyright © Oracle Corporation, 2002. All rights reserved.

Schedule: Timing Topic

40 minutes Lecture
40 minutes Total
Instructor Note
Make sure to mention the suggested prerequisite for this course: Oracle9i DBA Fundamentals
I. Do not offer too much detail about indexing and clustering techniques; just use them, and
measure performance. That is the main objective of this course. Also, the main target audience
for this course is developers, not database administrators.
Do not explain and demonstrate everything while teaching the lessons; the students should
discover certain things while participating in the workshops.
The first lesson provides participants with a framework for the course. It explains why tuning
is necessary, putting tuning firmly in the realm of solving business problems. It also presents a
tuning methodology and shows how the topic of this course fits into that methodology.
 Objectives

After completing this lesson, you should be able to

do the following:
• State the procedural steps in managing
performance
• Describe the causes of performance problems
• Identify the main system areas that you can
address by using the tuning process
• Describe the tuning methodology
• Explain the advantage of following the tuning
methodology in its proper order
• List the tuning steps that are the responsibility of
the application developer

Copyright © Oracle Corporation, 2002. All rights reserved.

Objectives
Applying a tuning methodology to your design and tuning efforts optimizes performance.
Tuning SQL is a major component of a tuning methodology.
After completing this lesson, you should be able to:
• State the procedural steps in managing performance
• Describe the causes of performance problems
• Identify the main system areas that you can address by using the tuning process
• Describe the tuning methodology
• Explain the advantage of following the tuning methodology in its proper order
• List the tuning steps that are the responsibility of the application developer

Oracle9i: SQL Tuning Workshop 1-2

 Overview

• Performance management
• Performance problems
• The tuning methodology
• SQL statement tuning
• Methodology application

Copyright © Oracle Corporation, 2002. All rights reserved.

Overview
Anyone involved in tuning should follow a tuning methodology to achieve maximum
performance. Tuning SQL statements is an important step that is least expensive when
performed at its proper moment within the methodology.
In addition to tuning at the right time, you should also have a good understanding of the issues
involved in performance management and the types of performance problems that might arise.

Oracle9i: SQL Tuning Workshop 1-3

 Managing Performance

• Start early
• Set objectives
• Tune and monitor conformance
• Work together
• Handle exceptions and changes

Copyright © Oracle Corporation, 2002. All rights reserved.

Managing Performance
Tuning requires several steps.
Start Early
Performance management must span the application or project continuously to be fully
effective. You should consider it at the design stage.
Set Objectives
Do not attempt to tune everything. Use the ROI (return on investment) methodology to identify
the portions of the application or project to tune. Define your objectives in a form accepted and
agreed to by all the interested parties. Service level agreements (SLAs) form an increasingly
important part of setting the objectives of operations groups and facilities management teams.
You can successfully extend this concept into the specification of application requirements.
It is better to set a measurable objective such as “this report will complete in 5 seconds or less”
than to aim for “all queries will be faster.”
Tune and Monitor Conformance
After you have set out and agreed upon the objectives, you are ready to tune and to reach those
objectives, monitoring your progress as you tune. You should keep detailed records about the
level of conformance with the requirements. You should publish indicative measures at regular
intervals, highlighting any deviations or trends.

Oracle9i: SQL Tuning Workshop 1-4

Managing Performance (continued)
Work Together
Database administrators, system administrators, and programmers should work together as a
team, not as adversaries.
Handle Exceptions and Changes
When effective monitoring is in place and supported at the same level as the rest of the
application, you must react promptly to any changes when they are reported. Use the
available data to select a course of action and then analyze the resulting performance to
determine whether your actions are successful. If overall changes in use patterns or
equipment capability occur, then you should consider setting new objectives.
The 80/20 Rule
Typically, 20% of your transactions (including select statements) account for 80% of the
system usage. The transactions that make up this 20% are the ones that must be tuned, and
tuning anything else is probably not worth your time and energy.
You can also formulate this phenomenon with the following corollary: 50% of the
utilization comes from 5% of your transactions.

Oracle9i: SQL Tuning Workshop 1-5

 Factors to Be Managed

• Schema
– Data design
– Indexes
• Application
– SQL statements
– Procedural code
• Instance
• Database
• User expectations

Copyright © Oracle Corporation, 2002. All rights reserved.

Instructor Note
Point out to the participants that:
• Schema design, as covered in Oracle9i Database Fundamentals for Developers, is a
suggested prerequisite for this course.
• Tuning SQL statements is the focus of this course.
• Tuning procedural code (PL/SQL, 3GLs) is part of learning that particular coding
environment.
• Instance and database tuning are discussed in the Oracle9i Performance Tuning course.
Refer to them briefly, when necessary.

Oracle9i: SQL Tuning Workshop 1-6

Factors to Be Managed
Performance management can be divided into the following four areas. Although the areas are
separate, they are also interdependent and require different skill sets.
• Schema tuning deals with the physical structure of the data.
• Application tuning deals with the business functions and the program modules that
implement the functions. Tuning the procedural code for the type of application and
tuning the embedded SQL statements are also included in this area.
• Instance tuning deals with the installation of the Oracle9i Server and how it uses memory.
• Database tuning deals with managing the physical arrangement of data on the disk.
This course concentrates on schema and SQL statement tuning. Other parts of application
tuning should be part of learning that particular type of application. Instance and database
tuning are covered in courses and literature for database administrators.
Schema
If an application has inadequate or inappropriate data design, then tuning the physical
allocation, providing indexes, or rewriting programs will not overcome the problem.
SQL Statements
If an application is well designed, it may still perform badly. A common reason for this is
badly written SQL.
User Expectations
Users expect consistent performance. They can cope with slower application functions if they
understand why the application is slower than usual. The project team should try to build a
realistic user expectation of performance, possibly including application messages to warn
operators that they are requesting operations that are resource-hungry. The best time to do this
is before the design and build phases and as part of the transition phase.

Oracle9i: SQL Tuning Workshop 1-7

 Performance Problems

• Inadequate consumable resources

– CPU
– I/O
– Memory (may be detected as an I/O problem)
– Data communications resources
• Design limitations
• Locking

Copyright © Oracle Corporation, 2002. All rights reserved.

Instructor Note
Before turning to the next page, you could ask for suggestions about performance
dependencies and write them on the board.

Oracle9i: SQL Tuning Workshop 1-8

Performance Problems
Performance problems occur when a function takes longer to perform than the time allowed.
This is because a resource of a particular type is insufficient or inadequate. The resource may
be a physical resource, such as available memory buffers that store data blocks, or an artificial
resource, such as a lock.
Inadequate Consumable Resources
A resource may simply be inadequate to meet the need under any circumstances. For example,
if you want a function to complete in under one second, a network with a message turnaround
time of two seconds will never meet the target.
If the limiting factor is a consumable resource, such as CPU power, all the users of that
resource are affected.
Inadequate Design Resources
If the limiting factor is the contention of processes for a design resource, such as a lock, then
only users of those specific processes are likely to be affected.
Locking
Contention due to locking by other transactions or applications might be a problem.

Oracle9i: SQL Tuning Workshop 1-9

 Critical Resource

• Performance depends on the following:

– How many clients need the resource
– How long they must wait for it
– How long they hold it
• Consider limiting demand to maintain acceptable
response times.

Copyright © Oracle Corporation, 2002. All rights reserved.

Critical Resource
Response time is defined as the service time plus the wait time to accomplish a certain task.
As demand for a resource with a single server increases toward the service rate, queues build up
with the queue length, increasing exponentially for each increase in demand.
Even if there are many servers, you can observe the same effect with a single queue. However,
multiple servers provide a valuable smoothing effect when there is a wide variation in the time
for which a resource is occupied by one of its clients before it is available for reallocation.
The goal is to design, engineer, and tune the system so that the load is never permitted to slow
service completion times below the acceptable level.

Oracle9i: SQL Tuning Workshop 1-10

 Excessive Demand

• Greatly increases response time and reduces

throughput
• Should be prevented as much as possible by
limiting demand to a level that still allows
reasonable throughput

Copyright © Oracle Corporation, 2002. All rights reserved.

Excessive Demand
Throughput is defined as the total amount of work accomplished by the system in a given
amount of time. Too many processes using a system simultaneously may result in the
following symptoms.
Increased Response Time
Most users know and understand the effects of queues on increasing response time. They may
be prepared to accept slower response at peak times if the effect is linear. However, both
statistical theory and experience show that when response time starts to deteriorate, small
increases in load can have a severe effect, which is unacceptable to users.

Oracle9i: SQL Tuning Workshop 1-11

Excessive Demand (continued)
Reduced Throughput
Any marked degradation in response time is likely to disrupt the work rate of the users who are
affected. Many people do not understand that adding more users to a system significantly
decreases the overall throughput of the system.
If system throughput is important, you must ensure that the number of users does not exceed the
threshold at which the throughput starts to decline. It is best to limit the number of users
mechanically.
This may force a change in working patterns to cope with the restriction, but spreading the peak
periods across the working day can improve the way the system is used.

Oracle9i: SQL Tuning Workshop 1-12

 Understanding Scalability

• A system’s ability to process more workload, with

a proportional increase in system resource usage
• Aimed at application designers and performance
specialists
• Poor scalability leads to system resource
exhaustion to the point where no more throughput
is possible when its workload is increased.

Copyright © Oracle Corporation, 2002. All rights reserved.

Scalability
Scalability is a system’s ability to process more workload, with a proportional increase in
system resource usage. In other words, in a scalable system, if you double the workload, then
the system must use twice as many system resources.
Examples of bad scalability due to resource conflicts include the following:
• Applications requiring significant concurrency management as user populations increase
• Increased locking activities
• Increased data consistency workload
• Increased operating system workload
• Transactions requiring increases in data access as data volumes increase
• Poor SQL and index design resulting in a higher number of logical I/Os for the same
number of rows returned
• Reduced availability because database objects take longer to maintain

Oracle9i: SQL Tuning Workshop 1-13

 Understanding Scalability

Resource exhaustion due to:

• Hardware exhaustion
• Table scans in high-volume transactions, causing
inevitable disk I/O shortages
• Excessive network requests, resulting in network
and scheduling bottlenecks
• Memory allocation, causing paging and swapping
• Excessive process and thread allocation, causing
operating system thrashing

Copyright © Oracle Corporation, 2002. All rights reserved.

Understanding Scalability
If an application exhausts a system resource to the point where no more throughput is possible
when its workload is increased it is said to be unscalable. This can result in fixed throughputs
and poor response times.

Oracle9i: SQL Tuning Workshop 1-14

 Internet Scalability

Characteristics of Internet applications:

• Availability 24 hours a day, 365 days a year
• An unpredictable and imprecise number of
concurrent users
• Difficulty in capacity planning
• Availability for any type of query
• Multitier architectures
• Stateless middleware
• Rapid development timescale
• Minimal time for testing

Copyright © Oracle Corporation, 2002. All rights reserved.

Internet Scalability
Applications must scale with the increase in workload and also when additional hardware is
added to support increasing demand. Design errors can cause the implementation to reach its
maximum, regardless of additional hardware resources or redesign efforts.
Internet applications are challenged by very short development timeframes with limited time
for testing and evaluation. A badly designed application generally means that at some point in
the future, the system will need to be rearchitected or reimplemented. From a business
perspective, poor performance can mean a loss of customers. If a Web user does not get a
response in seven seconds, then the user’s attention could be lost forever.
In many cases, the cost of redesigning a system with the associated down time costs in
migrating to new implementations exceeds the costs of properly building the original system.
It is critical to design and implement with scalability in mind from the start.

Oracle9i: SQL Tuning Workshop 1-15

 Scalability with Application Design,
Implementation, and Configuration

Applications have a large impact on scalability.

• Poor schema design can cause expensive SQL that
does not scale.
• Poor transaction design can cause locking and
serialization problems.
• Poor connection management can cause poor
response times and unreliable systems.

Copyright © Oracle Corporation, 2002. All rights reserved.

Scalability with Application Design, Implementation, and Configuration

Poor application design, implementation, and configuration have a large impact on scalability.
However, the design is not the only problem. The physical implementation of the application
can be the weak link. For example:
• Systems can move to production environments with bad I/O strategies.
• The production environment could use different execution plans than those generated in
testing.
• Memory-intensive applications that allocate a large amount of memory without much
thought for freeing the memory at run time can cause excessive memory usage.
• Inefficient memory usage and memory leaks put a high stress on the operating virtual
memory subsystem. This impacts performance and availability.

Oracle9i: SQL Tuning Workshop 1-16

 The Tuning Methodology

1. Tune business function

2. Tune data design
3. Tune process design
4. Tune SQL statements
5. Tune physical structure
6. Tune memory allocation
7. Tune I/O
8. Tune memory contention
9. Tune operating system

Copyright © Oracle Corporation, 2002. All rights reserved.

The Tuning Methodology

The list in the slide shows the phases of the development cycle in which tuning can be applied.
Following the steps in this order is recommended for the following reasons:
• The earlier the step, the higher the potential for improving performance.
• The earlier the step, the higher the cost for performing or reworking it later. For example,
changes to the data structures and application code after the initial design tend to be
expensive and require the additional expense of retesting the affected components.
• Decisions made at one step can influence subsequent steps. For example, the SQL
statement that you write in Step 4 will have significant bearing on parsing and caching
issues that are addressed in Step 6.
The more extensively that you use object-oriented design techniques and multitiered
architecture, the better the chances that you will safely achieve any application changes at a
reasonable cost.
Instructor Note
About object-oriented design techniques: on the other hand, encapsulation tends to make you
lose control over events occurring at a lower level.

Oracle9i: SQL Tuning Workshop 1-17

 Tuning Roles

Business analyst 1. Tune business function

Designer 2. Tune data design

3. Tune process design

Application developer 4. Tune SQL statements

5. Tune physical structure

Database administrator 6. Tune memory allocation

7. Tune I/O
8. Tune memory contention

Operating system administrator 9. Tune operating system

Network administrator 10. Networking issues

Copyright © Oracle Corporation, 2002. All rights reserved.

Tuning Roles
The business analyst, designer, application developer, database administrator, and operating
system administrator are responsible for different steps in the tuning process. In some cases, one
person may fulfill several of these roles.
The steps performed by the database administrator and operating system administrator have a
lesser effect on performance than earlier steps, but they can be performed at relatively low cost
with immediately available and observable results.
The fourth step in the methodology is primarily the responsibility of the application developer.
However, understanding how to tune SQL can enable designers to design schemas that tune
easily. With this understanding, database administrators can help pinpoint SQL tuning needs
and solutions, thus easing the burden on their databases. This is especially useful if the
application is already in production and the application developer is no longer available.
A large part of tuning in client-server or Web environments relates to network issues, as well.

Oracle9i: SQL Tuning Workshop 1-18

 Tuning SQL Statements

Use performance analysis tools to check the results

of the following steps:
1. Tune the schema.
– Add indexes as required
– Consider index-organized tables vs. heap-organized
tables in certain situations
– Create clusters if needed
2. Choose the language: SQL, PL/SQL, or Java.

Copyright © Oracle Corporation, 2002. All rights reserved.

Tuning SQL Statements

During the steps within SQL statement tuning, analysis techniques should be used frequently to
determine goals and progress.
1. Tuning the Schema
The designer is responsible for determining which tables should be used, their contents, and
similar issues. The application developer may then need to decide when to use indexing,
when to use index-organized tables, and when to use clustering. Denormalization may be
needed for good performance, especially in data warehouse environments. The decisions that
are made at this step greatly affect the performance of any SQL statements that use these
structures.
2. Choosing a Language: SQL, PL/SQL, or Java
In some cases, you can achieve better performance by using the PL/SQL or Java language to
perform a task.

Oracle9i: SQL Tuning Workshop 1-19

 Tuning SQL Statements

3. Design for the reuse of SQL optimization.

4. Design and tune SQL statements.
5. Maximize performance with the optimizer.

Copyright © Oracle Corporation, 2002. All rights reserved.

Tuning SQL Statements (continued)

3. Designing for Reuse of SQL Parsing, Optimization, and Tuning Efforts
The Oracle9i Server is often able to reuse some of its efforts at parsing and optimization
when it sees an identical statement repeated. Therefore, creating SQL statements that appear
identical can improve performance. It also enables the application developer to focus and
tune individual SQL statements that are used repeatedly.
4. Designing and Tuning SQL Statements
There is a wide variety of methods for designing high-performance SQL statements.
Knowing the overall functions of the optimizer helps to show where tuning efforts can be
effective. Consider rewriting SQL statements into semantically equivalent statements. Also,
use the correct language (SQL or PL/SQL) in each situation.
5. Achieving Maximum Performance with the Optimizer
To make the best use of the Oracle cost-based optimizer, you must understand how it
chooses access paths to data. You must assist the cost-based optimizer by using the
ANALYZE command and by sometimes providing it hints about the best access path. The
rule-based optimizer is also available. Both the cost-based optimizer and the rule-based
optimizer will be discussed in the “Rule-Based Optimization Versus Cost-Based
Optimization” lesson.

Oracle9i: SQL Tuning Workshop 1-20

 Applying the Methodology

• Be proactive: work from the top down.

• If you must be reactive, work from the bottom up,
using these hints:
– Establish quantifiable objectives.
– Create a minimum repeatable test.
– Ask questions of affected users, and avoid
preconceptions.
– Test hypotheses and keep notes.
– Stop when you meet the target.

Copyright © Oracle Corporation, 2002. All rights reserved.

Quantifiable Objectives
Establish quantifiable objectives and avoid vague ones. An example of an acceptable objective
would be the following:
“We must be able to support up to 20 operators, each entering 20 orders per hour, and the
picking lists must be produced within 30 minutes of the end of the shift.”
Minimum Repeatable Test
If you are trying to cut a four-hour run down to two hours, repeated timings take too long.
Perform your initial trials against a test environment similar to the real one (or provide input
for a restrictive condition, such as processing one department instead of 500 of them). The
ideal test case should run for more than one minute so that improvements are demonstrated
intuitively and can be measured using timing features.
• Ask questions of affected users and avoid preconceptions.
• Test hypotheses and keep notes.
• Stop when you meet the target.

Oracle9i: SQL Tuning Workshop 1-21

Tuning a Specific Problem for an Application Already in Production
Often, performance specialists are summoned late in the life cycle of a project, when the
system is in production and performing unacceptably.
In this situation, start at the bottom of the method list and work up. The items at the end are
the cheapest and fastest to perform. If they do not solve the problem, you will have to work
back “uphill,” and this starts to increase costs and time delays.

Oracle9i: SQL Tuning Workshop 1-22

 Summary

In this lesson, you should have learned how to:

• Manage performance
– Start early: be proactive
– Set measurable objectives
– Monitor requirements conformance
– Handle exceptions and changes
• Identify performance problems
– Inadequate consumable resources
– Inadequate design resources
– Critical resources
– Excessive demand

Copyright © Oracle Corporation, 2002. All rights reserved.

Summary
This lesson introduced you to following a tuning methodology. By applying a tuning
methodology, you can optimize performance, and improve response time and scalability.
Tuning is an on going process in the life cycle of a project that must start in the development
phase and continue through the maintenance phase.
• Manage performance at an early stage
• Identify performance problems
• Determine tuning roles
• Tune your SQL statements
• Apply the tuning methodology

Oracle9i: SQL Tuning Workshop 1-23

 Summary

• Tune SQL statements

– Manage user expectations
– Analyze the results at each step
– Tune the physical schema
– Choose when to use SQL
– Reuse SQL statements when possible
– Design and tune the SQL statement
– Get maximum performance with the optimizer

Copyright © Oracle Corporation, 2002. All rights reserved.

Summary (continued)
Tuning SQL Statements
Tuning SQL statements involves using analysis tools, SQL tuning techniques, and the optimizer
effectively. In addition, tuning the schema can create additional access paths for the optimizer,
such as the use of an index.

Oracle9i: SQL Tuning Workshop 1-24

 SQL Statement Processing

Copyright © Oracle Corporation, 2002. All rights reserved.

Schedule: Timing Topic

60 minutes Lecture
15 minutes Practice
75 minutes Total
Instructor Note
This lesson provides insight into the way SQL statements are processed by the Oracle9i Server.
It explains the Parse, Bind, Execute, and Fetch phases, concentrating on minimizing parsing and
optimizing shared cursor usage. It also discusses bind variables in that perspective.
 Objectives

After completing this lesson, you should be able to

do the following:
• Describe the basic steps involved in processing a
SQL statement
• Monitor the use of shared SQL areas
• Write SQL statements to take advantage of shared
SQL areas
• Understand how to use the CURSOR_SHARING
parameter
• Use automatic PGA memory management

Copyright © Oracle Corporation, 2002. All rights reserved.

Objectives
Knowing how the Oracle9i Server processes SQL statements can help you design SQL
statements that can reuse parsing and optimization efforts.
After completing this lesson, you should be able to:
• Describe the basic steps involved in processing a SQL statement
• Monitor the use of shared SQL areas
• Write a SQL statement to take advantage of shared SQL areas (This will enable you to
reuse parsing and optimization efforts).
• Understand how and when to set the CURSOR_SHARING parameter to SIMILAR or
FORCE.
• Use the automatic PGA memory management.

Oracle9i: SQL Tuning Workshop 2-2

 Overview

• System Global Area (SGA)

• Shared SQL areas
• Program Global Area (PGA)
• SQL processing phases
• SQL coding standards
• Shared cursors
• Automatic PGA memory management

Copyright © Oracle Corporation, 2002. All rights reserved.

Overview
By knowing about shared SQL areas, the SQL processing phases, and shared cursors, you can
understand how coding standards enable you to minimize how often your statements must be
parsed and optimized.
Additionally, as the statements you write become more and more standardized, you can
identify statements that occur frequently and devote additional tuning to them.

Oracle9i: SQL Tuning Workshop 2-3

 System Global Area (SGA)
Instance
SGA Shared pool
DB buffer Redo log
cache buffer
Server
process

PGA

Data Control Redo log

User Parameter files files files
process file
Archived
Program log files
unit Password
file
Database

Copyright © Oracle Corporation, 2002. All rights reserved.

System Global Area

The System Global Area (SGA) is a memory area used to store database information that is
shared by database processes. It is allocated in the virtual memory of the computer where the
Oracle server resides. The SGA contains several subareas that are used to organize and hold
different types of information. The Oracle server uses memory to store various types of
information, such as program code being executed, information about connected sessions,
SQL statements, and so on. One area is the program global area (PGA), another is the shared
pool. Various processes manage this area of memory.
Instructor Note (for page 2-5)
A parse tree is a structure in memory where each node on the tree represents processing
instructions for a step in the execution plan. In executing the plan, the processor follows these
nodes sometimes recursively.

Oracle9i: SQL Tuning Workshop 2-4

 The Shared Pool

Contains:
• Data dictionary cache
• Library cache
– SQL statements
– Parsed or compiled PL/SQL blocks Application
Code
– Java classes

Copyright © Oracle Corporation, 2002. All rights reserved.

The Shared Pool

The shared pool is a portion of the system global area that contains shared memory constructs
such as shared SQL areas, the data dictionary cache, and the fully parsed or compiled
representations of PL/SQL blocks.
The main components of the shared pool are the library cache and the dictionary cache. The
library cache stores the executable (parsed or compiled) form of recently referenced SQL and
PL/SQL code. The dictionary cache stores information such as usernames, segment
information, tablespace information, and sequence numbers.
• A shared SQL area is required to process every unique SQL statement submitted to a
database.
• A shared SQL area contains information such as the parse tree and execution plan for the
corresponding statement.
• A single shared SQL area is used by multiple applications that issue the same statement,
and therefore needs fewer hard parses which results in performance improvement as well
as leaving more shared memory for other uses.
The database administrator (DBA) can change the available dictionary and shared SQL areas
space by modifying the SHARED_POOL_SIZE initialization parameter. The DBA does this
as part of an overall database tuning effort, which is subject to various constraints discussed in
the Oracle9i Performance Tuning course.

Oracle9i: SQL Tuning Workshop 2-5

 Shared SQL Areas

Cursor for Cursor for

SELECT statement 2 SELECT statement 1

Shared SQL

SGA

User A User B User C

SELECT SELECT SELECT
statement 1 statement 2 statement 1

Copyright © Oracle Corporation, 2002. All rights reserved.

Shared SQL Areas

When you invoke application code, Oracle attempts to reuse existing code if it has been
executed previously so that it can be shared. If the parsed representation of the statement does
exist in the library cache and it can be shared, Oracle reuses the existing executable.
When the shared pool in the SGA fills up, the Oracle server makes room for newly requested
programs by using a least recently used (LRU) algorithm, so that items in the shared pool that
have not been used often are aged out to make room for the new programs.
Cursors
Inside the shared SQL area, each SQL statement is parsed in its own part, known as a context
area or cursor.
Each cursor holds the following information:
• The parsed statement (static, dynamic, and recursive SQL, plus program units such as
procedures and database triggers)
• The execution plan
• A list of referenced objects
If two users issue the same SQL statement, then they will use the same cursor. The statement
is reparsed if the representation in the shared pool is invalid. This happens, for example, if a
data definition language (DDL) statement such as ALTER TABLE was used on one of the
objects in the statement, or if a dependent table is analyzed.
Oracle9i: SQL Tuning Workshop 2-6
 The Program Global Area (PGA)

• Is a memory buffer that contains data and control

information for a server process.
• Can control and tune the size of a work area.
– Optimal size
– One-pass size
– multipass size

Copyright © Oracle Corporation, 2002. All rights reserved.

PGA
A program global area (PGA) is a memory region containing data and control information for
a single process (server or background). Access to it is exclusive to that server process and is
read and written only by the Oracle code acting on behalf of it. PGA is a nonshared memory
area to which a process can write. One PGA is allocated for each server process. A PGA is
allocated by the Oracle server when a user connects to an Oracle database and a session is
created, though this varies by operating system and configuration.
An example of such information is the run-time area of a cursor. Each time a cursor is
executed, a new run-time area is created for that cursor in the PGA. With complex queries, a
big portion of the run-time area is dedicated to work areas allocated by memory-intensive
operators such as sort-based operators. A sort operator uses a work area (the sort area) to
perform the in-memory sort of a set of rows
The size of the work area can be controlled and tuned. Ideally, the size of a work area is big
enough that it can accommodate the input data and auxiliary memory structures allocated by
its associated SQL operator. This is called the optimal size of a work area. When the size of
the work area is smaller than optimal, response time increases, because an extra pass is
performed over part of the input data. This is called the one-pass size of the work area. Under
a one-pass threshold, when the size of a work area is much too small compared to the input
data size, multiple passes over the input data are needed. This is called the multipass size of
the work area.
Oracle9i: SQL Tuning Workshop 2-7
 SQL Statement Processing Phases

Open Close

Parse Bind Execute Fetch

Copyright © Oracle Corporation, 2002. All rights reserved.

Processing Phases
The four most important phases in SQL statement processing are parsing, binding, executing,
and fetching.
The reverse arrows indicate processing scenarios; for example, Fetch—(Re)Bind—Execute—
Fetch.
The Fetch phase applies only to queries and DML statements with a returning clause.
Note: A detailed description of SQL statement processing can be found in the Oracle9i
Application Developers Guide - Fundamentals, “How Oracle Processes SQL Statements” and
the Oracle9i Concepts, “SQL and PL/SQL.”
Instructor Note
There is no real Rebind phase; the same pointer is simply used for a re-Execute. The value
found at the corresponding memory address is used; Oracle9i uses a “bind by reference”
technique.
Three more phases are missing from this slide: Define, Describe, and Parallelize.
Define and Describe are not very important for tuning. Parallelize is not mentioned here
because it is discussed in the Oracle9i: Implementing Scalable Systems course.

Oracle9i: SQL Tuning Workshop 2-8

 SQL Statement Processing Phases

• Parse
– Searches for the statement in the shared pool
– Checks syntax
– Checks semantics and privileges
– Merges view definitions and subqueries
– Determines execution plan

Copyright © Oracle Corporation, 2002. All rights reserved.

Parse Phase
The Oracle9i Server does the following:
• Searches for the statement in the shared pool
• Checks the statement syntax, given the grammar and specifications of the SQL language
• Checks the semantics, ensuring that objects referenced in the SQL statement are valid
and satisfy security constraints
• Determines whether the process issuing the statement has appropriate privileges to
execute it
• Transforms a SQL statement on a view into an equivalent SQL statement on its
underlying definition, and attempts to simplify a statement with a subquery by rewriting
it into a join
• Determines and stores the execution plan, or uses an existing execution plan, if possible
Note: The Parse phase also checks whether materialized views can be used. Materialized
views are discussed in a subsequent lesson.
Instructor Note
Determining the execution plan is the most costly part of the parsing and may take longer than
executing the plan.

Oracle9i: SQL Tuning Workshop 2-9

 SQL Statement Processing Phases

• Bind
– Scans the statement for bind variables
– Assigns (or reassigns) a value

Copyright © Oracle Corporation, 2002. All rights reserved.

Bind Phase
• The Oracle9i Server checks the statement for references of bind variables.
• The Oracle9i Server assigns or reassigns a value to each variable.
Note: This phase order implies that the Oracle9i Server does not know bind variable values
when optimizing a statement. This enables a fast rebind-execute without the need for
reparsing, thus saving time and memory; a disadvantage is that it is impossible for the
optimizer to estimate predicate selectivity. This will be discussed in more detail in the
“Collecting Statistics” lesson.
Instructor Note (for page 2-11)
Note the subtle difference between sorts for DML statements (during the Execute phase) and
queries (during the first fetch). There is no need to stress this, but students might ask.
With an INSERT, UPDATE, or DELETE statement, executing the statement results in the
SQL being run immediately during the Execute phase.
When a SELECT statement where rows are sorted or locked is issued, opening the cursor
retrieves all of the rows to be returned. If there is no locking or sorting, opening the cursor
locates the record pointer at the first row.

Oracle9i: SQL Tuning Workshop 2-10

 SQL Statement Processing Phases

• Execute
– Applies the execution plan
– Performs necessary I/O and sorts for data
manipulation language (DML) statements
• Fetch
– Retrieves rows for a query
– Sorts for queries when needed
– Uses an array fetch mechanism

Copyright © Oracle Corporation, 2002. All rights reserved.

Execute Phase
• The Oracle9i Server applies the parse tree to the data buffers.
• Multiple users can share the same parse tree.
• The Oracle9i Server performs physical reads or logical reads/writes for DML statements
and also sorts the data when needed.
Fetch Phase
The Oracle9i Server retrieves rows for a SELECT statement during the Fetch phase. Each
fetch typically retrieves multiple rows, using an array fetch.
Each Oracle tool offers its own ways of influencing the array size; in SQL*Plus you do so by
using the ARRAYSIZE setting:
SQL> show arraysize
arraysize 15
SQL> set arraysize 1
With this setting, SQL*Plus will process one row at a time. The default value is 15.

Oracle9i: SQL Tuning Workshop 2-11

 Sharing Cursors: Benefits

• Reduces parsing
• Dynamically adjusts memory
• Improves memory usage

Copyright © Oracle Corporation, 2002. All rights reserved.

Shared Cursor Benefits

When a SQL statement is found in the shared SQL area, then the Parse phase is curtailed and
the existing cursor is used. These are the main benefits of sharing cursors:
• Sharing cursors reduces parsing and saves time.
• Memory dynamically adjusts to the SQL being executed.
• Memory usage may improve dramatically, even for tools that store SQL within the
application.

Oracle9i: SQL Tuning Workshop 2-12

 Sharing Cursors: Requirements

Cursors can be shared only by SQL statements that

have the following identical elements:
• Text
– Uppercase and lowercase
– White space (spaces, tabs, carriage returns)
– Comments
• Referenced objects
• Bind-variable data types

Copyright © Oracle Corporation, 2002. All rights reserved.

Shared Cursor Requirements

• Only identical SQL statements can use the same cursor.
• The text of the SQL statements must be exactly the same, including case, spaces, tabs,
carriage returns, and comments.
• The objects referenced in the SQL statements must resolve to the same objects in the
database.
• The types of the bind variables used in the SQL statements must be the same.
Note: Before sending SQL statements to the Oracle9i Server, most Oracle tools (such as
PL/SQL, the precompilers, and Oracle Developer) preprocess SQL statements to make them
as identical as possible by removing comments, squeezing white space, and converting to
uppercase or lowercase.
SQL*Plus, however, sends SQL statements to the Oracle9i Server in the same format as they
are entered.
Instructor Note (for page 2-14)
Monitoring shared SQL is discussed at the end of this lesson. You can demonstrate these four
SQL statements at the end of this lesson—not now.

Oracle9i: SQL Tuning Workshop 2-13

 Writing SQL to Share Cursors

• If the case or the amount of white space is

different, then the statements are not identical.
select * from customers where cust_id = 180;

select * from CUSTOMERS where CUST_ID = 180;

• If the objects belong to different users, then the

statements are not identical.
select * from customers where CUST_ID = 180;

select * from customers where CUST_ID = 180;

Copyright © Oracle Corporation, 2002. All rights reserved.

Writing SQL to Share Cursors

SQL statements must be identical to be able to share cursors. Note that sharing statements is
unimportant in a decision support system (DSS) environment, because most statements will be
different anyway.
Identical Case
The first two examples in the slide are not identical. Notice the case difference for the table
and column names. Because of this case difference, the statements are not identical and thus
cannot share a SQL area. Because there is case difference between the two statements, the
parser gives different internal codes (HASH_VALUEs) for them. This can be checked
watching the V$SQL view.
Identical Objects
Even when two statements look identical, if the objects actually refer to different database
objects, then the two statements are not identical.
In the last two examples in the slide, the statements are issued by two different users who each
have their own CUSTOMERS table. Thus the statements are not identical and cannot share a
SQL area.

Oracle9i: SQL Tuning Workshop 2-14

 Bind Variables and Shared Cursors

• Suppose you enter the following two statements:

select * from customers where cust_id = :c

select * from customers where cust_id = :d

• Both statements are translated into:

select * from customers where cust_id = :b1

Copyright © Oracle Corporation, 2002. All rights reserved.

Bind Variables
If two bind variables have different data types, then the statements are not identical. If the
bind-variable data types match but their names are not identical, as in the example above,
there is no problem, because bind variables will be renamed internally. The first variable is
always called :b1, the second is :b2, and so on.
Instructor Note
Demo scripts: demo02_01.sql
demo02_01cursors.sql
Note that the script actually uses the PL/SQL environment of SQL*Plus. SQL*Plus itself does
not translate bind-variable references in SQL statements.
Note that starting with version 8.1.6, there is a powerful feature to force cursor sharing, which
you can use to set CURSOR_SHARING=FORCE at the session or instance level. This will
replace all literals in your SQL statements by system-generated bind variables. This option is
discussed at the end of the lesson.

Oracle9i: SQL Tuning Workshop 2-15

 Writing SQL to Share Cursors

Create generic code using the following:

• Stored procedures and packages
• Database triggers
• Referenced Oracle Forms Services procedures
• Any other library routines and procedures

Copyright © Oracle Corporation, 2002. All rights reserved.

Writing SQL to Share Cursors

Develop coding conventions for SQL statements in ad hoc queries, SQL scripts, and Oracle
Call Interface (OCI) calls.
Using Generic Shared Code
• Write and store procedures that can be shared across applications.
• Use database triggers.
• Write referenced triggers and procedures when using Oracle Developer.
• Write library routines and procedures in other environments.

Oracle9i: SQL Tuning Workshop 2-16

 Writing SQL to Share Cursors

Follow SQL coding standards:

• Case
• White space
• Comments
• Object references
• Bind variables

Copyright © Oracle Corporation, 2002. All rights reserved.

Writing SQL to Share Cursors (continued)

Writing to Format Standards
• Develop format standards for all statements, including those in PL/SQL code.
• Develop rules for use of uppercase and lowercase.
• Develop rules for use of white space (spaces, tabs, carriage returns).
• Develop rules for use of comments, preferably keeping them out of the SQL statements
themselves.
• Use the same names to refer to identical database objects.

Oracle9i: SQL Tuning Workshop 2-17

 Monitoring Shared Cursors

V$LIBRARYCACHE provides general information

about the library cache.
• Information about individual SQL statements is
contained in the following views:
– V$SQLTEXT
– V$SQLTEXT_WITH_NEWLINES
– V$SQL_BIND_DATA
– V$SQL and V$SQLAREA

Copyright © Oracle Corporation, 2002. All rights reserved.

Monitoring Shared Cursors

When there is no room to parse a statement in the shared SQL area, the oldest cursor is closed
and its space is reused. If the original statement is needed again, the server must parse it again.
The shared pool should be large enough to keep the number of statements that are parsed
more than once to a minimum.
You can monitor your system to see how often the server cannot find a statement in memory.
You can use Oracle Enterprise Manager (OEM) or query the appropriate data dictionary
views: V$LIBRARYCACHE and V$SQLAREA.
Instructor Note
Querying V$ views is usually a DBA task. Tell the participants that they may not have access
to these data dictionary views on their home systems. The following pages give examples of
V$LIBRARYCACHE and V$SQLAREA usage only.
Another interesting view is V$DB_OBJECT_CACHE.
Note that V$SQL lists all copies of identical SQL statements (child cursors) on a separate line,
whereas V$SQLAREA groups those statements together.
Do not go into too much detail here. Use these pages to show shared SQL. The Oracle9i
Performance Tuning course covers this in more detail.

Oracle9i: SQL Tuning Workshop 2-18

 V$LIBRARYCACHE Columns

NAMESPACE The name of the library cache area

GETS Total number of requests (lookups)

GETHITS The number of times that an object’s handle

was found in memory

GETHITRATIO The ratio of GETHITS to GETS

PINS The number of objects in the library cache

The number of times that all the pieces of the

PINHITS
object were found in memory

PINHITRATIO The ratio of PINHITS to PINS

RELOADS The number of library cache misses

Copyright © Oracle Corporation, 2002. All rights reserved.

The V$LIBRARYCACHE View

This view holds information about library cache management. Values for PINHITRATIO
and GETHITRATIO close to 1 indicate a good library cache performance. Note that this slide
does not show all columns of the V$LIBRARYCACHE view.
The V$LIBRARYCACHE view is used in the query below to check on the amount of caching.
select gethitratio, pinhitratio
from v$librarycache
where namespace = ’SQL AREA’;
GETHITRATIO PINHITRATIO
----------- -----------
.968244859 .997191864
Low GETHITRATIO could indicate that objects are getting swapped out of memory. Low
PINHITRATIO could indicate that the session is not executing the same cursor multiple
times (even though it might be shared across different sessions), or the session is not finding
the cursor shared.
Instructor Note
Demo script: demo02_02.sql
Instead of the PINHITRATIO, you could also use sum(reloads)/sum(pins) ratio to determine
when to increase the size of the shared pool.
Oracle9i: SQL Tuning Workshop 2-19
 V$SQLAREA Columns

SQL_TEXT Text of the SQL statement

VERSION_COUNT Number of versions of this cursor

LOADS Number of times the cursor has been loaded

INVALIDATIONS Number of times the contents have been

invalidated

PARSE_CALLS Number of times a user has called this cursor

SORTS Number of sorts performed by the statement

COMMAND_TYPE Command type

PARSING_USER_ID Parsing user ID (SYS = 0)

Copyright © Oracle Corporation, 2002. All rights reserved.

Shared Cursor Use

The V$SQLAREA view holds information about all shared cursors in the cache.
VERSION COUNT >1 Indicates that the same text is used by different users
on their own version of a table
LOADS > 1 Indicates cursor reloads after aging out or cursor invalidation
COMMAND_TYPE 1: CREATE TABLE
2: INSERT
3: SELECT
6: UPDATE
7: DELETE
Note: Only the most important columns of the V$SQLAREA view are listed above.
Instructor Note
V$SQL provides more detailed information than V$SQLAREA, because it does not group by the
SQL statement text. For a full list of command type values, see description of the V$SESSION
view in Oracle9i Server Reference.

Oracle9i: SQL Tuning Workshop 2-20

 Monitoring Shared Cursor Use

• One load per statement is ideal.

• One load per version/invalidation is optimal.
• More than one load per version indicates a
potential benefit from increasing the shared pool
size.

Copyright © Oracle Corporation, 2002. All rights reserved.

Monitoring Shared Cursor Use

In the best case scenario, there should be one version of each statement that is never
invalidated or aged out.
If the number of loads is significantly higher than the sum of the versions and invalidations,
especially if the number of loads is similar to the number of calls, then the cursor has probably
been reloaded because of aging, and the system may benefit from increasing the size of the
shared pool.

Oracle9i: SQL Tuning Workshop 2-21

 Monitoring Shared Cursor Use

select sql_text, version_count, loads,

invalidations, parse_calls, sorts
from v$sqlarea
where parsing_user_id > 0
and command_type = 3
Order by sql_text;

version invali parse

sql_text count loads dations calls sorts
--------------- -------- ----- ------- ----- -----
select * from 1 2 0 3 0
customers where
CUST_ID = 180

select * from 1 1 0 1 0
customers where
cust_id = 180

Copyright © Oracle Corporation, 2002. All rights reserved.

Monitoring Shared Cursor Use (continued)

The statement above excludes information about recursive SQL (parsing user SYS:
user_id=0) and displays only SELECT commands (command type 3).
There are two versions of the first statement, probably because they reference two different
CUSTOMERS objects. However, each version has been loaded only once. The statement has
been issued three times (PARSE_CALLS).
There is only one version of the second statement, but it has been loaded twice, having been
invalidated once (probably by some DDL on the table or related index).
Note: Oracle SQL Analyze, an add-on component of the Oracle Enterprise Manager (OEM)
Tuning Pack, offers an excellent graphical interface on top of V$SQLAREA.
Instructor Note
V$SQL would distinguish between the two versions of the first SQL statement.
Demo scripts: demo02_03.sql
demo02_01cursors.sql
Edit script demo02_03.sql to fill in the correct user ID and location prior to running the
script.

Oracle9i: SQL Tuning Workshop 2-22

 Cursor Sharing

Use the parameter

CURSOR_SHARING=SIMILAR|FORCE when:
• You want to match to previous statements that
differ only in literal values.
• Response time is low due to very high library
cache misses.

Copyright © Oracle Corporation, 2002. All rights reserved.

Cursor Sharing
The parsing phase compares the text of the statement with existing statements in the shared
pool to see whether the statement can be shared. If the statement differs textually in any way,
then Oracle does not share the statement.
The CURSOR_SHARING parameter enables the DBA to specify the level of matching that
will be considered acceptable by the SQL parser. The default value is EXACT, which forces
the parser to seek a statement that exactly matches the current statement before reusing an
existing cursor.
Setting CURSOR_SHARING to SIMILAR allows the parser to use a cursor for a statement
that is identical in all aspects other than literal values. After the statement is identified, the
current statement will continue to be parsed to ensure that the cursor’s execution plan is
applicable to the current statement. CURSOR_SHARING = FORCE makes the parser behave
the same way as for SIMILAR, except the execution plan in the cursor is always used,
regardless of its applicability.
CURSOR_SHARING = SIMILAR or FORCE is not recommended for certain environments
such as DSS or for complex queries.
See the Oracle9i Database Performance Guide and Reference for more information.

Oracle9i: SQL Tuning Workshop 2-23

 Automated SQL Execution Memory
Management

• Allocation and tuning of PGA memory is simplified

and improved.
• DBAs can use parameters to specify the policy for
PGA sizing.

Copyright © Oracle Corporation, 2002. All rights reserved.

Automated SQL Execution Memory Management

This feature, available with Oracle9i and later, provides an automatic mode for allocating
memory to working areas in the PGA. It simplifies and improves the way PGA memory is
allocated and tuned by Oracle. In Oracle9i DBAs can use the PGA_AGGREGATE_TARGET
parameter to specify the total amount of memory that should be allocated to the PGA areas of
the instance’s sessions. In automatic mode, working areas used by memory-intensive
operators can be automatically and dynamically adjusted.
This feature offers several performance and scalability benefits for DSS workloads or mixed
workloads with complex queries. The overall system performance is maximized and the
available memory is allocated more efficiently among queries to optimize both throughput
and response time. In particular, saving gains from improved use of memory translate to better
throughput at high load.

Oracle9i: SQL Tuning Workshop 2-24

 Dynamic Performance Views

Several dynamic performance views provide

information to help you tune the value of the
PGA_AGGREGATE_TARGET parameter:
• V$SYSSTAT and V$SESSTAT
• V$PGASTAT
• V$PROCESS
• V$SQL_WORKAREA_ACTIVE
• V$SQL_WORKAREA

Copyright © Oracle Corporation, 2002. All rights reserved.

Dynamic Performance Views

View Description
V$SYSSTAT and Statistics in the V$SYSSTAT and V$SESSTAT views show the
V$SESSTAT total number of work areas executed with optimal memory size,
one-pass memory size, and multi-pass memory size. These
statistics are cumulative since the instance or the session was
started.
V$PGASTAT The V$PGASTAT view gives instance-level statistics (in
bytes)on the PGA usage and the automatic memory manager.
V$PROCESS The V$PROCESS view has one row per Oracle process
connected to the instance. The col umns PGA_USED_MEM,
PGA_ALLOC_MEM and PGA_MAX_MEM can be used to
monitor the PGA memory usage of these processes.
V$SQL_WORKAREA Query V$SQL_WORKAREA_ACTIVE to display the work areas
_ACTIVE that are active (or executing) in the instance. Small active sor ts
(under 64KB) are excluded from the view. Use this view to
precisely monitor the size of all active work areas
V$SQL_WORKAREA Oracle maintains cumulative work area statistics for each loaded
cursor whose execution plan uses one or more work areas.
Every time a work area is deallocated, the V$SQL_WORKAREA
table is updated with execution statistics for that work area.
Oracle9i: SQL Tuning Workshop 2-25
 V$SYSSTAT and V$SESSTAT

Find the total number and percentage of times work

areas were executed in the optimal, one-pass, and
multipass modes.
SELECT name profile, cnt,
decode(total, 0, 0, round(cnt*100/total))
percentage
FROM (SELECT name, value cnt,
(sum(value) over ()) total
FROM V$SYSSTAT
WHERE name like ’workarea exec%’);
PROFILE CNT PERCENTAGE
----------------------------------- --------- ----------
workarea executions - optimal 5395 95
workarea executions - onepass 284 5
workarea executions - multipass 0 0

Copyright © Oracle Corporation, 2002. All rights reserved.

V$SYSSTAT and V$SESSTAT

The query shown gives the total number and the percentage of times work areas are executed
in the modes optimal, one-pass, and multipass, since the instance was started.
The output tells the DBA whether the value of the PGA_AGGREGATE_TARGET parameter
must be changed. This can be changed with the ALTER SYSTEM statement.
Because the number of multipass executions is not zero, consider increasing the value of this
parameter. If the value percentage of one-pass executions is high compared to optimal,
increase the value of the PGA_AGGREGATE_TARGET parameter.
Instructor Note
Do not spend a lot of time of explaining these two views. These views are used by DBAs to
help tune the database. Application developers must be aware of the existence of these views,
but typically it is not an application developer’s task to modify the settings.
One-pass and multipass definitions are discussed on page 7.

Oracle9i: SQL Tuning Workshop 2-26

 Summary

In this lesson, you should have learned that the four

most important processing phases are
• Parse
• Bind
• Execute
• Fetch

Copyright © Oracle Corporation, 2002. All rights reserved.

Summary
Use identical SQL statements to reuse SQL areas and thus increase performance.
Realistically, this will be easiest if you use shared code such as packaged procedures, because
the same code will be executed multiple times. This also increases maintainability of the code
and helps narrow your tuning efforts to one set of shared code instead of multiple, scattered
similar SQL statements.
Processing Phases
There are four main stages in processing a SQL statement:
• In the Parse stage, the Oracle9i Server checks the syntax of the statement, determines the
execution plan if the statement is not already in the shared SQL area, and verifies the
information in the data dictionary.
• In the Bind stage, any unresolved bind variables are assigned a value.
• In the Execute stage, the Oracle9i Server executes the statement performing required
reads and writes (except for queries).
• In the Fetch stage (for queries only), rows are retrieved using an array fetch mechanism.

Oracle9i: SQL Tuning Workshop 2-27

 Summary

To optimize the use of shared SQL areas:

• Write generic code
• Follow coding standards
• Use bind variables
• Set the CURSOR_SHARING parameter
• Use automatic PGA memory management

Copyright © Oracle Corporation, 2002. All rights reserved.

Shared SQL Areas

To avoid parsing every SQL statement issued to the Oracle9i Server, develop a coding
standard to take advantage of shared SQL areas:
• Follow coding standards for spacing, uppercase and lowercase, comments, object
references, and so on. The main advantage comes from applications that use the same
SQL statements.
• Use generic code, such as packaged procedures, database triggers, and so on.
• Use the same names to refer to identical database objects.
Try to use bind variables as opposed to literal values in your SQL statements, unless
maximum performance for a single SQL statement is needed. There is a tradeoff between
using shared SQL (with bind variables) and maximum performance (using literal values,
enabling the optimizer to come up with an optimal execution plan). Additionally, use the
automatic PGA memory management.

Oracle9i: SQL Tuning Workshop 2-28

 Practice Overview

This practice covers the following topics:

• SQL statement processing
• Using the V$LIBRARYCACHE view to analyze library
cache performance
• Using the V$SQLAREA view to see information
about all shared cursors in the cache

Copyright © Oracle Corporation, 2002. All rights reserved.

Practice Overview
In this practice, use the data dictionary views to analyze cache management:
• Use the V$LIBRARYCACHE view to analyze library cache performance
• Use the V$SQLAREA view to see information about all shared cursors in the cache
Instructor Note
This practice session is optional. Use it when you want to have a break between the initial
(theoretical) lessons. Usually, you should try to go through the first seven lessons as quickly
as possible, so that the first workshop can be done. During the three days of this class, the
theory-to-practice ratio changes considerably.

Oracle9i: SQL Tuning Workshop 2-29

Practice 2 (optional)
1. Log on with the user ID of SH and password SH. Your instructor will provide the
connection information.
Determine the arraysize set in your SQL*Plus environment.
2. Issue the following SQL statement, then view the information generated by this
statement in the V$SQLAREA view.
SQL> select channel_id, channel_desc
2 from channels;
SQL> select sql_text, sorts
2 from v$sqlarea
3 where command_type = 3
4 and lower(sql_text) like ’%channels%’;
3. Modify the query on the CHANNELS table and add a sort criterion. Then view the
information generated by this statement in the V$SQLAREA view.
SQL> select channel_id, channel_desc
2 from channels
3 order by channel_id;
SQL> select sql_text, sorts
2 from v$sqlarea
3 where command_type = 3
4 and lower(sql_text) like ’%channels%’;
4. Insert a new record into the CHANNELS table. View the information generated by the
this statement in the V$SQLAREA view. Roll back your transaction to restore the
original situation.
SQL> insert into channels
2 values(’W’, ’Wholesale’, ’Direct’);
SQL> select sql_text, sorts
2 from v$sqlarea
3 where command_type in (2,3)
4 and lower(sql_text) like ’%channels%’;
SQL> rollback;
5. Use the V$LIBRARYCACHE view to check the amount of SQL caching.

Oracle9i: SQL Tuning Workshop 2-30

 EXPLAIN and AUTOTRACE

Copyright © Oracle Corporation, 2002. All rights reserved.

Schedule: Timing Topic

45 minutes Lecture
10 minutes Practice
55 minutes Total
Instructor Note
This lesson introduces EXPLAIN and SQL*Plus AUTOTRACE. SQL*Plus AUTOTRACE is the
diagnostic tool that is used in most workshops. Unfortunately, you must introduce statistics
when using AUTOTRACE. Indicate that statistics will be discussed in later lessons and
introduce the most important ones.
Keep the practice session short; there will be plenty of time to use the diagnostic tools in later
lessons and workshops. The practice session is a short break between this lesson and the
following one. You may also want to discuss the other diagnostic tools first, and combine the
practice sessions.
 Objectives

After completing this lesson, you should be able to

do the following:
• Create a table named PLAN_TABLE
• Use EXPLAIN PLAN to show how a statement will
be processed
• Query the performance view V$SQL_PLAN
• Use SQL*Plus AUTOTRACE to show SQL statement
execution plans and statistics

Copyright © Oracle Corporation, 2002. All rights reserved.

Objectives
You can use various analysis tools to evaluate the performance of SQL statements.
After completing this lesson, you should be able to:
• Create the PLAN_TABLE by using the utlxplan.sql script
• Use the EXPLAIN PLAN command to show how a statement is processed
• Query the V$SQL_PLAN performance view to examine the execution plan for cursors
that were recently executed.
• Use the SQL*Plus AUTOTRACE setting to show SQL execution plans and statistics

Oracle9i: SQL Tuning Workshop 3-2

 Creating the Plan Table

• Run a standard script to create a plan table.

@utlxplan.sql

• Create your own plan table with a different name,

by editing the utlxplan.sql script and changing
the table name within the script.

Copyright © Oracle Corporation, 2002. All rights reserved.

Creating the Plan Table

The EXPLAIN PLAN command uses a table to store information about the execution plan
chosen by the optimizer. If you examine the plan, you can see how the Oracle9i Server
executes the statement. You must follow three steps to use EXPLAIN PLAN:
1. Create the plan output table.
2. Use the EXPLAIN PLAN command.
3. Retrieve the plan steps by using SQL statements against the plan table.
The PLAN_TABLE
You have several options in creating a plan table:
• You can use the default output table named PLAN_TABLE.
• You can use the utlxplan.sql script to create PLAN_TABLE.
In a Windows environment, this script can be found in
%ORACLE_HOME%\rdbms\admin; and under UNIX in
$ORACLE_HOME/rdbms/admin. In a client-server environment, be sure to use the
correct (server side) version of this script.
• You can also copy and edit the utlxplan.sql script; note that you should not change
the structure of the table.

Oracle9i: SQL Tuning Workshop 3-3

The Create Plan Table Syntax
The entire syntax for the plan table is shown below:
create table PLAN_TABLE
(statement_id varchar2(30)
,timestamp date
,remarks varchar2(80)
,operation varchar2(30)
,options varchar2(255)
,object_node varchar2(128)
,object_owner varchar2(30)
,object_name varchar2(30)
,object_instance numeric
,object_type varchar2(30)
,optimizer varchar2(255)
,search_columns number
,id numeric
,parent_id numeric
,position numeric
,cost numeric
,cardinality numeric
,bytes numeric
,other_tag varchar2(255)
,partition_start varchar2(255)
,partition_stop varchar2(255)
,partition_id numeric
,other long
,distribution varchar2(30)
,cpu_cost numeric
,io_cost numeric
,temp_space numeric);
Column descriptions can be found in Appendix B, “Diagnostic Tools Reference.” The most
important columns will be discussed later in this lesson.
Instructor Note
The plan table creation syntax is for student interest only. Usually, the table is created using
utlxplan.sql. However, if a participant wants to call the table by a different name, the full
table definition is included here.

Oracle9i: SQL Tuning Workshop 3-4

 The EXPLAIN PLAN Command

EXPLAIN PLAN
SET STATEMENT_ID
= ’text’

INTO your plan table

FOR statement

Copyright © Oracle Corporation, 2002. All rights reserved.

The EXPLAIN PLAN Command

This command inserts a row into the plan table for each step of the execution plan.
In the syntax diagram shown in the slide, the fields in italics have the following meanings:

Field Meaning
text This is an optional identifier for the
statement. You should enter a value
to identify each statement so that you can
later specify the statement that you want
explained. This is especially important
when you share the plan
table with others, or when you keep
multiple execution plans in the same
plan table.
schema.table This is the optional name of the output
table. The default is PLAN_TABLE.

statement This is the text of the SQL statement.

Oracle9i: SQL Tuning Workshop 3-5

 EXPLAIN PLAN Example

SQL> EXPLAIN PLAN

2 set statement_id = ’demo01’ for
3 select *
4 from products
5 where prod_category = ’Men’
6 and prod_subcategory = ’Jeans - Men’;

Explained.

Note: The EXPLAIN PLAN command does not

actually execute the statement.

Copyright © Oracle Corporation, 2002. All rights reserved.

EXPLAIN PLAN Example

This command inserts the execution plan of the SQL statement into the plan table, and adds the
name tag demo01 for future reference.
Instructor Note
Demo script: demo03_01.sql
Note that this script could take some time to complete, because it must delete statistics (to force
RBO) and to (re)create two indexes, to make the result consistent with the next slides.
Instructor Note for the next page
Demo script: demo03_02.sql (output on next slide)
Do not forget to execute “delete from plan_table” afterwards; you can also show the
rp.sql script that automatically deletes all rows from the plan table.

Oracle9i: SQL Tuning Workshop 3-6

 Displaying the Execution Plan

SQL> column "Query Plan" format a60

SQL> select id
2 , lpad(’ ’, 2*level)||operation
3 ||decode(id,0,’ Cost = ’||position)
4 ||’ ’||options
5 ||’ ’||object_name as "Query Plan"
6 from plan_table
7 where statement_id = ’demo01’
8 connect by prior id = parent_id
9 start with id = 0;

Copyright © Oracle Corporation, 2002. All rights reserved.

Displaying the Execution Plan

The select statement in the slide is designed to display the steps that would be taken if the
SQL statement were executed. The following plan table columns are used:
STATEMENT_ID The value of the optional STATEMENT_ID parameter
specified in the EXPLAIN PLAN statement
OPERATION The name of the operation performed at this step;
contains a value in the first row generated for each statement
OPTIONS The options for the operation (For example, this will say “remote”
if a remote table is referenced in the SQL statement).
OBJECT_NAME The name of the object
ID The number of this step in the execution plan
PARENT_ID The ID of the step that operates on the results of this step
POSITION The order in which steps with the same parent are processed
Note: Usually, you put this statement into a SQL script with a variable for the statement ID
value. This makes using the explain plan much more feasible. In that same script, you can
also delete (or truncate) all rows from the plan table after you have retrieved them.
Alternatively, use SQL*Plus AUTOTRACE to eliminate the need to issue queries against the
plan table.

Oracle9i: SQL Tuning Workshop 3-7

 Displaying the Execution Plan

SQL> select id
2 , lpad(’ ’,2*level)||operation||
3 decode(id, 0,’ Cost = ’||position)
4 ||’ ’||options
5 ||’ ’||object_name as "Query Plan"
6 from plan_table
7 where statement_id = ’demo01’
8 connect by prior id = parent_id start with id=0;

ID Query Plan
----- -----------------------------------------
0 SELECT STATEMENT Cost =
1 TABLE ACCESS BY INDEX ROWID PRODUCTS
2 AND-EQUAL
3 INDEX RANGE SCAN PRODUCTS_PROD_CAT_IX
4 INDEX RANGE SCAN PRODUCTS_PROD_SUBCAT_IX

Copyright © Oracle Corporation, 2002. All rights reserved.

Displaying the Execution Plan (continued)

Some other useful plan table columns (not used in the example) are:
TIMESTAMP The date and time when the EXPLAIN PLAN statement was issued
COST The cost-based optimizer’s estimate of the cost of performing the
operation, including the cost of any operations that are required to
complete this operation (The value of this column does not have a
particular unit of measurement. It is merely a weighted value used to
compare costs of execution plans.)
CARDINALITY The cost-based optimizer’s estimate of the number of rows processed
BYTES The cost-based optimizer’s estimate of the number of bytes
OTHER The SQL text for remote cursors and parallel query slaves,
as well as parallel query dataflow-operation information
OTHER_TAG The function descriptions of the SQL text in the OTHER column
OPTIMIZER The current mode of the optimizer
OBJECT_OWNER The name of the schema that contains the table or index
OBJECT_TYPE A modifier that describes the object, such as NON-UNIQUE for an index
Instructor Note
A NULL value for COST always implies rule-based optimization.
Oracle9i: SQL Tuning Workshop 3-8
 Interpreting the Execution Plan

TABLE ACCESS
1 (BY INDEX ROWID)
classes

AND_EQUAL
2

INDEX INDEX
(RANGE_SCAN) 3 4 (RANGE_SCAN)
Prod category Prod subcategory
index index

Copyright © Oracle Corporation, 2002. All rights reserved.

Interpreting the Execution Plan

Each step of the execution plan either retrieves rows from the database or accepts rows as
input from one or more other steps, also known as row sources. In the diagram, the numbers
correspond to the ID values in the plan table (see previous page).
The AND_EQUAL operation (2) accepts the ROWIDs obtained by the scans of the
STAT_INDEX (3) and LOC_INDEX (4), and returns the ROWIDs that come from both
sources. This results in a set of ROWIDs that allow TABLE ACCESS (1) of the CLASSES
table to return the rows that satisfy the query. ROWIDs are explained in the “Indexes and
Basic Access Methods” lesson.
The cost of an execution plan is a value proportional to the expected elapsed time needed to
execute the statement, using that execution plan. If the rule-based optimizer is used, then the
cost is not available and shows a NULL value.

Oracle9i: SQL Tuning Workshop 3-9

Comparing Execution Plans
You can determine the execution plan used by the optimizer by using EXPLAIN PLAN. If a
SQL statement does not perform satisfactorily, you may decide to introduce an index. You
should check the execution plan again to make sure that the index is used. You can also check
the effect of using various optimizer hints, which are covered in the “Influencing the
Optimizer” lesson.
Reference
For more information about interpreting execution plans, see Oracle9i Database Performance
Guide and Reference, “Using EXPLAIN PLAN” and Appendix B of this student guide.

Oracle9i: SQL Tuning Workshop 3-10

 Using V$SQL_PLAN

• Provides a way of examining the execution plan for

cursors that were recently executed.
• Information in this view is very similar to the
output of an EXPLAIN PLAN statement:
– EXPLAIN PLAN shows a theoretical plan that can be
used if this statement were to be executed.
– V$SQL_PLAN contains the actual plan used.

Copyright © Oracle Corporation, 2002. All rights reserved.

Using V$SQL_PLAN
This view provides a way of examining the execution plan for cursors that were recently
executed.
The information in this view is very similar to the output of an EXPLAIN PLAN statement.
However, EXPLAIN PLAN shows a theoretical plan that can be used if this statement were
to be executed, whereas V$SQL_PLAN contains the actual plan used. The execution plan
obtained by the EXPLAIN PLAN statement can be different from the execution plan used to
execute the cursor, because the cursor might have been compiled with different values of
session parameters (for example, HASH_AREA_SIZE).
V$SQL_PLAN shows the plan for a cursor, not for a SQL statement. The difference is that a
SQL statement can have more than one cursor associated with it, with each cursor further
identified by a CHILD_NUMBER. For example the same statement executed by different
users will have different cursors associated with it if the object being referenced is in a
different schema. Similarly different hints can cause different cursors. The V$SQL_PLAN
table can be used to see the different plans for different child cursors of the same statement.

Oracle9i: SQL Tuning Workshop 3-11

 V$SQL_PLAN Columns

ADDRESS Address of the handle to the parent of the cursor

OBJECT# Object number of the table or the index

OBJECT_NAME Name of the table or index

OPERATION Name of the internal operation performed in this

step; for example, TABLE ACCESS
OPTIONS A variation on the operation described in the
OPERATION column; for example, FULL
OPTIMIZER Current mode of the optimizer for the first row in
the plan (statement line)

COST Cost of the operation as estimated by the

optimizer’s cost-based approach

Copyright © Oracle Corporation, 2002. All rights reserved.

Using V$SQL_PLAN (continued)

There are several uses for V$SQL_PLAN:
• Determining the current execution plan
• Identifying the effect of creating an index on a table
• Finding cursors containing a certain access path (for example, full table scan or index
range scan)
• Identifying indexes that are, or are not, selected by the optimizer
• Determining whether the optimizer selects the particular execution plan (for example,
nested loops join) expected by the developer
This view can also be used as a key mechanism in plan comparison to observe the changes in:
• Dropping or creating indexes
• Running the ANALYZE statement on the database objects
• Modifying initialization parameter values
• Switching from the rule-based optimizer to the cost-based optimizer
• Migrating the application or the database to a new release
If previous plans are kept, then it is then possible to identify how changes in the performance
of a SQL statement can be correlated with changes in the execution plan for that statement.

Oracle9i: SQL Tuning Workshop 3-12

 V$SQL_PLAN Columns

HASH_VALUE Hash value of the parent statement in the library

cache
ADDRESS Object number of the table or the index

CHILD_NUMBER Child cursor number using this execution plan

POSITION Order of processing for operations that all have

the same PARENT_ID
PARENT_ID ID of the next execution step that operates on
the output of the current step

ID A number assigned to each step in the

execution plan

Note: This is a partial listing of columns

Copyright © Oracle Corporation, 2002. All rights reserved.

V$SQL_PLAN Columns
The view contains almost all PLAN_TABLE columns, in addition to new columns. The
columns that are also present in the PLAN_TABLE have the same values:
• ADDRESS
• HASH_VALUE
The two columns ADDRESS and HASH_VALUE can be used to join with V$SQLAREA to add
the cursor-specific information.
The ADDRESS, HASH_VALUE and CHILD_NUMBER columns can be used to join with
V$SQL to add the child cursor-specific information.

Oracle9i: SQL Tuning Workshop 3-13

 Querying V$SQL_PLAN
SELECT id
, lpad (’ ’, depth) || operation operation
, options , object_name , optimizer , cost
FROM V$SQL_PLAN
WHERE hash_value = 912244748
AND address = ’67D419DC’
START WITH id = 0
CONNECT BY
( prior id = parent_id
AND prior hash_value = hash_value
AND prior child_number = child_number
)
ORDER SIBLINGS BY id, position;

ID OPERATION OPTIONS OBJECT_NAME OPT COST

-- --------------- --------------- ---------------------- --- -----
0 SELECT STATEMENT CHO
1 TABLE ACCESS BY INDEX ROWID PRODUCTS
2 AND-EQUAL
3 INDEX RANGE SCAN PRODUCTS_PROD_CAT_IX
4 INDEX RANGE SCAN PRODUCTS_PROD_SUBCAT_I

Copyright © Oracle Corporation, 2002. All rights reserved.

Querying V$SQL_PLAN
The above statement shows the execution plan for the SQL SELECT statement shown
on page 3-6. Looking at the plan for a SQL statement is one of the first steps in tuning a SQL
statement. The SQL statement that is used to return the plan is identified by the statement’s
HASH_VALUE and ADDRESS from V$SQL_PLAN.
Note: For statements that use the rule-based approach, the COST column is null.
Instructor Note
These are a few examples of how a SQL statement can result in more than one cursor:
When the same table name resolves to two separate tables:
User1: SELECT * FROM CUSTOMERS;
User2: SELECT * FROM CUSTOMERS;
Where User2 has his or her own customer table, and User1 uses the table referenced by a
public synonym.

Oracle9i: SQL Tuning Workshop 3-14

 SQL*Plus AUTOTRACE

OFF
SET AUTOTRACE ON
TRACE[ONLY]

EXPLAIN
STATISTICS

SHOW AUTOTRACE

Copyright © Oracle Corporation, 2002. All rights reserved.

SQL*Plus AUTOTRACE
In SQL*Plus, you can automatically obtain the execution plan and some additional statistics
on the running of a SQL command by using the AUTOTRACE setting. Unlike the EXPLAIN
PLAN command, the statement is actually run, even if you choose to suppress the statement
output; thus you can display realistic statistics. AUTOTRACE, a feature available since Oracle
Server release 7.3, is an excellent diagnostic tool for SQL Statement Tuning. Because it is
purely declarative, it is easier to use than EXPLAIN PLAN.
Command Options
OFF Disables autotracing SQL statements
ON Enables autotracing SQL statements
TRACEONLY Enables autotracing SQL statements and suppresses statement output
EXPLAIN Displays execution plans but does not display statistics
STATISTICS Displays statistics but does not display execution plans
Note: If both command options EXPLAIN and STATISTICS are omitted, execution plans
and statistics are displayed by default.
Instructor Note (for page 3-16)
There are some more settings for the (optional) second part of AUTOTRACE output, for
parallel and distributed queries. However, they are not relevant at this point.

Oracle9i: SQL Tuning Workshop 3-15

 SQL*Plus AUTOTRACE Examples

• To start tracing statements using AUTOTRACE:

set autotrace on

• To hide statement output:

set autotrace traceonly

• To display execution plans only:

set autotrace traceonly explain

• Control the layout with column settings.

Copyright © Oracle Corporation, 2002. All rights reserved.

Prerequisites for Using AUTOTRACE

To use the EXPLAIN option of AUTOTRACE, you must first create the PLAN_TABLE table
in your schema, just as you would need to do before using the EXPLAIN PLAN command.
To access STATISTICS data, you must have access to several dynamic performance tables.
The DBA can grant access by using the role plustrace created in the plustrce.sql
script. (This script name may vary by operating system.) The DBA must run the script as the
SYS user and grant the role to users who want to use the STATISTICS option of
AUTOTRACE.
Controlling the AUTOTRACE Execution Plan Layout
The execution plan consists of four columns displayed in the following order:
ID_PLUS_EXP The line number of each step
PARENT_ID_PLUS_EXP Parent step line number
PLAN_PLUS_EXP The report step
OBJECT_NODE_PLUS_EXP Database links or parallel query servers used
You can change the format of these columns, or suppress them, by using the SQL*Plus
COLUMN command. For further information, see the Oracle9i SQL*Plus User’s Guide and
Reference.

Oracle9i: SQL Tuning Workshop 3-16

 SQL*Plus AUTOTRACE Statistics

set autotrace traceonly statistics

select * from dual;

Statistics
---------------------------------------------------
0 recursive calls
2 db block gets
1 consistent gets
0 physical reads
0 redo size
367 bytes sent via SQL*Net to client
430 bytes received via SQL*Net from client
2 SQL*Net roundtrips to/from client
0 sorts (memory)
0 sorts (disk)
1 rows processed

Copyright © Oracle Corporation, 2002. All rights reserved.

AUTOTRACE Statistics
AUTOTRACE displays several statistics, not all of them relevant to the discussion at this
stage. In the next lesson, when discussing SQL Trace and TKPROF, this course goes into
more detail about SQL statement execution statistics. The most important ones are the
following:
db block gets The number of logical I/Os for current gets
consistent gets The number of logical I/Os for read-consistent gets
physical reads The number of blocks read from disk
redo size The amount of redo generated (for DML statements)
sorts (memory) The number of sorts performed in memory
sorts (disk) The number of sorts performed using temporary disk storage
Note: Current gets are reads of the current block’s buffer cache; consistent gets are reads of
rollback segments due to changes in the buffer. The db block gets, consistent gets, and
physical reads are the three statistics that are usually monitored. These should be low
compared to the number of rows retrieved. Sorts should be performed in memory rather than
disk.
Instructor Note
Demo script: demo03_03.sql
See Instructor Note on page 3-19.
Oracle9i: SQL Tuning Workshop 3-17
 Summary

In this lesson, you should have learned how to:

• Create a PLAN_TABLE
• Use EXPLAIN PLAN to view execution plans
• Query V$SQL_PLAN to see the execution plan for
cursors that were recently executed
• Use SQL*Plus AUTOTRACE to run statements
and display execution plans and statistics

Copyright © Oracle Corporation, 2002. All rights reserved.

Summary
Tuning involves careful analysis of execution plans and execution statistics. Several tools
are available to help you test and diagnose performance problems with a SQL statement:
• The PLAN_TABLE stores execution plans and can be created using the
utlxplan.sql script.
• EXPLAIN PLAN determines how the optimizer processes a statement (the statement
is not executed) and stores the execution plan in the PLAN_TABLE.
• You can also use the view V$SQL_PLAN to view the execution plan for recently
executed cursors.
• The SET AUTOTRACE command in SQL*Plus enables you to view the execution
plan and a variety of statistics about the execution whenever you execute a SQL
statement.
You can use several other tools to monitor performance, alerting you to the need to tune
SQL statements. Some of these tools will be covered in future lessons.

Oracle9i: SQL Tuning Workshop 3-18

 Practice Overview

This practice covers the following topics:

• Creating a PLAN_TABLE
• Using the EXPLAIN PLAN command to view
execution plans
• Formatting execution plan results by using the
rp.sql script
• Setting the SQL*Plus AUTOTRACE command to
display execution plans and statistics for SQL
statements

Copyright © Oracle Corporation, 2002. All rights reserved.

Practice Overview
In this practice, you create a PLAN_TABLE, analyze a SQL statement, and format the
analyzed results by using the rp.sql script. You also use the SQL*Plus AUTOTRACE
command to view execution plans.
Instructor Note (for page 3-17)
Demo script: demo03_03.sql
Note that consistent gets can be higher than expected due to array processing; each new array
needs one additional get. The number of array fetches equals (SQL*Net roundtrips - 2).
AUTOTRACE performs an implicit query on v$sesstat; see script demo03_04.sql for a
similar query on v$sesstat. This query shows the results for the session.
demo03_03.sql shows the results from the last statement.
Instructor Note (for the Practice)
The optimizer_mode parameter for the session should be set to rule to ensure that the
results are easier to interpret. At this point, the goal is to have the students use autotrace
and understand basic execution plans.

Oracle9i: SQL Tuning Workshop 3-19

Practice 3
1. Verify the existence of a PLAN_TABLE table in your schema. Create this table when
needed, using the utlxplan.sql script. If the table already exists, ensure that the
table is empty.
2. Ensure that your session is set to rule-based optimization by executing the following
command. Then explain the following SQL SELECT statement:
Note: Rule-based optimization is discussed in detail in a later lesson.
SQL> alter session set optimizer_mode = RULE;
SQL> select cust_first_name, cust_last_name
2 from customers
3 where cust_id = 100;
3. Now query the PLAN_TABLE table:
SQL> select * from plan_table;
Use the rp.sql script to query the PLAN_TABLE table in a more sophisticated
manner:
SQL> @rp
4. Enable SQL*Plus AUTOTRACE to display execution plans and run the command from
step 2 again.
5. Set AUTOTRACE to suppress the command output and show both execution plans and
statistics, and run step 2 again.
6. If you have some time left, enter your own SQL statements and experiment with
EXPLAIN PLAN and AUTOTRACE.

Oracle9i: SQL Tuning Workshop 3-20

 SQL Trace and TKPROF

Copyright © Oracle Corporation, 2002. All rights reserved.

Schedule: Timing Topic

45 minutes Lecture
10 minutes Practice
55 minutes Total

Instructor Note
This lesson introduces the SQL Trace utility and the associated initialization parameters.
It explains how to enable SQL Trace for a session and how to find your trace files. It also
covers the TKPROF command to format trace files.
Keep the practice session short; the students will have enough time to use SQL Trace and
TKPROF in the workshops.
 Objectives

After completing this lesson, you should be able to do

the following:
• Configure SQL Trace to collect session statistics
• Enable SQL Trace and locate your trace files
• Format trace files using TKPROF
• Interpret the output of the TKPROF command

Copyright © Oracle Corporation, 2002. All rights reserved.

Objectives
You can use various analysis tools to evaluate the performance of SQL statements. After
completing this lesson, you should be able to:
• Configure SQL Trace to gather statistics
• Set up appropriate initialization parameters
• Format the trace statistics using the TKPROF utility
• Interpret the output of the TKPROF command
Background Reading
For more information about SQL Trace and TKPROF, see the following:
• Oracle9i Performance Guide and Reference chapters in this part are: “Using
EXPLAIN PLAN,”, “Using SQL Trace and TKPROF,” “Using Autotrace in
SQL*Plus,” “Using Oracle Trace”
• Oracle9i Performance Methods
• Oracle9i Application Developer’s Guide, “Using Procedures and Packages”
• Oracle9i Reference, “Initialization Parameters”

Oracle9i: SQL Tuning Workshop 4-2

 SQL Trace Facility

• Is set at the instance or session level

• Gathers statistics for SQL statements
• Produces output that can be formatted by TKPROF

Trace
SQL Trace
file

Report
TKPROF
file
Database

Copyright © Oracle Corporation, 2002. All rights reserved.

SQL Trace Facility

A good way to compare two execution plans is to run them and compare the statistics to see
which one performs better. You can use the SQL Trace facility to obtain performance
information. SQL Trace writes its output to a file, and you use TKPROF to format it.
The SQL Trace facility:
• Can be switched on for an instance or a session
• Reports on volume and time statistics for the Parse, Execute, and Fetch phases
• Produces output that can be formatted by TKPROF
When the SQL trace facility is enabled for a session, Oracle generates a trace file containing
statistics for traced SQL statements for that session. When the SQL trace facility is enabled
for an instance, Oracle creates a separate trace file for each process.
Note: SQL Trace involves some overhead, so you usually do not want to enable SQL Trace at
the instance level.

Oracle9i: SQL Tuning Workshop 4-3

 How to Use the SQL Trace Facility

1. Set the initialization parameters.

2. Switch on tracing.
3. Run the application.
4. Switch tracing off and format the trace file.
5. Interpret the output.
Trace
SQL Trace
file

Report
TKPROF
file
Database

Copyright © Oracle Corporation, 2002. All rights reserved.

How to Use the SQL Trace Facility

You must complete five steps to use the SQL Trace tool:
1. Set the appropriate initialization parameters.
2. Switch on SQL Trace.
3. Run the application (and switch off tracing when done).
4. Format the trace file produced by SQL Trace with TKPROF.
5. Interpret the output and tune the SQL statements when needed.
Running SQL Trace increases system overhead. Use SQL Trace only when required, and use
it at the session level rather than at the instance level.
Instructor Note
On certain platforms you must exit your session after step 3 to get the full trace file on disk.
Sometimes the last part is missing from a trace file; exiting your session resolves that
problem. Statistics lines in the trace file are written only when the corresponding cursor is
closed.
Note: Unlike previous releases TKPROF no longer needs to log back into the database to get
an explain plan, because this is already provided in the trace files.

Oracle9i: SQL Tuning Workshop 4-4

 Initialization Parameters

TIMED_STATISTICS = {false|true}
MAX_DUMP_FILE_SIZE = {n|500}
USER_DUMP_DEST = directory_path

• MAX_DUMP_FILE_SIZE is measured in operating

system blocks
• Default USER_DUMP_DEST is the location of the
background dump destination

Copyright © Oracle Corporation, 2002. All rights reserved.

Initialization Parameters
Several initialization parameters relate to SQL Trace.
TIMED_STATISTICS
The SQL Trace facility provides a variety of information about processes, optionally including
timing information. If you want timing information, you must turn on this parameter. You can
do so for the entire database by setting the following initialization parameter in the parameter
file before starting up and opening the database:
TIMED_STATISTICS = TRUE
The parameter also can be set dynamically for a particular session with the following
command:
SQL> ALTER SESSION SET timed_statistics=true;
The timing statistics have a resolution of one-hundredth of a second. This means that any
operation that finishes quickly may not be timed accurately, such as simple queries that execute
quickly.
Having TIMED_STATISTICS turned on affects performance slightly because the Oracle
server must do some additional work. Therefore, this parameter is commonly turned off until
specifically desired for tuning purposes. However, keeping TIMED_STATISTICS turned on
could make trace files more useful for support engineers during a system crash.
Note: The underlined values in the slide indicate the default values for the parameters.
Oracle9i: SQL Tuning Workshop 4-5
Initialization Parameters (Continued)
MAX_DUMP_FILE_SIZE and USER_DUMP_DEST
These two parameters control the size and destination of the output file:
MAX_DUMP_FILE_SIZE = n
USER_DUMP_DEST = directory_path
The MAX_DUMP_FILE_SIZE default value is 500, and this parameter value is
expressed in operating system blocks. MAX_DUMP_FILE_SIZE can also be changed at
the session level by using the ALTER SESSION command.
The default USER_DUMP_DEST location varies by operating system; it typically is the
default destination for system dumps on your system. USER_DUMP_DEST is a system-
level parameter that cannot be changed at the session level. It can only be changed
dynamically by a database administrator using the ALTER SYSTEM command.
Obtain Information about Parameter Settings
You can display current parameter values by querying the V$PARAMETER view:
SQL> SELECT name, value
2 FROM v$parameter
3 WHERE name LIKE ’%dump%’;
Note: The directory path specifications that show up in the VALUE column are operating
system dependent. This view is accessible only by users with DBA privileges
Instructor Note
You can also use the show parameter <search string> command; all
commands that were specific for the Server Manager tool are now available in
SQL*Plus, provided you have the correct privileges.
If the MAX_DUMP_FILE_SIZE is reached, *** Trace File Full *** is written to the
trace file. No notification or error message is displayed, and the background process
stops writing to the trace file.
Note that there is an additional initialization parameter, SQL_TRACE, that influences
SQL Trace. However, it is not a parameter to use when preparing for SQL Trace, but
rather when enabling SQL Trace. That is why it is not on the previous slide but is
introduced on the next slide. You should discourage use of that parameter and suggest
enabling SQL Trace at the session level instead.
Note that when you are using SQL Trace in Multi-Threaded Server (MTS), the shared
server processes will create trace files in the BACKGROUND_DUMP_DEST, not in the
USER_DUMP_DEST directory. It is recommended to use a dedicated server process for
tuning.

Oracle9i: SQL Tuning Workshop 4-6

 Switching On SQL Trace

• For an instance, set the following parameter:

SQL_TRACE = TRUE

• For your current session:

SQL> ALTER SESSION SET sql_trace = true;

SQL> EXECUTE dbms_session.set_sql_trace(true);

• For any session:

SQL> EXECUTE dbms_system.set_sql_trace_in_session
2 (session_id, serial_id, true);

Copyright © Oracle Corporation, 2002. All rights reserved.

Switching On SQL Trace for an Instance

In order to switch on SQL Trace for an entire instance, set the initialization parameter:
SQL_TRACE = TRUE
Switching On SQL Trace for a Session
You can use a SQL command to switch on SQL Trace for your session.
SQL> ALTER SESSION SET sql_trace = true;
You can also use a supplied package to switch on SQL Trace for your session. This is useful
when you want to turn SQL Trace on and off from within a PL/SQL unit.
SQL> EXECUTE dbms_session.set_sql_trace(true);
You can also switch on SQL Trace for another user’s session with a supplied package.
SQL> EXECUTE dbms_system.set_sql_trace_in_session
2 (session_id, serial_id, true);
In this procedure call, session_id and serial_id are the values in the SID and SERIAL#
columns of V$SESSION, a data dictionary view commonly used by database administrators.
Switching Off SQL Trace
When you have finished tuning, you can switch off SQL Trace by using any of the above
methods, substituting the word FALSE for TRUE. If you switch on SQL Trace for a single
session, exiting that session also turns off SQL Trace.
Oracle9i: SQL Tuning Workshop 4-7
 Finding Your Trace Files

• Look in the directory specified by USER_DUMP_DEST.

• If only a few people are running SQL Trace, look for
the most recent time stamp.
• Otherwise, consider writing a script.

SQL> @readtrace.sql

SQL> EXECUTE gettrace;

PL/SQL procedure successfully completed.

Copyright © Oracle Corporation, 2002. All rights reserved.

Finding Your Trace Files

Identifying your trace file in the directory specified by USER_DUMP_DEST is usually a matter
of finding the trace file with the most recent time stamp. However, this task becomes more
difficult when several users generate trace files at the same time.
When Several Users Create Trace Files
When several users create trace files simultaneously, the readtrace.sql script can be
useful. It creates a procedure that opens your current trace file by using the UTL_FILE
package. The default output file name is username.trc, but you can also specify a name
yourself.
SQL> @readtrace.sql
SQL> ALTER SESSION SET sql_trace = true;
SQL> SELECT * FROM dual;
SQL> execute gettrace(’output_filename’);
Instructor Note
Tell the students that readtrace is not a standard Oracle utility that comes with the
software, but is only developed and provided for this course. The script is written for an NT
environment and must be modified for other environments.
By default, the .trc file is named after the SPID. You can find the SPID in v$process.
To find your SPID for the session, run the demo04_01spid.sql script.
Oracle9i: SQL Tuning Workshop 4-8
 Formatting Your Trace Files

OS> tkprof tracefile outputfile [options]

TKPROF command examples:

OS> tkprof
OS> tkprof ora_901.trc run1.txt
OS> tkprof ora_901.trc run2.txt sys=no
sort=execpu print=3

Copyright © Oracle Corporation, 2002. All rights reserved.

Formatting Your Trace Files

Use the TKPROF command to format your trace files into a readable output.
This is the TKPROF syntax:
OS> tkprof tracefile outputfile [options]
tracefile The name of the trace output file (input for TKPROF)
outputfile The name of the file to store the formatted results
When the TKPROF command is entered without any arguments, it generates a usage message
together with a description of all TKPROF options. See the next page for a full listing.
Note: Trace files generated immediately after instance startup contain data that reflects the
activity of the startup process. In particular, they reflect a disproportionate amount of physical
I/O activity as caches in the SGA are filled. For the purposes of tuning, you should ignore
such trace files.
Instructor Note
These are just some examples; the full TKPROF syntax will be explained later. In previous
releases, TKPROF had a version component in the command name on Windows platforms (for
example, tkprof73 or tkprof80). With Oracle8i and higher, there is no tkprof8i
command both in UNIX and NT anymore.
Instructor Note (for page 4-10)
This is the (formatted) output you get when executing the TKPROF command without
arguments; some options are explained in more detail on the following pages.
Oracle9i: SQL Tuning Workshop 4-9
Output of the TKPROF Command
This is the output you get when you enter the TKPROF command without any arguments:
Usage: tkprof tracefile outputfile [explain= ] [table= ]
[print= ] [insert= ] [sys= ] [sort= ]

table=schema.tablename Use schema.tablename with the explain option.

explain=user/password Connect to the database and issue EXPLAIN PLAN.
print=integer List only the first integer SQL statements.
aggregate=no Do not aggregate identical SQL statements into one
result.
insert=filename List SQL statements and data inside INSERT
statements.
sys=no TKPROF does not list SQL statements run as user SYS.
record=filename Record nonrecursive statements found in the trace file.
sort=option A set of zero or more of the following sort options:
Options: prscnt Number of times parse was called
prscpu CPU time parsing
prsela Elapsed time parsing
prsdsk Number of disk reads during parse
prsqry Number of buffers for consistent read during parse
prscu Number of buffers for current read during parse
prsmis Number of misses in library cache during parse
execnt Number of execute was called
execpu CPU time spent executing
exeela Elapsed time executing
exedsk Number of disk reads during execute
exeqry Number of buffers for consistent read during execute
execu Number of buffers for current read during execute
exerow Number of rows processed during execute
exemis Number of library cache misses during execute
fchcnt Number of times fetch was called
fchcpu CPU time spent fetching
fchela Elapsed time fetching
fchdsk Number of disk reads during fetch
fchqry Number of buffers for consistent read during fetch
fchcu Number of buffers for current read during fetch
fchrow Number of rows fetched
userid ID of user who parsed the cursor

Oracle9i: SQL Tuning Workshop 4-10

 TKPROF Command Options

SORT = option
PRINT = n
EXPLAIN = username/password
INSERT = filename
SYS = NO
AGGREGATE = NO
RECORD = filename
TABLE = schema.tablename

Copyright © Oracle Corporation, 2002. All rights reserved.

TKPROF Command Options

SORT The order in which to sort the statements in the report (See previous page
for a list of values.)
PRINT Produces a report on this (sorted) number of statements only (This option
is especially useful in combination with the SORT option.)
EXPLAIN Logs on and executes EXPLAIN PLAN in the specified schema
INSERT Creates a SQL script to load the TKPROF results into a database table
SYS Disables the listing of recursive SQL statements, issued by the user SYS
AGGREGATE Disables or enables the (default) behavior of TKPROF, aggregating
identical SQL statements into one record
RECORD Creates a SQL script with all the nonrecursive SQL statements found in
the trace file (This script can be used to replay the tuning session later.)
TABLE Specifies the table to temporarily store execution plans before writing them
to the output file (This parameter is ignored if EXPLAIN is not specified.
It is useful when several individuals use TKPROF to tune the same
schema concurrently, to avoid destructive interference.)

Oracle9i: SQL Tuning Workshop 4-11

 Output of the TKPROF Command

• Text of the SQL statement

• Trace statistics (for statement and recursive calls)
separated into three SQL processing steps:

PARSE Translates the SQL statement

into an execution plan

EXECUTE Executes the statement

(This step modifies the data for INSERT,
UPDATE, and DELETE statements.)

FETCH Retrieves the rows returned by a query (Fetches

are performed only for SELECT statements.)

Copyright © Oracle Corporation, 2002. All rights reserved.

Output of the TKPROF Command

The TKPROF output lists the statistics for a SQL statement by the SQL processing step.
The step for each row that contains statistics is identified by the value of the call column.
PARSE This step translates the SQL statement into an execution plan and includes
checks for proper security authorization and checks for the existence of
tables, columns, and other referenced objects.
EXECUTE This step is the actual execution of the statement by the Oracle9i Server.
For INSERT, UPDATE, and DELETE statements, this step modifies the
data (including sorts when needed). For SELECT statements, this step
identifies the selected rows.
FETCH This step retrieves rows returned by a query, and sorts them when needed.
Fetches are performed only for SELECT statements.

Note: The PARSE value includes both hard and soft parses. A hard parse refers to the
development of the execution plan (including optimization); it is subsequently stored in the
library cache. A soft parse means that a SQL statement is sent for parsing to the kernel, but
the kernel finds it in the library cache, and only needs to verify things such as access rights.
Hard parses can be expensive, in particular due to the optimization; a soft parse is mostly
expensive in terms of library cache activity.

Oracle9i: SQL Tuning Workshop 4-12

 Output of the TKPROF Command

Seven categories of trace statistics:

Count Number of times procedure was executed

CPU Number of seconds to process
Elapsed Total number of seconds to execute
Disk Number of physical blocks read
Query Number of logical buffers read for consistent read
Current Number of logical buffers read in current mode
Rows Number of rows processed by the fetch or execute

Copyright © Oracle Corporation, 2002. All rights reserved.

Oracle9i: SQL Tuning Workshop 4-13

Output of the TKPROF Command (continued)
Next to the CALL column, TKPROF displays the following statistics for each statement:
Count Number of times a statement was parsed, executed, or fetched (Check this
column for values >1 before interpreting the statistics in the other columns.
Unless the AGGREGATE = NO option is used, TKPROF aggregates
identical statement executions into one summary table.)
CPU Total CPU time in seconds for all parse, execute, or fetch calls
Elapsed Total elapsed time in seconds for all parse, execute, or fetch calls
Disk Total number of data blocks physically read from the data files on disk for
all parse, execute, or fetch calls
Query Total number of buffers retrieved in consistent mode for all parse, execute,
or fetch calls (Buffers are usually retrieved in consistent mode for queries.)
Current Total number of buffers retrieved in current mode (Buffers typically are
retrieved in current mode for DML statements. However, segment header
blocks are always retrieved in current mode.)
Rows Total number of rows processed by the SQL statement (This total does not
include rows processed by subqueries of the SQL statement. For SELECT
statements, the number of rows returned appears for the fetch step. For
UPDATE, DELETE, and INSERT statements, the number of rows processed
appears for the execute step.)
Note:
• DISK is equivalent to physical reads from v$sysstat or AUTOTRACE.
• QUERY is equivalent to consistent gets from v$sysstat or AUTOTRACE.
• CURRENT is equivalent to db block gets from v$sysstat or AUTOTRACE.

Oracle9i: SQL Tuning Workshop 4-14

 Output of the TKPROF Command

The TKPROF output also includes the following:

• Recursive SQL statements
• Library cache misses
• Parsing user ID
• Execution plan
• Optimizer mode or hint

Copyright © Oracle Corporation, 2002. All rights reserved.

Recursive Calls
Sometimes in order to execute a SQL statement issued by a user, the Oracle server must
issue additional statements. Such statements are called recursive SQL statements. For
example, if you insert a row into a table that does not have enough space to hold that row,
the Oracle server makes recursive calls to allocate the space dynamically. Recursive calls
are also generated when data dictionary information is not available in the data dictionary
cache and must be retrieved from disk.
If recursive calls occur while the SQL trace facility is enabled, TKPROF marks them
clearly as recursive SQL statements in the output file. You can suppress the listing of
recursive calls in the output file by setting the SYS=NO command-line parameter. Note
that the statistics for recursive SQL statements are always included in the listing for the
SQL statement that caused the recursive call.
Library Cache Misses
TKPROF also lists the number of library cache misses resulting from parse and execute
steps for each SQL statement. These statistics appear on separate lines following the
tabular statistics.
Parsing User ID
This is the ID of the last user to parse the statement.

Oracle9i: SQL Tuning Workshop 4-15

Execution Plan
If you specify the EXPLAIN parameter on the TKPROF command line, TKPROF uses the
EXPLAIN PLAN command to generate the execution plan of each SQL statement traced.
TKPROF also displays the number of rows processed by each step of the execution plan.
Note: Be aware that the execution plan is generated at the time that the TKPROF command
is run and not at the time the trace file was produced. This could make a difference if, for
example, an index has been created or dropped since tracing the statements.
Optimizer Mode or Hint
This indicates the optimizer hint that is used during the execution of the statement. If there
is no hint, then it shows the optimizer mode that is used. Hints and optimizer modes are
discussed in more detail in the “Influencing the Optimizer” lesson.
Instructor Note
Note that TKPROF output also shows the row source operators (RSO) before the (optional)
full execution plan. Note that these row source operators are derived from the trace file
STAT lines without accessing the database, as opposed to the additional execution plan
generated with the EXPLAIN option.

Oracle9i: SQL Tuning Workshop 4-16

 TKPROF Output Example: No Index

...
select cust_first_name, cust_last_name, cust_city, cust_state_province
from customers
where cust_last_name = ’Smith’

call count cpu elapsed disk query current rows

------- ------ -------- ---------- ------- ---------- ---------- ----------
Parse 3 801.14 800.00 0 0 1 0
Execute 3 0.00 0.00 0 0 0 0
Fetch 21 1802.59 9510.00 48 4835 10 234
------- ------ -------- ---------- ------- ---------- ---------- ----------
total 27 2603.73 10310.00 48 4835 11 234

Misses in library cache during parse: 3

Optimizer goal: CHOOSE
Parsing user id: 44

Rows Row Source Operation

------- ---------------------------------------------------
78 TABLE ACCESS FULL CUSTOMERS
...

Copyright © Oracle Corporation, 2002. All rights reserved.

Example of TKPROF Output: No Index

The example in the slide shows 234 rows being fetched from the CUSTOMERS table. It
requires 4835 reads and 2603.73 seconds of CPU fetch time. The statement is executed
through a full table scan of the CUSTOMERS table, as you can see using the EXPLAIN
command-line option.
The statement must be optimized.
Note: If cpu or elapsed values are 0, then timed_statistics is not turned on.
Instructor Note
Demo script: demo04_02.sql
Note that this script could take some time to complete (approximately 3 minutes).
You are prompted from the script dai.sql to drop all indexes. Enter CUSTOMERS when
prompted.
After running the script, go to the OS, and use the TKPROF command to format the results.

Oracle9i: SQL Tuning Workshop 4-17

 TKPROF Output Example: Unique Index

select cust_first_name, cust_last_name, cust_city, cust_state_province

from customers
where cust_last_name = ’Smith’

call count cpu elapsed disk query current rows

------- ------ -------- ---------- ------- ---------- ---------- ----------
Parse 1 0.00 0.00 0 0 0 0
Execute 1 0.00 0.00 0 0 0 0
Fetch 7 100.14 1500.00 2 87 0 78
------- ------ -------- ---------- ------- ---------- ---------- ----------
total 9 100.14 1500.00 2 87 0 78

Misses in library cache during parse: 1

Optimizer goal: CHOOSE
Parsing user id: 44

Rows Row Source Operation

------- ---------------------------------------------------
78 TABLE ACCESS BY INDEX ROWID CUSTOMERS
78 INDEX RANGE SCAN (object id 32172)

Copyright © Oracle Corporation, 2002. All rights reserved.

Example of TKPROF Output: Unique Index

The same row is retrieved, but in this case I/O is substantially reduced to only 87 blocks
read. These results are achieved because the statement uses an index.
You can achieve significant improvements in performance by sensible indexing. Identify
areas for potential improvement using the SQL Trace facility.
Note: Indexes should not be built until required. Indexes slow down the processing of
INSERT, UPDATE, and DELETE commands, because references to rows must be added,
changed, or removed. Unused indexes should be removed. If possible, process all
application SQL through EXPLAIN PLAN and remove any indexes that are not
referenced.
Instructor Note
This demonstration can be done using either SQL*Plus AUTOTRACE or TKPROF:
Demo script: demo04_03.sql
Note that this script may take some time to complete (approximately 3 minutes).
This script creates an index on the CUST_LAST_NAME column.
After running the script, go to the OS, and use the TKPROF command to format the
results.

Oracle9i: SQL Tuning Workshop 4-18

 Some TKPROF Interpretation Pitfalls

• Read consistency trap

• Schema trap
• Time trap
• Trigger trap

Copyright © Oracle Corporation, 2002. All rights reserved.

Read Consistency Trap

Sometimes other transactions hold uncommitted changes against a table. This increases the
number of block visits because additional blocks must be constructed and visited for read
consistency.
Schema Trap
The TKPROF statistics show a high number of block visits whereas the execution plan
indicates an index access. Especially if the output shows a nonzero value for current, the table
is probably accessed by a full table scan (compare the current columns on the two previous
pages). The index is probably built after the trace file is produced, or the table and column
statistics may be recomputed.
Time Trap
If a particular DML statement shows high timing statistics, the explanation may be
interference from another transaction holding (shared) locks on the table. That is why CPU
time is a better indicator than elapsed time.
Trigger Trap
The resources reported for a statement include those for all of the SQL issued while the
statement was being processed. Therefore, they include any resources used within a trigger,
along with the resources used by any other recursive SQL (such as SQL used for space
allocation).
Oracle9i: SQL Tuning Workshop 4-19
 Summary

In this lesson, you should have learned how to:

• Set SQL Trace initialization parameters
– SQL_TRACE, TIMED_STATISTICS
– USER_DUMP_DEST, MAX_DUMP_FILE_SIZE
• Switch on SQL Trace for a session
ALTER SESSION set sql_trace = true
dbms_session.set_sql_trace(…)
dbms_system.set_sql_trace_in_session(…)

• Format trace files with TKPROF

• Interpret the output

Copyright © Oracle Corporation, 2002. All rights reserved.

Summary
This lesson introduced you to the SQL Trace facility. By using SQL Trace, you can evaluate the
performance of SQL statements. After SQL Trace is turned on, you view execution plans and
statistics on SQL statements generated in a trace file.
To use SQL Trace:
• Identify the appropriate initialization parameters that you want to use
• Turn SQL Trace on
• Format the trace file that is generated by using TKPROF
• Interpret the output of TKPROF

Oracle9i: SQL Tuning Workshop 4-20

 Practice Overview

This practice covers the following topics:

• Viewing and changing initialization parameters
• Using the SQL Trace facility
• Configuring SQL Trace
• Using TKPROF to format trace files
• Interpreting TKPROF results

Copyright © Oracle Corporation, 2002. All rights reserved.

Practice Overview
In this practice, you view and change initialization parameters that influence tracing SQL
statements. Then you use the SQL_TRACE facility to analyze a SQL statement. The results of
the analysis are written to a trace file. Find the trace file and use TKPROF to format the trace
statistics. Interpret the formatted trace results.

Oracle9i: SQL Tuning Workshop 4-21

Practice 4
1. Check the following initialization parameters:
– TIMED_STATISTICS
– MAX_DUMP_FILE_SIZE
– USER_DUMP_DEST
– SQL_TRACE
Make sure that both TIMED_STATISTICS and SQL_TRACE are set to TRUE for
your session. Turn SQL*Plus AUTOTRACE off.
2. Execute the following SQL command:
SQL> select prod_name, prod_desc, supplier_id
2 from products
3 where prod_id = 200;
3. Now disable tracing for your session and try to find your trace file. Open the
trace file with an operating system text editor and investigate the contents of
your trace file.
Note: The edit command in SQL*Plus will default to the "ex" editor. To use vi, type
the command define_editor=vi in your SQL*Plus session.
4. Start TKPROF without command-line arguments. TKPROF displays a usage
message, listing all command-line options.
5. Start TKPROF again, this time with the name of your trace file as an argument.
Also specify a name for the TKPROF output report. Check the result with an
operating system editor again, and see how the readability is improved.
Note: If you are using UNIX to edit the file, use the command vi filename.
6. Finally, start TKPROF in such a way that execution plans are added to the
report and recursive SQL statements are suppressed.

Oracle9i: SQL Tuning Workshop 4-22

 Rule-Based Optimization
Versus Cost-Based Optimization

Copyright © Oracle Corporation, 2002. All rights reserved.

Schedule: Timing Topic

30 minutes Lecture
15 minutes Practice (optional)
45 minutes Total
Instructor Note
This short lesson introduces both the rule-based optimizer and the cost-based optimizer, and
discusses how to influence the optimizer mode at the instance and the session level. Do not
spend too much time on the rule-based optimizer.
 Objectives

After completing this lesson, you should be able to do

the following:
• Describe the functions of the Oracle9i optimizer
• Distinguish between rule-based optimization (RBO)
and cost-based optimization (CBO)
• Identify when RBO and CBO are used
• Identify the factors that CBO considers
• Identify and influence RBO behavior
• Set the optimizer approach at the instance and
session level

Copyright © Oracle Corporation, 2002. All rights reserved.

Objectives
After completing this lesson, you should be able to:
• Describe the functions of the Oracle9i optimizer
• Distinguish between RBO and CBO
• Identify how the optimizer chooses between RBO and CBO
• Identify the factors that CBO considers when it selects an execution plan
• Identify RBO behavior and the use of the RBO ranking scheme
• Influence RBO behavior
• Set the optimizer approach at the instance and session level
Instructor Note
This lesson introduces the Oracle9i optimizer and explains the differences between RBO and
CBO and the default behavior when choosing between RBO and CBO. This is the only lesson
that specifically addresses RBO behavior. Statistics collection and how to influence CBO are
covered later in the course.

Oracle9i: SQL Tuning Workshop 5-2

 Overview

• Rule-based optimization (RBO)

• Cost-based optimization (CBO)
• Setting an optimizer approach
• The RBO ranking scheme
• Influencing the RBO

Copyright © Oracle Corporation, 2002. All rights reserved.

Overview
There are various methods and access paths that can be used to execute a SQL statement. It is
the optimizer’s job to choose which ones to use.
• RBO decides based on a set of rules.
• CBO bases this choice on statistics that it holds about the tables.
This lesson explains how the Oracle optimizer decides between these two approaches.
To use cost-based optimization, you should collect statistics about the tables involved. This is
covered in a later lesson. Cost-based optimization can be influenced in several sophisticated
ways; this is also covered in a later lesson.
Although it is still possible to use rule-based optimization, you should not use this method.
This lesson shows how rule-based optimization uses a ranking scheme to produce execution
plans. You will also see coding techniques that have been popular in the past to influence rule-
based optimization.
Finally, this lesson details the Oracle9i optimizer’s four basic approaches, which can be set at
the instance and session level.

Oracle9i: SQL Tuning Workshop 5-3

 Functions of the Oracle9i Optimizer

• Evaluates expressions and conditions

• Transforms statements into equivalent statements
• Decides how to access the data
• Decides how to join tables
• Decides which path is most efficient

Copyright © Oracle Corporation, 2002. All rights reserved.

Functions of the Oracle9i Optimizer

The optimizer is the part of the Oracle9i Server that creates the execution plan for a SQL
statement. An execution plan is a series of operations that are performed in sequence to execute
the statement. Most of these steps were discussed in the “SQL Statement processing” lesson.
The optimizer uses various pieces of information to work out the best path:
• Hints supplied by the developer
• Statistics
• Information in the dictionary
• The WHERE clause
Usually, the optimizer works in the background. However, with diagnostic tools such as
EXPLAIN, and SQL*Plus AUTOTRACE, you can see the decisions that the optimizer makes.

Oracle9i: SQL Tuning Workshop 5-4

 Rule-Based Optimization

• Is supported for backward compatibility

• Is syntax driven
• Uses no statistics about the data
• Calculates no costs

Copyright © Oracle Corporation, 2002. All rights reserved.

Rule-Based Optimization
Rule-based optimization is supported in Oracle9i, but only for compatibility reasons with
Oracle version 6 and earlier. If you still have existing OLTP applications developed and
carefully tuned using Oracle version 6, you might want to continue using rule-based
optimization when you upgrade these applications to Oracle9i.
Rule-based optimization is syntax driven, which means that changing the syntax of SQL
statements could change the performance. Be aware of the following:
• No statistics are used, so the optimizer does not know anything about the number of rows in
a table and the values in the table columns.
• No costs of execution plans are calculated; the decision about an execution plan is fully
based on a set of rules.
Note: When using diagnostic tools such as EXPLAIN and SQL*Plus AUTOTRACE, the rule-
based optimization approach is easy to recognize: the COST will always display a NULL value.
As soon as a COST value appears, you know that the cost-based optimizer approach is used.

Oracle9i: SQL Tuning Workshop 5-5

 Cost-Based Optimization

• Was introduced with Oracle7

• Is statistics driven
• Bases decisions on cost calculation
– Number of logical reads
– Network transmissions
• Is continuously enhanced
• Typically outperforms RBO
• Estimates CPU usage. In Oracle9i cost includes
estimated usage of CPU for SQL functions and
operators.

Copyright © Oracle Corporation, 2002. All rights reserved.

Cost-Based Optimization
Oracle7 was the first release that implemented cost-based optimization. Unlike rule-based
optimization, the cost-based optimizer bases its decisions on cost estimates, which in turn are
based on statistics stored in the data dictionary.
In calculating the cost of execution plans, cost-based optimization considers the following:
• The number of logical reads (This is the most important factor.)
• CPU utilization
• Network transmissions
Cost-based optimization is continually enhanced with each new Oracle server release, and
several newer features—such as hash joins, star queries, histograms, and index-organized
tables—are available only to cost-based optimization.
With Oracle9i the estimated usage of CPU for SQL functions and operators is included in the
overall estimate of computer resources usage, together with disk I/O and memory, and is used
to compute the cost of access paths and join orders. Estimate of network usage is also included
when data is shipped between query servers running on different nodes.
The result is an improved accuracy of the cost and size models used by the query optimizer.
This helps the optimizer produce better execution plans, improving query performance.

Oracle9i: SQL Tuning Workshop 5-6

Cost-Based Query Optimization Challenges
The task of the optimizer is to bridge the gap between WHAT is specified in a SQL statement
and HOW this can be solved by the execution engine of the database.
For a single SQL statement there can be a huge number of alternative execution plans that all
provide the correct answer but differ significantly in execution time. The challenge for the
optimizer is to quickly find a cost-effective plan. Here it should balance the search time and the
predicted execution time to find the best plan so far.
To estimate the cost of an execution plan it uses a costmodel and the obtained statistics. The
costmodel consists of a costfunction for each possible operator of an execution plan and a
mechanism for estimating the intermediate result sizes.
The problem of query optimization is known to be a mathematically complex problem. And
because the decisions made by the optimizer are based on statistics and estimations it can never
guarantee to find the most cost-effective plan possible. At best it can try to find a reasonably
cost-effective plan within a reasonable amount of time.

Oracle9i: SQL Tuning Workshop 5-7

 Choosing Between RBO and CBO

• Oracle9i default optimizer behavior:

– RBO when no statistics are available
– CBO when statistics are available for at least one
referenced object
• Behavior can be influenced at the following levels:
– Instance level
– Session level
– Statement level

Copyright © Oracle Corporation, 2002. All rights reserved.

Choosing Between RBO and CBO

The default Oracle9i optimizer behavior is as follows:
• When no statistics are available, rule-based optimization is used.
• When statistics are available for at least one referenced object, cost-based optimization is
used; any missing statistics are estimated or replaced by hard-coded, built-in values.
Influencing the Default Behavior
This default behavior can be influenced—for example, to force rule-based optimization under
all circumstances—at three different levels:
• Instance level
• Session level
• Statement level
Influencing the optimizer at the statement level is discussed in the “Influencing the Optimizer”
lesson.
Note: If any table referenced in a query has a degree of parallelism greater than 1, CBO will
always be used regardless of any hints or settings. Also make sure that you have table statistics,
otherwise collecting column statistics for a table will not result in cost-based optimization.

Oracle9i: SQL Tuning Workshop 5-8

 Setting the Optimizer Approach

• At the instance level, set the following parameter:

OPTIMIZER_MODE =
{CHOOSE|RULE|FIRST_ROWS_n|ALL_ROWS}

• For a session, use the following SQL command:

ALTER SESSION SET optimizer_mode =
{choose|rule|first_rows(_n)|all_rows}

Copyright © Oracle Corporation, 2002. All rights reserved.

Setting the Optimizer Approach

Oracle9i has four basic optimizer approaches:
CHOOSE Behaves as described earlier in this lesson: cost-based when statistics
are available, rule-based when they are not (which is the default value)
RULE Forces rule-based optimization regardless of the presence of statistics
FIRST_ROWS[_n] Optimizes to reduce time until the first result comes back (such as for
interactive applications) n=1, 10, 100, or 1000
ALL_ROWS Optimizes overall throughput (such as for reports and batch jobs)
Optimizer Approach Levels
At the instance level, you can set the optimizer approach by using the OPTIMIZER_MODE
initialization parameter. When this parameter is not set, it defaults to CHOOSE. When statistics
are available, CHOOSE is equivalent to ALL_ROWS.
At the session level, the optimizer approach set at the instance level can be overruled by using
the ALTER SESSION command.
Instructor Note
For backward compatibility reasons, OPTIMIZER_GOAL is still supported as a synonym for
OPTIMIZER_MODE in the ALTER SESSION command. However, OPTIMIZER_GOAL is
no longer documented in Oracle9i. There is no need to mention this to the students
Oracle9i: SQL Tuning Workshop 5-9
 First Rows Optimization

OPTIMIZER_MODE =
• FIRST_ROWS_1
• FIRST_ROWS_10
• FIRST_ROWS_100
• FIRST_ROWS_1000
ALTER SESSION
SET OPTIMIZER_MODE = FIRST_ROWS_n
/*+ FIRST_ROWS(x) */

Copyright © Oracle Corporation, 2002. All rights reserved.

New First Rows Optimization

Oracle9i introduces a new way of doing first rows optimization. The old mode was partially
based on heuristics, such as always using an index if possible. These heuristics sometimes led
to bad plans.
The new approach is completely cost based and allows optimizing for a particular number of
first rows, for example, the first 10 rows.
It is invoked as an argument to the optimizer_mode initialization parameter or the
optimizer_goal session parameter, which allows the following values:
• FIRST_ROWS_1
• FIRST_ROWS_10
• FIRST_ROWS_100
• FIRST_ROWS_1000
In addition, the FIRST_ROWS hint now takes a numeric argument that is not limited to the
values for the parameter. For instance, you could specify /*+ FIRST_ROWS(20)*/ as a
hint.
Without a numeric argument, both the parameter and the hint imply the old first rows behavior,
which is retained for backwards compatibility and plan stability.

Oracle9i: SQL Tuning Workshop 5-10

 Rule-Based Optimization

• Uses indexes unconditionally, because it has no

awareness of statistics
• Uses a built-in ranking scheme with scores for
access paths
• Is difficult to influence

Copyright © Oracle Corporation, 2002. All rights reserved.

Rule-Based Optimization
Rule-based optimization always uses indexes when possible, even when tables are small or
when queries typically return a significant percentage of the rows, so that a full table scan is
faster. Rule-based optimization is not aware of statistics, such as the number of rows in a table
or the selectivity of a predicate.
Rule-based optimization uses a ranking scheme to decide which access path to the data to
choose.
Influencing RBO
You cannot influence rule-based optimization, except by creating indexes, dropping indexes, or
making indexes unusable by changing the statement syntax.
If the RBO finds two execution plans that score the same on the ranking scheme, it makes an
arbitrary decision. For example, suppose each available plan can use only one of two different
indexes. In that case the RBO decision is not documented, but dropping and recreating one of
the involved indexes will influence the RBO behavior: it will use the newly created index.
Thus, you should be aware that recreating a series of indexes in a different order will probably
influence the RBO behavior. Alternatively, you can change the SQL statement to make it
impossible for the RBO to use one of the indexes. This is discussed later in this lesson.

Oracle9i: SQL Tuning Workshop 5-11

 RBO Ranking Scheme
Rank Access Path
1 Single row by ROWID
2 Single row by cluster join
3 Single row by hash cluster key with unique or primary key
4 Single row by unique or primary key
5 Cluster join
6 Hash cluster key
7 Indexed cluster key
8 Composite index
9 Single-column index
10 Bounded range search on indexed columns
11 Unbounded range search on indexed columns
12 Sort-merge join
13 MAX or MIN of indexed column
14 ORDER BY on indexed column
15 Full table scan

Copyright © Oracle Corporation, 2002. All rights reserved.

RBO Ranking Scheme

Rule-based optimization applies rules based on the statement syntax to establish the access
path. The optimizer determines all available access paths for a statement and then chooses the
path with the lowest ranking number from the ranking scheme displayed in the slide.
This approach always assumes that a full table scan (rank 15) is the worst access method. This
might not actually be the case, especially with small tables or with queries that return a large
percentage of the rows in a table.
These (and more) access paths are also available to cost-based optimization. However, the
optimizer ignores the rankings and determines the expected cost for each possible access path,
then chooses the one with the smallest estimated cost. Many features (such as hash joins, star
queries, histograms, and index-organized tables) are available only through cost-based
optimization.
Note: For more information about RBO access paths, see the lesson titled “The Optimizer” in
Oracle9i Tuning and Oracle9i Concepts.
Instructor Note (for page 5-13)
Demo script: demo05_01.sql.
This script creates indexes on LAST_NAME, COUNTRY_ID, and CREDIT_LIMIT, and forces
RBO.

Oracle9i: SQL Tuning Workshop 5-12

 Rule-Based Optimization Example

SQL> select cust_last_name

2 from customers
3 where country_id = ’FR’
4 and cust_credit_limit between 11000 and 15000
5 order by cust_credit_limit;

RBO has the following access paths available:

• Full table scan Rank = 15
• ORDER BY on indexed column Rank = 14
• COUNTRY_ID index equality (range) scan Rank = 9
• CUST_CREDIT_LIMIT index bounded-range search
Rank = 10

Copyright © Oracle Corporation, 2002. All rights reserved.

Rule-Based Optimization Example

The rule-based approach chooses a path based on the syntax of the statement. The optimizer
chooses the access path with the lowest ranking number.
If the CUST_LAST_NAME, COUNTRY_ID, and CUST_CREDIT_LIMIT columns are
indexed, then the rule-based optimizer has the following choices:
Full table scan This path is available to all SQL statements. It ranks 15.
ORDER BY This path could make sorting unnecessary. It ranks 14.
Country ID index Use the index on COUNTRY_ID. It ranks 9, a single-column
scan index.
Credit limit index Use the index on CUST_CREDIT_LIMIT. It ranks 10, a
scan search bounded range search.

The lowest rank is 9, so the index on COUNTRY_ID will be used to access the rows of the
CUSTOMERS table. Note that although the index on COUNTRY_ID is used, a filter operation
must be performed on the CUST_CREDIT_LIMIT column to return the correct results.

Oracle9i: SQL Tuning Workshop 5-13

 Influencing Rule-Based Optimization

• Change the table order in the FROM clause

• Change the predicate order in the WHERE clause
• Create or drop indexes
• Suppress index usage with syntax tricks:
– numeric_column + 0
– date_column + 0
– char_column || ’’

Copyright © Oracle Corporation, 2002. All rights reserved.

Influencing Rule-Based Optimization

Rule-based optimization is highly sensitive to syntax. When you join tables, the order in which
the tables are specified in the FROM clause can have a significant performance impact. For
example, you should always put the most restrictive table at the end, because the optimizer
evaluates the FROM clause from right to left. Also, the order of the predicates in a compound
WHERE clause can influence performance.
The optimizer always uses indexes when they are available. Thus it is important to create
appropriate indexes and drop indexes that deteriorate performance. Unfortunately, most of the
unique indexes cannot be dropped because they protect integrity constraints.
The standard way to prevent rule-based optimization from using existing indexes is to use a
trick: Indexes are usable only when the associated column names appear clean in the
predicates. By applying a dummy operator (adding zero or concatenating an empty string), you
suppress the index usage. You will investigate this optimizer behavior in more detail in the
workshops.
Note: Cost-based optimization provides a more flexible, readable, and acceptable method to
influence the optimizer. Modifying predicates in the way described above still works, but is not
recommended. However, you should understand RBO behavior when maintaining or upgrading
existing applications.

Oracle9i: SQL Tuning Workshop 5-14

 Summary

In this lesson, you should have learned:

• Rule-based optimization (RBO)
• Cost-based optimization (CBO)
• Cost factors considered by CBO
• Setting the optimizer approach
• RBO ranking scheme
• Influencing rule-based optimization

Copyright © Oracle Corporation, 2002. All rights reserved.

Summary
This lesson introduced both rule-based and cost-based optimization. Rule-based optimization is
still supported for backward compatibility. You have seen how the RBO uses a ranking
scheme, and how you can influence its behavior.
Cost-based optimization was introduced in Oracle7. The optimizer uses statistics to calculate
the cost of access paths, taking into account the number of logical I/Os, CPU usage, and
network traffic.
You also learned how the optimizer approach can be set at the instance and session level,
supporting four different settings:
• CHOOSE Default behavior: CBO when statistics are available, RBO when not
• RULE Forces RBO behavior
• FIRST_ROWS Optimizes response time for the first result
• ALL_ROWS Optimizes overall throughput

Oracle9i: SQL Tuning Workshop 5-15

 Practice Overview

This practice covers the following topics:

• Analyzing results with RBO
• Altering your session to use CBO

Copyright © Oracle Corporation, 2002. All rights reserved.

Practice Overview
In this practice, you analyze execution plans using RBO. After changing your session to use the
CBO, you generate new execution plans and examine the results.
Note: By default, statistics exist on the tables in the SH schema.

Oracle9i: SQL Tuning Workshop 5-16

Practice 5 (optional)
1. Use the ALTER SESSION statement to force your session to use the RULE
optimizer mode.
2. Run the dai.sql script to drop all nonprimary key indexes on the CUSTOMERS
and COUNTRIES tables.
SQL> @dai
on which table: customers
SQL> @dai
on which table: countries
3. Enable SQL*Plus AUTOTRACE to display execution plans and analyze the SQL
statement shown below.
SQL> select c.cust_last_name
2 , c.cust_first_name
3 , co.country_name
4 , co.country_region
5 from customers c
6 , countries co
7 where c.country_id = co.country_id
8 and co.country_subregion = ’Africa’;
4. Analyze the SQL statement shown below to see which optimizer mode is used.
SQL> SELECT *
2 FROM customers
3 WHERE cust_credit_limit = 15000;
5. Use the ALTER SESSION statement to force your session to use the CHOOSE
optimizer mode. Remember to turn SQL*Plus AUTOTRACE off prior to the ALTER
SESSION command.
6. Turn SQL*Plus AUTOTRACE on. Re-execute the query in step 3 and examine the
results. Are the results any different?
Instructor Note
This practice session is optional. Use it when you want a break between the initial
(theoretical) lessons. Ideally, you should try to go through the first six lessons as quickly
as possible, so that the first workshop can be started. During the three days of this course,
the theory/practice ratio changes considerably.

Oracle9i: SQL Tuning Workshop 5-17

Oracle9i: SQL Tuning Workshop 5-18
 Indexes and Basic Access Methods

Copyright © Oracle Corporation, 2002. All rights reserved.

Schedule: Timing Topic

45 minutes Lecture
45 minutes Total
Instructor Note
Spend some time on constraints and implicit indexes. With every new release, the optimizer
becomes more intelligent in its use and reuse of existing indexes. Instead of relying on unique
constraints to create unique indexes as a side effect, you should manually create all indexes
needed for queries before constraints are enabled.
Note that indexes will be revisited in the “Advanced Indexes” lesson; you can point this out to
the students. This lesson is a short one, so that students can begin the first workshops as soon
as possible.
 Objectives

After completing this lesson, you should be able to do

the following:
• Identify ROWIDs
• Identify index types
• Show the relationship between indexes and
constraints
• Create indexes manually
• Identify index issues with foreign keys
• Identify basic access methods
• Skip scan of indexes
• Monitor index usage
• Describe an overview of clusters

Copyright © Oracle Corporation, 2002. All rights reserved.

Objectives
After completing this lesson, you should be able to do the following:
• Identify ROWIDs
• Identify index types
• Show the relationship between indexes and constraints
• Create indexes manually
• Identify index issues with foreign keys
• Identify basic access methods
• Skip scan of indexes
• Index usage monitoring
• Overview of clusters
Instructor Note (for page 6-3)
Index-organized tables have logical ROWIDs instead of real ROWIDs. This is discussed in the
“Alternative Storage Techniques” lesson.
Demo scripts: demo06_01.sql and demo06_02.sql
Purpose: To show the students ROWIDs.
Oracle9i: SQL Tuning Workshop 6-2
 ROWIDs

• ROWIDs indicate the address of a row.

• Extended and restricted ROWIDs are available.
• Each table has a ROWID pseudocolumn.
SQL> select rowid, c.*
from channels c;

ROWID C CHANNEL_DESC CHANNEL_CLASS

------------------ - ------------ --------------
AAABQDAABAAAJ4yAAA S Direct Sales Direct
AAABQDAABAAAJ4yAAB T Tele Sales Direct
AAABQDAABAAAJ4yAAC C Catalog Indirect
AAABQDAABAAAJ4yAAD I Internet Indirect
AAABQDAABAAAJ4yAAE P Partners Others

Copyright © Oracle Corporation, 2002. All rights reserved.

ROWIDs
Oracle uses ROWIDs to store addresses of table rows, such as in indexes. The extended
ROWID format was introduced with the new partitioning capabilities of Oracle 8.0. A
restricted ROWID data type is still available for backward compatibility. Oracle still uses
restricted ROWIDs internally, when the extended format is not required.
Each Oracle table has a pseudocolumn named ROWID. ROWIDs are not stored in the
database. However, you can create tables that contain columns having the ROWID data type,
although Oracle does not guarantee that the values of such columns are valid ROWIDs.
The extended ROWID has a four-piece format, using a base-64 encoding technique. In the
example in the slide, following are the fields and their meaning:
AAABDL Data object number; identifies the database segment
AAC Relative file number; unique within the tablespace
AAAF0Z Data block number; relative to its data file
AAA-AAE Row numbers in the block
ROWID Manipulation
You can use the DBMS_ROWID package to extract information from an extended ROWID or to
convert a ROWID from extended format to restricted format (or vice versa). See the Oracle9i
Supplied Packages Reference for more information.
Oracle9i: SQL Tuning Workshop 6-3
 Indexes

• Unique and nonunique indexes

• Composite indexes
• Index storage techniques:
– B*-tree
Reverse key
Descending
Function-based
– Bitmap
– Domain indexes

Copyright © Oracle Corporation, 2002. All rights reserved.

Indexes
An index is a database object that is logically and physically independent of the table data. The
Oracle9i Server may use an index to access data required by a SQL statement, or use indexes
to enforce integrity constraints. You can create and drop indexes at any time. The Oracle9i
Server automatically maintains indexes when the related data changes.
Index Types
Indexes can be unique or nonunique. Unique indexes guarantee that no two index entries have
the same value. A composite index (also called a concatenated index) is an index that you
create on multiple columns in a table (up to 32). Columns in a composite index can appear in
any order, and need not be adjacent in the table.
Index Storage Techniques
For standard indexes, Oracle uses B*-tree indexes that are balanced to equalize access times.
Bitmap indexes are discussed in the “Advanced Indexes” lesson.
Reverse key indexes reverse the bytes of the column values. They have benefits when used for
monotonically-increasing values that can lead to skewed indexes in which older values are
dropped over time. They will not be covered in this course.
Function-based indexes are also covered in the “Advanced Indexes” lesson.
Domain indexes are application-specific indexes. They are created, managed, and accessed by
routines supplied by an index type. They will not be covered in this course.
Oracle9i: SQL Tuning Workshop 6-4
 B*-Tree Indexes

Index entry

Root

Branch

Index entry header

Key column length
Leaf Key column value
ROWID

Copyright © Oracle Corporation, 2002. All rights reserved.

A Description of B*-Tree Indexes

Each B*-tree index has a root block as a starting point. Depending on the number of entries,
there are one or more branch block layers. A B*-tree index also consists of a set of leaf blocks
containing all values plus ROWIDs pointing to the rows in the associated data segment. The
leaf blocks are connected by previous and next block pointers, so that they can be traversed
from left to right and vice versa.
Indexes are always balanced and they grow from the bottom and up. In certain situations, the
balancing algorithm can cause the B*-tree height to increase unnecessarily.
You should reorganize your indexes regularly to guarantee optimal storage efficiency and
performance. This can be done using the ALTER INDEX … REBUILD command. Since
Oracle8i it is possible to rebuild some indexes online, which does not impact the performance
of DML statements.
Instructor Note
You should not go into too much detail here. The internal index algorithms are undocumented
and may change (and get smarter) with each new release, without notification. Just point out
what the last sentence above states: after a period of heavy DML activity, rebuilding your
indexes is a good idea.

Oracle9i: SQL Tuning Workshop 6-5

Format of Index Leaf Entries
An index entry is composed of the following components:
• An entry header, which stores the number of columns and locking information
• Key column length-value pairs, which define the column size in the key that is followed
by the column value (the number of such pairs is a maximum of the number of columns
in the index)
• ROWID of a row, which contains the key values
Branch block Characteristics
Branch blocks are for searching and they store the following:
• The minimum key prefix needed to make a branching decision between two keys
• The pointer to the child block that contains the key
If the blocks have n keys then they have n+1 pointers. The number of keys and pointers is
limited by the block size.
Index Leaf Entry Characteristics
• Key values are repeated if there are multiple rows that have the same key value.
• There are no index entries to rows that have all key columns containing NULL values.
• For nonpartitioned tables, restricted ROWIDs are used to point to the rows of the table,
because all rows belong to the same segment.
Instructor Note
Root blocks and branch blocks are compressed by default. The leaf blocks take
approximately 99% of index storage. Oracle9i also allows key compression (for
concatenated indexes only) with the COMPRESS option of the CREATE INDEX or ALTER
INDEX REBUILD commands. See the Oracle9i SQL Reference for more details. However,
you should not offer too much technical detail here.

Oracle9i: SQL Tuning Workshop 6-6

 B*-Tree Index Example
Root block
JONES

Branch
blocks
ALLEN MARTIN
FORD SMITH
Leaf
blocks

ADAMS BLAKE - rowid JAMES KING MILLER TURNER

ALLEN CLARK - rowid JONES MARTIN SCOTT WARD
FORD - rowid SMITH

ROWID Data segment Next/previous

lookup 7782 CLARK leaf pointers
7698 BLAKE
7902 FORD

Copyright © Oracle Corporation, 2002. All rights reserved.

B*-Tree Index Example

The internal structure of a B*-tree index allows rapid access to the indexed values. The
Oracle9i Server can directly access rows after it has retrieved the address (the ROWID) from
the index leaf blocks.
If the optimizer decides to use an index, then the Oracle9i Server searches the index for the
column values accessed by the statement and takes the following steps:
• If the statement accesses only columns that are present in the index, it retrieves the
column values directly from the index (fast full index scan) without going to the data
segment.
• If the statement also accesses other columns, it retrieves the addresses from the index, and
then accesses the data blocks by ROWID.
Instructor Note:
Fast full index scans (also covered on page 14) have the following restrictions:
• At least one indexed column of the table must have the NOT NULL constraint.
• There must be a parallel clause on the index if you want to perform fast full index scan in
parallel. The parallel degree of the index is set independently. The index does not inherit
the degree of parallelism of the table.
• Make sure that you have analyzed the index; otherwise, the optimizer might decide not to
use it.
Oracle9i: SQL Tuning Workshop 6-7
 CREATE INDEX Syntax

UNIQUE
CREATE INDEX index_name
BITMAP
,

ON table ( column )
expression
ASC
DESC

physical attributes NOSORT

REVERSE
TABLESPACE tablespace
NOLOGGING

Copyright © Oracle Corporation, 2002. All rights reserved.

CREATE INDEX Syntax

The NOSORT option can be used only with standard (B*-tree) indexes, and only if the rows in
the table are exactly in the right order. It will skip the sort phase, speeding up the index
creation. If the rows are not in the right order, the following error will be returned:
ORA-01409: NOSORT option may not be used; rows are not in
ascending order
The NOLOGGING option reduces the undo generated during the index creation.
The physical attributes clause contains details about the physical storage for the index to be
created. The details are not discussed in this course.
You can use the column expression to create function-based indexes, which are discussed in
the “Advanced Indexes” lesson. In the most simple case, the column expression is just a single
column name.
With the ASC and DESC options, you can specify whether the index values should be in
ascending or descending order. Oracle treats descending indexes as if they were function-based
indexes. You do not need the QUERY REWRITE or GLOBAL QUERY REWRITE privileges
to create them, as you do with other function-based indexes. However, as with other function-
based indexes, Oracle does not use descending indexes until you first analyze the index and the
table on which the index is defined.
The default index is is nonunique B*-tree if UNIQUE and BITMAP are not specified.
Note: This is not the full CREATE INDEX syntax; refer to the Oracle9i SQL Reference for
more details.
Oracle9i: SQL Tuning Workshop 6-8
 Composite Index Guidelines

• Put the most frequently queried column first (also

called the leading part of the index)
• Put the most restrictive column first
• Add extra columns to the index

Copyright © Oracle Corporation, 2002. All rights reserved.

Composite Index Guidelines

When creating a composite index, use the following (possibly conflicting) guidelines:
• Put the most frequently queried column first (also called the leading part of the index),
because the optimizer can use it as an index in its own right.
• Put the most restrictive column first if you expect to specify the full key.
• Consider adding extra columns to the index, so that you can retrieve query results without
accessing the base table. This requires additional storage capacity but saves I/O.
When creating composite indexes you might consider using the COMPRESS option to
eliminate repeated occurrences of key column values. This can substantially reduce storage.
For example:
SQL> create index cust_name_idx
2 on customers(cust_first_name, cust_last_name)compress 1;
This compresses repeated occurrences of cust_first_name values. NOCOMPRESS is the
default.
Note: Remember that the Oracle server must maintain indexes, thus slowing down DML
statements; see the next page for more details. Create only those indexes that you are sure the
optimizer will use to speed up queries. It is recommended that indexes be placed in tablespaces
separate from tables. This can be done using the storage clause while creating the index. This
will improve performance by reducing I/O and fragmentation.

Oracle9i: SQL Tuning Workshop 6-9

 Index Statistics

Index statistics can be collected with the following

commands:
SQL> analyze index <index_name>
2 validate structure;

SQL> select * from index_histogram;

SQL> select * from index_stats;

Copyright © Oracle Corporation, 2002. All rights reserved.

Index Statistics
In the “Collecting Statistics” lesson, statistics are discussed in detail. You can collect and
retrieve some useful index statistics with the following commands:
SQL> analyze index <index_name> validate structure;
SQL> select * from index_histogram;
SQL> select * from index_stats;
The first command populates the two views that are queried by the last two commands. The
first view gives information about the selectivity of the index (cardinality of key values), and
the second view can tell you the height of the B*-tree, the number of blocks, the number of
distinct key values, and the number of leaf block entries.
ALTER INDEX Statement
You can use the ALTER INDEX statement to change an index definition, or rebuild an existing
index. On the next page you see some guidelines about when this could be needed. See the
Oracle9i SQL Reference for the full ALTER INDEX statement syntax.
Instructor Note
Typically, a DBA uses these views to monitor and maintain statistics.

Oracle9i: SQL Tuning Workshop 6-10

 Effect of DML Operations on Indexes

• Inserts result in the insertion of an index entry in the

appropriate block (Block splits might occur.)
• Deletes result in a logical deletion of the index entry
(Empty blocks become available.)
• Updates to the key columns result in a logical delete
and an insert to the index.

Copyright © Oracle Corporation, 2002. All rights reserved.

Effect of DML Operations on Indexes

The Oracle server maintains all the indexes when DML operations are carried out on the table.
Here is an explanation of the effect of a DML command on an index:
• Insert operations result in the insertion of an index entry in the appropriate block. If there
is no room for the entry in the block, then it is split into two blocks, and an entry is added
in the parent block to point to the new block.
• Deleting a row results only in a logical deletion of the index entry. The space used by the
deleted row is not available for new entries until all the entries in the block are deleted.
• Updates to the key columns result in a logical deletion and an insertion to the index. The
PCTFREE setting has no effect on the index except at the time of creation. A new entry
can be added to an index block even if it has less space than that specified by PCTFREE.
After periods of high DML activity, you should reorganize your B*-tree indexes. This can be
done online in Oracle8i and later. Indexes created (or rebuilt) on existing data are always more
efficient than indexes that are implicitly maintained by the Oracle Server.
Instructor Note
Space released by a deleted index entry can be reused by a new entry if its value falls between
the logically prior and logically subsequent values of the deleted value. A large number of
deletions on an indexed table can lead to a large amount of wasted space. This can negatively
impact performance during operations such as range scans of the index.
Oracle9i: SQL Tuning Workshop 6-11
 Indexes and Constraints

The Oracle server implicitly creates or uses B*-tree

indexes when you define the following:
• Primary key constraints
• Unique constraints

CREATE TABLE channels

( channel_id CHAR(1)
CONSTRAINT chan_pk PRIMARY KEY
, channel_desc VARCHAR2(20)
CONSTRAINT chan_desc_nn NOT NULL
, channel_class VARCHAR2(20)
, channel_total VARCHAR2(13)
);

Copyright © Oracle Corporation, 2002. All rights reserved.

Indexes and Constraints

When you define a primary or unique key constraint on a table, the Oracle database
automatically generates an index, or uses an existing index, to support it. Indexes created for
this purpose are physically, but not logically, independent of the table structure.
The Oracle server gives the index the same name as the constraint.
You cannot drop an index that enforces a constraint. Instead, you must disable the constraint,
in most cases causing the Oracle database to automatically drop the index.
Be aware that the Oracle database supports deferrable constraints. Unique deferrable
constraints are always enforced using nonunique indexes. Also, when disabling such a
constraint, the Oracle server does not drop the associated index, thus offering an easy check
when the constraint is enabled again.
Indexes serve at least two purposes: to make queries faster and to enforce unique and primary
keys. The Oracle optimizer uses existing indexes when new constraints are defined. Note that
the Oracle database can use nonunique indexes to check unique constraints.
Note: Instead of relying on unique constraints to create unique indexes as a side effect, you
should manually create all indexes needed for queries before constraints are enabled, including
unique indexes.
For more information about this topic, see Oracle9i Concepts and Oracle9i Administrator’s
Guide.
Oracle9i: SQL Tuning Workshop 6-12
 Indexes and Foreign Keys

• Indexes are not created automatically.

• There are locking implications to DML activity on
parent-child tables.

CUSTOMERS
# cust_id

CHANNELS SALES PRODUCTS

#channel_id #prod_id

Copyright © Oracle Corporation, 2002. All rights reserved.

Indexes and Foreign Keys

The Oracle database does not automatically create an index for a foreign key constraint.
If the foreign key columns are often used in join conditions, then you should create an index on
them to enhance the join process.
In earlier releases deleting or updating data in parent-child tables (with a foreign key
constraint) caused different locking behavior, depending on the presence of an index on the
foreign key column. If you do not index a foreign key, the child table is locked (with a table-
level lock) whenever the following occurs:
• A row is deleted from the parent table
• The columns pointed to by the foreign key are updated in the parent table
For this reason, you had to index foreign key columns, even if they are not normally used in
join conditions and WHERE clauses.
In Oracle9i this locking behavior is no longer an issue. A lock is temporarily placed, and then
immediately removed, before it can cause any kind of contention.
Note: For more detailed information about indexes, foreign keys, and locking behavior, see the
Oracle9i Application Developer’s Guide, “Maintaining Data Integrity.”

Oracle9i: SQL Tuning Workshop 6-13

 Basic Access Methods

• Full table scans:

– Can use multiblock I/O
– Can be parallelized
• Index scans:
– Allow index access only
– Are followed by access by ROWID
• Fast full index scans:
– Can use multiblock I/O
– Can be parallelized

Copyright © Oracle Corporation, 2002. All rights reserved.

Full Table Scans

A full table scan may involve reading many sequential data blocks from database files on disk
into the buffer cache. However, the Oracle server can read several sequential blocks in a single
I/O. This is referred to as multiblock I/O. You can influence multiblock I/O by setting the
DB_FILE_MULTIBLOCK_READ_COUNT parameter. Using the parallel query capability can
help speed up full table scans by allocating several processes to the work.
Index Scans
Indexes improve the performance of many SQL statements. However, an indexed query may
read from both index and data blocks. These blocks are not usually sequential, so the query
cannot use multiblock reads. Use indexes only when they improve performance.
As a general rule, use B*-tree indexes for queries that select fewer than 5% of the rows from a
table. However, this break-even point varies with your data. Make sure to test this for all
critical processes by using diagnostic tools.
Fast Full Index Scans
The fast full index scan is an alternative to a full table scan when there is an index that contains
all the columns that are needed for a query. It is faster than a normal index scan because it can
use multiblock I/O and can be parallelized in the same way as a table scan. Unlike regular
index scans, however, the rows will not necessarily come back in sorted order. If there is a
predicate that narrows the index search to a range of values, then the optimizer will not
consider a fast full index scan.
Oracle9i: SQL Tuning Workshop 6-14
 Skip Scanning of Indexes

• Index skip scanning enables access through a

composite index when the prefix column is an
unknown value
• Is supported by:
– Cluster indexes
– Descending scans
– CONNECT BY clauses
• Is not supported by reverse key or bitmap indexes

Copyright © Oracle Corporation, 2002. All rights reserved.

Skip Scanning Definition

In prior releases, a composite index would be used only if the index prefix (leading) column
was included in the predicate of the statement. With Oracle9i, the optimizer can use a
composite index even if the prefix column value is not known. The optimizer uses an algorithm
called skip scanning to retrieve ROWIDs for values that do not use the prefix column.
Skip scans reduce the need to add an index to support occasional queries that do not reference
the prefix column of an existing index. This can be useful when high levels of DML activity
are degraded by the existence of too many indexes used to support infrequent queries. The
algorithm is also valuable in cases where there are no clear advantages as to which column to
use as the prefix column in a composite index. The prefix column should be the most
discriminating, but also the most frequently referenced in queries. Sometimes, these two
requirements are met by two different columns in a composite index, forcing a compromise or
the use of multiple indexes.

Oracle9i: SQL Tuning Workshop 6-15

Skip Scanning Definition (continued)
During a skip scan, the B*-tree is probed for each distinct value in the prefix column. Under
each prefix column value, the normal search algorithm takes over. The result is a series of
searches through subsets of the index, each of which appears to result from a query using a
specific value of the prefix column. However, with the skip scan, the value of the prefix column
in each subset is obtained from the initial index probe rather than from the command predicate.
In addition to standard B*-tree indexes, the optimizer can use skip scans to process:
• Cluster indexes
• Descending scans
• Statements with CONNECT BY clauses
Reverse key and bitmap indexes do not support the skip scan algorithm.

Oracle9i: SQL Tuning Workshop 6-16

Skip Scanning Example
Skip scanning is useful when the initial column of the composite index is not specified in the
query. In other words, it is “skipped.”
For example, suppose you have the following index data:
(’F’,98)
(’F’,100)
(’F’,102)
(’F’,104)
(’M’,101)
(’M’,103)
(’M’,105)
Skip scanning is advantageous if there are few distinct values of the leading column of the
composite index and many values of the nonleading key of the index.
The index is split logically in the following two subindexes:
1. The first one has the keys with the value ’F’.
2. The second one has all the values with the value ’M’.
If you have an EMPLOYEES table with a gender column, in the following query,
SELECT *
FROM employees
WHERE empno = 101;
the column gender is skipped. A complete scan of the index is not performed, but the
subindex with the value ’F’ is searched first, then the subindex with the value ’M’ is
searched.

Oracle9i: SQL Tuning Workshop 6-17

 Identifying Unused Indexes

• Oracle9i provides the capability to gather statistics

about the usage of an index.
• Benefits include:
– Space conservation
– Improved performance by eliminating unnecessary
overhead during DML operations

Copyright © Oracle Corporation, 2002. All rights reserved.

Identifying Unused Indexes

Oracle9i provides the capability to gather statistics within the database about the usage of an
index.
Benefits
If you find an index that is never used, you can drop the index and free the space it used.
Indexes must be maintained during most inserts, deletes, and some updates. Eliminating an
unused index eliminates this overhead.
Parsing is simplified when the optimizer has fewer indexes to consider.

Oracle9i: SQL Tuning Workshop 6-18

 Enabling and Disabling the Monitoring of
Index Usage

• To start monitoring the usage of an index:

ALTER INDEX customers_pk MONITORING USAGE;

• To stop monitoring the usage of an index:

ALTER INDEX customers_pk NOMONITORING USAGE;

• V$OBJECT_USAGE and V$OBJECT_STATS contain

information about the usage of an index.

Copyright © Oracle Corporation, 2002. All rights reserved.

Enabling and Disabling the Monitoring of Index Usage

The V$OBJECT_USAGE view displays information about the usage of an index. Each time
the index is altered with the MONITORING USAGE clause, V$OBJECT_USAGE is reset for
the specified index. Previous information is lost, and a new start time is recorded.
The new V$OBJECT_USAGE view supports the identification of unused indexes with the
following columns:
• INDEX_NAME: The index name
• TABLE_NAME: The corresponding table
• MONITORING: Indicates whether monitoring is on or off
• USED: Indicates whether an index has been used during the monitoring time
• START_MONITORING: Time monitoring began on an index
• STOP_MONITORING: Time monitoring stopped on an index

Oracle9i: SQL Tuning Workshop 6-19

 Clusters

• Rows are grouped and physically stored together.

• Tables in a cluster are prejoined.
• Clusters are transparent to users and applications.
• DML and full table scans are more expensive.

Copyright © Oracle Corporation, 2002. All rights reserved.

Clusters: Overview
A cluster is a structure that is used to physically locate rows together because they share
common column values. There are two type of clusters: index clusters and hash clusters.
The columns used to cluster the rows are called the cluster key:
• The cluster key can consist of one or more columns.
• Tables in a cluster have columns that correspond to the cluster key. Clustering is a
mechanism that is transparent to the applications using the tables. Data in a clustered
table can be manipulated as though it were stored in a regular table.
• Updating one of the columns in the cluster key may cause the Oracle server to physically
relocate the row.
• The cluster key is independent of the primary key. The tables in a cluster can have a
primary key, which may be the cluster key or a different set of columns.
Clusters can waste disk space if the cluster key distribution is skewed. Choosing an appropriate
cluster size is one of the most important issues when you consider whether to cluster your
tables.

Oracle9i: SQL Tuning Workshop 6-20

 Cluster Example
CUST_ID CUST_LAST_NAME CO ...
---------- -------------- -- Cluster Key: COUNTRY_ID
10 Kessel UK -------------------------
20 Everett US UK United Kingdom
30 Odenwalld US 10 Kessel
40 Sampson US 60 Rhodes
50 Nenninger FR 280 Durby
60 Rhodes UK 650 Williamson
…
-------------------------
US United States of America
CO COUNTRY_NAME ... 20 Everett
-- ------------------------- 30 Odenwalld
US United States of America 40 Sampson
DE Germany ...
UK United Kingdom
NL The Netherlands

Unclustered CUSTOMERS and Clustered CUSTOMERS and

COUNTRIES tables COUNTRIES tables

Copyright © Oracle Corporation, 2002. All rights reserved.

Cluster Example
If they are stored as regular tables, CUSTOMERS and COUNTRIES are placed in different
segments. This means that the tables use their own set of blocks—a block that stores rows from
the CUSTOMERS table does not contain data from the COUNTRIES table, and vice versa.
Because customers are usually accessed by COUNTRY_ID, the developer may cluster both
tables or just the CUSTOMERS table on the COUNTRY_ID column. If the tables CUSTOMERS
and COUNTRIES are stored in a cluster, they share the same cluster segment; see the example
in the slide. A block in this segment stores rows from both tables. This means that a full table
scan on the COUNTRIES table takes more time, because its rows are interspersed with the
rows from the CUSTOMERS table.
If a table is stored in a cluster, the cluster becomes the physical unit of storage, and the table is
a logical entity; that is, the clustering is transparent to the users and applications.
Instructor Note
These two tables are not ideal for clustering; the slide is meant only for illustration purposes.
Note that this is an index cluster.

Oracle9i: SQL Tuning Workshop 6-21

 Index Clusters

• Cluster index required

to access rows
Cluster
index • NULL entries included

Cluster

Copyright © Oracle Corporation, 2002. All rights reserved.

Index Clusters
An index cluster uses a special index, known as the cluster index, to maintain the data within
the cluster. In the example on the previous page, when a user inserts a new customer, the
cluster index ensures that the new customer is placed in the same block as the other customers
with the same country_id.
Choose Appropriate Columns for the Cluster Key
Choose cluster key columns carefully. If you use multiple columns in queries that join the
tables, make the cluster key a composite key. In general, the same column characteristics that
make a good index apply to cluster indexes.
A good cluster key has enough unique values so that the group of rows corresponding to each
key value fills approximately one or two data blocks. Too many rows per cluster key value can
require extra searching to find rows for that key. Cluster key values that are too general (such
as MALE and FEMALE) result in excessive searching and can result in poor performance.
Instructor Note
The “Choose Appropriate Columns for the Cluster Key” paragraph is also true for hash
clusters.

Oracle9i: SQL Tuning Workshop 6-22

 Index Clusters: Performance
Characteristics

• Can save disk space

• Can reduce disk I/O
• Improve range scan access time
• Improve join performance for clustered tables

Copyright © Oracle Corporation, 2002. All rights reserved.

Index Clusters: Performance Characteristics

Index clusters can save disk space, because each cluster key value is stored only once for all
the rows that have the same key value.
When accessing rows with a common value, disk I/O is reduced and access time improves.
Tables that are often joined using a foreign key relationship are good choices for an index
cluster. Because rows with a given key value are stored together, using a cluster can reduce the
number of blocks read to satisfy such a request.
In some situations, it is useful to place only the detail table in a cluster. Consider a case where
customer information is always accessed by CUST_ID, and full table scans of the
CUSTOMERS table are frequent. In this case, placing only the COUNTRIES table in an index
cluster may be the best solution, thus the full table scans of the CUSTOMERS table do not
suffer any performance degradation.
The cluster index must be available to store, access, or maintain data in an index cluster. The
cluster index points to the block that contains the rows with a given key value. The structure of
a cluster index is similar to that of a normal index. Although a normal index does not store
NULL key values, cluster indexes do store NULL keys. There is only one index entry for each
key value in the cluster index. Therefore, cluster indexes are likely to be smaller than a normal
B*-tree index on the same set of data.
To store or retrieve rows from a cluster, the Oracle server uses the cluster index to locate the
first row that corresponds to the given key value and then retrieves the rows for the given key.

Oracle9i: SQL Tuning Workshop 6-23

 Index Clusters: Limitations and Guidelines

• DML does not perform as well.

• Avoid updating cluster key values.
• Index clusters are more suitable when the
occurrence of the key value is uniform.
• Index clusters are less suitable where large
amounts of data are being loaded.

Copyright © Oracle Corporation, 2002. All rights reserved.

DML Performance
DML does not perform as well on a cluster as it does with a table. When inserting a row into a
cluster, it must be inserted into the same block with the other rows with the same cluster key. If
several rows with different cluster keys are being inserted, then several blocks must be read.
However, for a regular table, the rows can be inserted into the single block that is the first
block on the free list.
Cluster Key Column Updates
When the key value in a cluster is updated, the row may need to be moved physically to
another block. This can cause degradation in performance. Therefore, columns that are
frequently updated are not good candidates for being part of the cluster key.
Row Frequency
Clusters are generally more suitable in situations where the frequency of occurrence of key
values is uniform. In the COUNTRIES example, if the countries contain almost an equal
number of customers, then clustering may be useful. If a certain country has only one or two
customers and others have several hundred, clustering may not be justified. If a country has
many customers, then some of the customer rows may need to be stored in overflow blocks.
Direct Loads
Clusters cannot be loaded using direct loads.

Oracle9i: SQL Tuning Workshop 6-24

 Hash Clusters

• Use a hash function

• Can bypass index I/O
Hash
function • May affect performance for range
scans or full table scans
• Should be used when:
– Optimizing queries with equality
(=) operator
– Tables are large
– Key has even row distribution

Copyright © Oracle Corporation, 2002. All rights reserved.

Hash Clusters
A hash cluster uses a hash function to calculate the location of a row. The hash function uses
the cluster key and can either be system generated or defined by the developer.
When a row is inserted into a table in a hash cluster:
• The hash key columns are used to compute a hash value
• The row is stored based on the hash value
The hash function is also used to find the row while retrieving the data from a hashed table.
Example
Customer rows are stored in a hash cluster using the Customer ID as the cluster key. When
Customer 501 is inserted into the cluster, the hash function determines where to store the row
using the cluster key value 501. Later, when Customer 501 is needed for retrieval, the cluster
key value 501 is input into the hash function, which points to the location where the row was
inserted.
Hash Cluster Performance Characteristics
The hash function is used to locate the row while retrieving the data from a hashed table
without the I/O and CPU time required to search through an index. Because large tables have
indexes with more B*-tree levels, hash clusters perform better when there are many rows in the
table.
Oracle9i: SQL Tuning Workshop 6-25
 Hash Clusters: Limitations

• Range scans cannot use the hash function.

• Unevenly distributed cluster key values can cause
overflow, which implies additional I/O.
• Volatile and unpredictable data can result in an
inefficient hash function.
• Full table scans are generally slower on clustered
tables.

Copyright © Oracle Corporation, 2002. All rights reserved.

Hash Clusters: Limitations

The hash function can be used to retrieve data only if the key value is known. For this reason, a
range search cannot use the hash function. If queries use equality predicates on the cluster key,
they can benefit from hashing. However, it is possible to create an index on the hash cluster
key if many queries use range searches on the hash cluster key columns.
If cluster keys are not evenly distributed throughout the hash cluster, then many of them hash
into the same blocks. When these blocks fill, rows get stored into overflow blocks. It takes
additional I/Os to access these rows.
In a hash cluster, the hash function is based on the parameters specified by the developer at the
time of its creation. If it is difficult to predict the key values or their size accurately because of
the volatility of the data, a wrong choice could result in an inefficient hash function, causing
either a lot of overflow or a waste of space. Therefore, hashing is useful only if the number of
rows per cluster key and the number of cluster keys are predictable.
The hashing algorithm uses the number of distinct cluster keys that will be stored in the cluster
when it determines where to store the row. There must be enough space allocated to the cluster
to store all of these rows. Because rows can be stored anywhere in the cluster, the Oracle server
reads the entire cluster when performing a full table scan, even though there may be few rows
in the cluster. Contrast this with a table that is loaded from the front to the back. The Oracle
server only needs to read to the block that stores the last row in the table. Therefore, full table
scans are generally slower on clustered tables than on nonclustered tables.
Oracle9i: SQL Tuning Workshop 6-26
 When to Use Clusters

• Use clusters:
– With tables that are primarily queried or joined on the
cluster key
– To even cluster key distribution
• Clusters can also reduce performance:
– DML statements
– Full table scans

Copyright © Oracle Corporation, 2002. All rights reserved.

When to Use Clusters

Use clusters to store one or more tables that are queried primarily on the cluster key (not
predominantly inserted or updated), and for which queries often join data of multiple tables in
the cluster or retrieve related data from a single table.
All of the rows for a single cluster key can fit into one or two Oracle blocks. If not, the Oracle
server must read multiple blocks when querying using the cluster key. To identify data that
would be better stored in clustered form than nonclustered, look for tables that are related
through referential integrity constraints and for tables that are frequently accessed together
using queries that join data from two or more tables.
If you cluster tables on the columns used to join table data, you reduce the number of data
blocks that must be accessed to process the query; all the rows needed for a join on a cluster
key are in the same block. Therefore, query performance for joins is improved.
Like indexes, clusters do not affect application design. The existence of a cluster is transparent
to users and applications. Data stored in a clustered table is accessed through SQL just like data
stored in a nonclustered table. Clusters can reduce the performance of:
• DML statements (INSERT, UPDATE, and DELETE) (Because of the overhead of
maintaining multiple tables in a single block, DML against a table is more efficient than
DML against a cluster.)
• Full table scans (When performing a full table scan on a cluster with multiple tables, all
of the tables are read, because rows from the tables are stored in the same blocks.)
Oracle9i: SQL Tuning Workshop 6-27
 When to Use Clusters

• When to use index clusters:

– Appropriate physical data structure
– Hash cluster cannot be used
• When to use hash clusters:
– Query performance for equality searches
is the most important goal
– Tables are large

Copyright © Oracle Corporation, 2002. All rights reserved.

When to Use Index Clusters

Consider using index clusters in those situations where the cluster is the appropriate physical
data structure, but a hash cluster cannot be used because of its limitations.
When to Use Hash Clusters
Consider hash clusters in these situations:
• SELECT performance is the primary tuning goal.
• The rows in the hash cluster are primarily queried with the equality (=) operator.
• The table is large, so a B*-tree index would become too high. Accessing rows through a
hash function bypasses index I/O. With a large table, a hash cluster can be chosen as the
data structure, even if the rows do not need to be clustered. In this situation, the hash
cluster is used because it eliminates index I/Os, not because it clusters rows.
• Cluster keys are evenly distributed throughout the hash cluster.
Do not use hash clustering when you access data often without an equality comparison on the
cluster key:
• Range scans cannot use the hashing algorithm. Thus, this type of access causes full table
scans. However, if range scans are required, the developer can create an additional index.
• Full table scans may perform poorly because hash clusters tend to have a lot of unused
space that is read during full table scans.
Oracle9i: SQL Tuning Workshop 6-28
 Summary

In this lesson, you should have learned about the

following:
• ROWIDs
• Indexes
– Index types
– B*-tree index structure
– DML operations and indexes
– Indexes and constraints
• The CREATE INDEX syntax

Copyright © Oracle Corporation, 2002. All rights reserved.

Summary
This lesson introduced you to indexes. An index is a database object that is logically and
physically independent of the table data. The Oracle server may use an index to access data
required by a SQL statement, or use indexes to enforce integrity constraints.
There are different index types. For standard indexes, Oracle uses B*-tree indexes that are
balanced to equalize access times. You can create an index manually by using the CREATE
INDEX syntax, or automatically by creating unique or primary key constraints on tables.
Indexes can impact performance. There are different methods to scan a table: full table scans,
index scans, and fast full index scans. Index scans can improve the performance of many SQL
statements.

Oracle9i: SQL Tuning Workshop 6-29

 Summary

• Basic access methods

– Full table scans
– B*-tree index scans
– Access by ROWID
– Fast full index scans
– Skip scan of indexes
– Index usage monitoring
– Overview of clusters

Copyright © Oracle Corporation, 2002. All rights reserved.

Instructor Note
After completing this lesson, it is possible to do workshops 1 (single table, single predicate)
and 2 (sorting, grouping, and set operators). Note that the next lesson has a short practice
session. Thus, a typical approach might be to do workshop 1 now, continue with the next two
lessons, and then do workshop 2. But, as indicated, you can do workshop 2 earlier as well.

Oracle9i: SQL Tuning Workshop 6-30

 Collecting Statistics

Copyright © Oracle Corporation, 2002. All rights reserved.

Schedule: Timing Topic

45 minutes Lecture
20 minutes Practice
65 minutes Total
Instructor Note
The main topics in this lesson are statistics, how to collect statistics with the ANALYZE
command, using DBMS_STATS, predicate selectivity, and histograms. The short practice
session reviews collecting and viewing some statistics. There is a new utility called statspack
which is similar to utlbstat/utlestat scripts, which collect all system and application statistics.
This is not covered in this course.
 Objectives

After completing this lesson, you should be able to

do the following:
• Use the ANALYZE command to collect statistics
• Identify table, index, column, and cluster statistics
• Identify predicate selectivity calculations
• Use the DBMS_STATS package
• Create histograms for columns with skewed data

Copyright © Oracle Corporation, 2002. All rights reserved.

Objectives
• Use the ANALYZE command to provide the Oracle9i cost-based optimizer with statistics.
• Identify table, index, column, and cluster statistics. View these statistics in the Oracle8i
data dictionary views.
• Use the DBMS_STATS package to collect and manage optimizer statistics.
• Identify predicate selectivity calculations, assuming even data distribution. Identify the
consequences of using bind variables for predicate selectivity.
• Create histograms for columns with skewed data.
• Choose appropriate values for:
– A sample size (when using the ESTIMATE option of the ANALYZE command)
– The number of histogram buckets

Oracle9i: SQL Tuning Workshop 7-2

 The ANALYZE Command

INDEX
ANALYZE TABLE name
CLUSTER schema.

for
clause
COMPUTE STATISTICS
DELETE STATISTICS
ESTIMATE STATISTICS
for sample
clause clause

Copyright © Oracle Corporation, 2002. All rights reserved.

The ANALYZE Command

You collect statistics on an object with the ANALYZE command. Although the optimizer is not
sensitive to minor changes in volume or selectivity, you may want to collect new statistics
periodically on frequently modified tables to ensure that the optimizer is using recent, accurate
information.
Using the ANALYZE command to collect new statistics overwrites any existing statistics in the
data dictionary and flushes any related execution plans from the shared pool.
The DBMS_DDL package contains a procedure called ANALYZE_OBJECT that allows an index,
table, or cluster to be analyzed. The DBMS_UTILITY package contains a procedure called
ANALYZE_SCHEMA that gathers statistics on all objects in the given schema.
COMPUTE STATISTICS Option
You calculate exact statistics with this option. It performs a full table scan and several
calculations. For large tables, this operation could take a considerable amount of time.
ESTIMATE STATISTICS Option
You estimate statistics with this option. However, if you use this option with a suitable sample of
the data, it is almost as reliable as the COMPUTE STATISTICS option.
DELETE STATISTICS Option
You clear out statistics with this option. You do not need to use this option before reanalyzing an
object, because existing statistics are overwritten.
Oracle9i: SQL Tuning Workshop 7-3
 The ANALYZE Command

for
COMPUTE and ESTIMATE: clause

FOR TABLE
FOR ALL INDEXES
FOR ALL COLUMNS
INDEXED SIZE n
FOR COLUMNS column
SIZE n SIZE n

sample
ESTIMATE: clause
ROWS
SAMPLE n
PERCENT

Copyright © Oracle Corporation, 2002. All rights reserved.

The ANALYZE Command: Sampling

If more than half the data is specified as a sample size, the Oracle9i Server reads all the data and
computes the statistics rather than estimating them. If the table has fewer than 1,064 rows and a
sample size is not specified, the Oracle9i Server will compute the statistics regardless of whether
ESTIMATE STATISTICS or COMPUTE STATISTICS is chosen.
Note: The default ANALYZE behavior is the combination of FOR TABLE and FOR ALL
INDEXES.
FOR Clause
The FOR clause is meant for histograms and will be discussed in more detail later in this lesson.

Oracle9i: SQL Tuning Workshop 7-4

 Table Statistics

• Number of rows
• Number of blocks (always exact)
• Number of empty blocks (always exact)
• Average available free space
• Number of chained or migrated rows
• Average row length
• Last ANALYZE date and sample size
• Data dictionary views:
– USER_TABLES
– ALL_TABLES

Copyright © Oracle Corporation, 2002. All rights reserved.

Table Statistics
Following are the table statistics collected by the ANALYZE command:
• Number of rows
• Number of blocks (always exact)
• Number of empty (never used) blocks (always exact)
• Average row length (in bytes)
• Average available free space (in bytes per block)
• Number of chained or migrated rows
• Last ANALYZE date and sample size
Note: Chaining occurs when a single row is too big to fit in any database block, usually because
column widths are too big, or there are too many columns defined for the table.
Migration happens when the space reserved in the block for updates is insufficient, which causes
the row being updated to be moved into another block, leaving a pointer in the original block to
the new address for the row.

Oracle9i: SQL Tuning Workshop 7-5

ALL_TABLES and DBA_TABLES Data Dictionary Views
The table statistics are stored in the data dictionary. They can be displayed by querying the
USER_TABLES or ALL_TABLES view.

Name Null? Type

----------------- -------- ------------
TABLE_NAME NOT NULL VARCHAR2(30)
TABLESPACE_NAME VARCHAR2(30)
NUM_ROWS NUMBER
BLOCKS NUMBER
EMPTY_BLOCKS NUMBER
AVG_SPACE NUMBER
CHAIN_CNT NUMBER
AVG_ROW_LEN NUMBER
SAMPLE_SIZE NUMBER
LAST_ANALYZED DATE
...
Example
SQL> analyze table products
2 estimate statistics for table;

SQL> select num_rows,blocks,empty_blocks

2 , avg_space,avg_row_len
3 , sample_size
4 from user_tables
5 where table_name = ’PRODUCTS’;

NUM_ROWS BLOCKS EMPTY_BLOCKS AVG_SPACE AVG_ROW_LEN SAMPLE_SIZE

---------- ---------- ------------ ---------- ----------- -----------
10000 714 53 292 260 1007
Instructor Note
Demonstration: First delete all non-primary key indexes from the PRODUCTS table by running
the dai.sql script.
Second, run the demo07_01.sql script to demonstrate the code shown.
Note that the SAMPLE_SIZE column contains a number that could indicate a number of rows or
a percentage, depending on the last sample type used. Also note that the NUM_ROWS and
BLOCKS values are the maximum and minimum value for the clustering factor of any index
created on the table. The clustering factor is introduced on the next page.

Oracle9i: SQL Tuning Workshop 7-6

 Index Statistics

• Index level (always exact)

• Number of leaf blocks
• Number of distinct keys
• Average number of leaf blocks per key
• Average number of data blocks per key
• Number of index entries
• Clustering factor
• Last ANALYZE date and sample size

Copyright © Oracle Corporation, 2002. All rights reserved.

Index Statistics
Note that the index B*-tree-level statistic is always exact, regardless whether you compute or
estimate index statistics. The number of distinct keys may include rows that have been deleted.
Index Clustering Factor
The index clustering factor is an important index statistic for the CBO to estimate index scan
costs. It is an indication of the number of (logical) data block visits needed to retrieve all table
rows via the index. If the index entries follow the table row order, this value approaches the
number of data blocks (each block is visited only once). Conversely, if the index entries
randomly point at different data blocks, the clustering factor could approach the number of rows.
Suppose a typical table contains 50 rows per data block and 5,000 rows in total. When a query
looks for 50 rows (selectivity is 1%), the clustering factor indicates whether you must visit one
block (best case: 1% of the data blocks) or 50 blocks (worst case: 50% of the data blocks).
Instructor Note
The index clustering factor is a measure of how well ordered the data is in the table in reference
to the key.

Oracle9i: SQL Tuning Workshop 7-7

 Index Statistics

• Collect index statistics while creating or rebuilding

indexes
• Data dictionary views:
– USER_INDEXES
– ALL_INDEXES
SQL> select uniqueness, blevel
2 , leaf_blocks, distinct_keys
3 , clustering_factor
4 from user_indexes
5 where index_name = ’PRODUCTS_PK’;

Copyright © Oracle Corporation, 2002. All rights reserved.

Index Statistics (continued)

You can collect index statistics with the ANALYZE command, but you can also collect index
statistics while creating or rebuilding them by using the following syntax:
SQL> create index … compute statistics
SQL> alter index … rebuild compute statistics

Oracle9i: SQL Tuning Workshop 7-8

USER_INDEXES and ALL_INDEXES Data Dictionary Views
The index statistics are stored in the data dictionary. They can be displayed by querying the
USER_INDEXES or ALL_INDEXES view.
Name Null? Type
----------------------- -------- ------------
INDEX_NAME NOT NULL VARCHAR2(30)
INDEX_TYPE VARCHAR2(12)
UNIQUENESS VARCHAR2(9)
BLEVEL NUMBER
LEAF_BLOCKS NUMBER
DISTINCT_KEYS NUMBER
AVG_LEAF_BLOCKS_PER_KEY NUMBER
AVG_DATA_BLOCKS_PER_KEY NUMBER
CLUSTERING_FACTOR NUMBER
NUM_ROWS NUMBER
SAMPLE_SIZE NUMBER
LAST_ANALYZED DATE
...
Example
SQL> analyze index products_pk compute statistics;

SQL> select uniqueness, blevel

2 , leaf_blocks, distinct_keys
3 , clustering_factor
4 from user_indexes
5 where index_name = ’PRODUCTS_PK’;

UNIQUENES BLEVEL LEAF_BLOCKS DISTINCT_KEYS CLUSTERING_FACTOR

--------- ---------- ----------- ------------- -----------------
UNIQUE 1 43 10000 709
Instructor Note
Demo: Run the demo07_02.sql script to analyze an index and display information about the
analyzed index. You can analyze the PRODUCTS_PK as shown in the example, or choose
another index to analyze.

Oracle9i: SQL Tuning Workshop 7-9

 Column Statistics

• Number of distinct values

• Lowest value
• Highest value
• Last ANALYZE date and sample size
• Data dictionary views:
– USER_TAB_COL_STATISTICS
– ALL_TAB_COL_STATISTICS

Copyright © Oracle Corporation, 2002. All rights reserved.

Column Statistics
These are the column statistics collected by the ANALYZE command:
• Number of distinct values
• The lowest value
• The highest value
• Last ANALYZE and sample size
Instructor Note
This statement causes histograms to be created for every column. This is not advisable.
SQL> analyze table products
2 compute statistics for all columns;
Instead, use the syntax FOR TABLE FOR COLUMNS, such as:
SQL> analyze table products compute statistics
2 for table for columns prod_list_price
3 size 50;
(This example is shown on page 7-22.)
Both the lowest and the highest value are stored in RAW (binary) format.

Oracle9i: SQL Tuning Workshop 7-10

USER/ALL_TAB_COL_STATISTICS Data Dictionary Views
The column statistics are stored in the data dictionary. They can be displayed by querying the
USER_TAB_COL_STATISTICS or ALL_TAB_COL_STATISTICS view.
Name Null? Type
------------------------------- -------- ------------
TABLE_NAME NOT NULL VARCHAR2(30)
COLUMN_NAME NOT NULL VARCHAR2(30)
NUM_DISTINCT NUMBER
LOW_VALUE RAW(32)
HIGH_VALUE RAW(32)
DENSITY NUMBER
NUM_NULLS NUMBER
NUM_BUCKETS NUMBER
LAST_ANALYZED DATE
SAMPLE_SIZE NUMBER
GLOBAL_STATS VARCHAR2(3)
USER_STATS VARCHAR2(3)
AVG_COL_LEN NUMBER
Note: For backward compatibility with Oracle7, the USER/ALL_TAB_COLUMNS data
dictionary views also show column statistic values. Also, the NUM_BUCKETS column shows that
regular column statistics are treated as a histogram with one bucket. Histograms are discussed
later in this lesson.
Example
SQL> select
column_name, num_distinct
2 , low_value, high_value
3 , num_nulls, num_buckets
4 from user_tab_col_statistics
5 wheretable_name = ’PRODUCTS’;
NUM NUM NUM
COLUMN_NAME DISTINCT LOW_VALUE HIGH_VALUE NULLS BUCKETS
-------------------- ---------- ---------- ---------- ------ --------
PROD_ID 10000 C106 C306 0 75
PROD_NAME 759 312D506965 5A69702D46 0 75
PROD_DESC 671 7468697320 7468697320 0 75
PROD_SUBCATEGORY 37 4361737561 556E646572 0 36
PROD_SUBCAT_DESC 31 7468697320 7468697320 0 30
PROD_CATEGORY 4 426F7973 576F6D656E 0 3
...
14 rows selected.
Instructor Note
Demo script: demo07_03.sql
If you have time, you can show the BOIL_RAW function here. You create this function by
running the boilraw.sql script. This is an unsupported script. The function is used by the
colstats.sql script, but you can also use the function directly; for example, see the
demo07_04.sql script, and enter PROD_ID when prompted.
Note that Oracle9i also provides a DBMS_STATS.CONVERT_RAW_VALUE procedure that has
the same functionality. The DBMS_STATS package is discussed later this lesson.
Oracle9i: SQL Tuning Workshop 7-11
 The DBMS_STATS Package

Database Schema Table Index

Parallel or serial Serial

Copyright © Oracle Corporation, 2002. All rights reserved.

The DBMS_STATS Package

You can use the DBMS_STATS supplied package to generate and manage statistics for cost-
based optimization. The statistics can reside in the data dictionary or in a table created in the
user’s schema. Only statistics stored in the dictionary impact the cost-based optimizer.
The procedures in the package can be divided into three main functions:
• To set or get individual statistics
• To transfer statistics between the data dictionary and user tables
• To gather certain classes of optimizer statistics
Note: The procedures of the third category have improved (or provide equivalent) performance
characteristics compared to the ANALYZE command.
Reference
For full details of the DBMS_STATS package, see the Oracle9i Supplied PL/SQL Packages and
Types Reference .

Oracle9i: SQL Tuning Workshop 7-12

 DBMS_STATS: Generating Statistics

dbms_stats.GATHER_TABLE_STATS
(’SH’ -- schema
,’CUSTOMERS’ -- table
, NULL -- partition
, 20 -- sample size(%)
, FALSE -- block sample?
,’FOR ALL COLUMNS’ -- column spec
, 4 -- degree of //
,’DEFAULT’ -- granularity
, TRUE -- cascade to indexes
);

Copyright © Oracle Corporation, 2002. All rights reserved.

DBMS_STATS Procedures
Use the following procedures to gather statistics on indexes, tables, columns, and partitions:
Procedure Description
GATHER_ INDEX_STATS Collects index statistics
GATHER_TABLE_STATS Collects table, column, and index statistics
GATHER_SCHEMA_STATS Collects statistics for all objects in a schema
GATHER_DATABASE_STATS Collects statistics for all objects in a database

DBMS_STATS does not gather cluster statistics, but you can use it to gather statistics on the
individual tables in the cluster. Other options for gathering statistics with DBMS_STATS include
the following:
• Collect statistics either serially or in parallel. Index statistics are gathered only serially.
• Compute or estimate statistics; use block samples or row samples to estimate statistics.
• Specify the columns for which statistics are needed with GATHER_TABLE_STATS.
• Use the CASCADE option to collect index statistics when generating table statistics.
• Hold statistics in statistics tables to experiment with different sets of statistics.
Instructor Note
Run the demo07_05.sql script to show the above code. This demonstration script works with
scripts demo07_06.sql and demo07_07.sql, which are run on the next pages.
Oracle9i: SQL Tuning Workshop 7-13
 Copy Statistics Between Databases

Data dictionary
User-defined
statistics table

Copy from
DD to user
Export
table Copy from
and import
user table user table
to DD

User-defined
statistics table
Data dictionary

Copyright © Oracle Corporation, 2002. All rights reserved.

Copying Statistics Between Databases

You can use the DBMS_STATS package to copy statistics from a production to a test database to
facilitate tuning. For example, to copy a schema’s statistics, perform the following steps:
1. Use the DBMS_STATS.CREATE_STAT_TABLE procedure in the production database to
create a user-defined statistics table.
2. Use the DBMS_STATS.EXPORT_SCHEMA_STATS procedure in the production database
to copy statistics from the data dictionary to the user-defined statistics table from Step 1.
3. Use the export and import utilities to transfer the statistics to a corresponding user-defined
statistics table in the test database.
4. Use the DBMS_STATS.IMPORT_SCHEMA_STATS procedure to import the statistics into
the data dictionary in the test database.
You can also use DBMS_STATS to back up statistics prior to analyzing objects. Use a backup to:
• Restore old statistics
• Study changes in data characteristics over time

Oracle9i: SQL Tuning Workshop 7-14

 Example: Copying Statistics

dbms_stats.CREATE_STAT_TABLE
(’SH’ -- schema
,’STATS’ -- statistics table name
,’DATA01’ -- tablespace
);

dbms_stats.EXPORT_TABLE_STATS
(’SH’ -- schema
,’CUSTOMERS’ -- table name
, NULL -- no partitions
,’STATS’ -- statistics table name
, NULL -- id for statistics
, TRUE -- index statistics
);

Copyright © Oracle Corporation, 2002. All rights reserved.

Example
To copy statistics for a table from the data dictionary to a user-defined statistics table, use the
CREATE_STAT_TABLE procedure to create the user-defined table and
EXPORT_TABLE_STATS to copy the statistics. DBMS_STATS includes the following
procedures for copying statistics:
Procedure Description
CREATE_STAT_TABLE Creates a user-defined table capable of
holding statistics
DROP_STAT_TABLE Drops a user-defined statistics table
EXPORT_object_STATS Exports statistics from the data dictionary to a
user-defined table
IMPORT_object_STATS Imports statistics from a user-defined table to the
data dictionary

where object can be COLUMN, INDEX, TABLE, or SCHEMA.

Instructor Note
Demonstration script: demo07_06.sql
You may need to change the name of the tablespace to your environment’s tablespace name.
Oracle9i: SQL Tuning Workshop 7-15
 Example: Gathering Statistics

begin
dbms_stats.CREATE_STAT_TABLE
(’SH’, ’STATS’);
dbms_stats.GATHER_TABLE_STATS
(’SH’, ’CUSTOMERS’
,stattab => ’STATS’);
end;

begin
dbms_stats.DELETE_TABLE_STATS
(’SH’, ’CUSTOMERS’);
dbms_stats.IMPORT_TABLE_STATS
(’SH’, ’CUSTOMERS’
,stattab => ’STATS’);
end;

Copyright © Oracle Corporation, 2002. All rights reserved.

Gathering Optimizer Statistics

The GATHER_TABLE_STATS procedure gathers table and column (and index) statistics. There
are several parameters for the procedure; most are optional and have a default value. See the
Oracle9i Supplied PL/SQL Packages and Types Reference for more information.
Example
There has been a lot of modification against the CUSTOMERS table since the last time statistics
were gathered. To ensure that the cost-based optimizer is still selecting the best plan, statistics
are gathered again using GATHER_TABLE_STATS. However, you are concerned that new
statistics will cause the optimizer to choose bad plans when the current ones are acceptable.
If you believe that the new statistics cause the optimizer to generate poor plans, you can restore
the original statistics using DELETE_TABLE_STATS and SET_TABLE_STATS.
Instructor Note
Demonstration script: demo07_07.sql.
This is a second example of the GATHER_TABLE_STATS procedure; note that this time you
use a user-defined statistics table. With the IMPORT_TABLE_STATS, you restore statistics
from the user-defined statistics table to the data dictionary.

Oracle9i: SQL Tuning Workshop 7-16

 Predicate Selectivity

• Unique or primary key = constant

single-row predicate
• Nonunique index = constant
selectivity = 1/distinct_keys
• Bounded or unbounded range scan
selectivity = (high-low+1)/(max-min+1)

– high: upper bound (or max)

– low: lower bound (or min)
– max,min: column statistics

Copyright © Oracle Corporation, 2002. All rights reserved.

Predicate Selectivity
The expected percentage of rows (the selectivity) depends on the sort of operation performed in
the WHERE clause. The cost-based approach is more likely to choose an index access path for a
query with good selectivity. The Oracle9i optimizer uses the following rules when calculating
selectivity.
Equality Conditions
Unique or primary key = constant
This is a single-row predicate, returning one row maximally; it has maximum selectivity.
Nonunique index = constant
The query may return more than one row, so the optimizer uses the number of distinct key values
in the column. The selectivity is 1 divided by the number of distinct values.
Bounded and Unbounded Ranges
Bounded and unbounded ranges need statistics to calculate selectivity, thus CBO and RBO yield
different results. The RBO must rely on the syntax and its ranking scheme; the CBO uses the
following formula:
high - low + 1 low high
selectivity = ----------------
max - min + 1
min max
Oracle9i: SQL Tuning Workshop 7-17
 Bind Variables and Predicate Selectivity

• Equality condition: No difference

• Bounded or unbounded range scan:
Built-in default selectivity assumed
• If high performance is an issue, consider using
dynamic SQL instead of bind variables:
– DBMS_SQL package
– EXECUTE IMMEDIATE
• Dynamic SQL implies less shared SQL

Copyright © Oracle Corporation, 2002. All rights reserved.

Bind Variables and Predicate Selectivity

In case of equality conditions (that is, if the operator is an equality sign), a bind variable is
treated as a constant. See the first two bullets in the previous slide.
Because the bind phase follows the parse phase, the Oracle9i optimizer does not know the actual
bind variable values when optimizing a SQL statement. This means that the optimizer must use a
built-in default selectivity in case a SQL statement contains bind variables. These built-in values
are not documented, cannot be influenced, and can change with each new release of the Oracle
server.

Oracle9i: SQL Tuning Workshop 7-18

Performance
If performance is an issue, consider using dynamic SQL to build statements with literal values
instead of bind variables. This means that each statement will be optimized separately (giving
optimal performance). However, there is one big disadvantage: because most dynamic SQL
statements are different from one another, you will have less benefit from shared SQL.
You can send dynamic SQL (that is, SQL statements that are built at run time) to the Oracle8i
Server by using the DBMS_SQL package, or by using the EXECUTE IMMEDIATE statement.
Dynamic SQL is not discussed in this class; see the PL/SQL User’s Guide and Reference for
more details.
Instructor Note
Be careful to talk about built-in default selectivities. As previously stated, they are
undocumented and can change without notice. Note that these default values are also used when
CBO is forced and no statistics are available. Since Oracle 7.3, the default selectivity for an
inequality is 5%.
Instructor Note (for page 7-20)
This slide is just an example; histograms are only useful on large sets of values. That is why it is
useless to give selectivity calculations based on this example. You can use the histdemo.sql
demonstration script at the end of this lesson, based on this example.

Oracle9i: SQL Tuning Workshop 7-19

 Histograms

10
8 Width-balanced Height-balanced
5
3
2 1 1 3 100
2 1 1 2 10
1 1 1 2 8
1 1 1 1 5
1
1 1 1 1 3 100
1
1 Height-balanced histograms give a
1 better sense of the data distribution.
1
1 100

1 25 50 75 100

Copyright © Oracle Corporation, 2002. All rights reserved.

Histograms
The optimizer must estimate the number of rows processed by a given query. This is achieved
partly by estimating the selectivities of the query’s predicates. The accuracy of these estimates
depends on the optimizer’s knowledge of the data distribution. Without histograms, an even
distribution is assumed.
Types of Histograms
Width-balanced histograms partition the attribute domain into equal width ranges, called
buckets, and count the number of rows, the value of which in a column makes them fall into each
bucket. If several rows contain the same value, they are all put into the same bucket, increasing
the height of that bucket.
In height-balanced histograms, each bucket has approximately the same number of rows. If
several rows contain the same value, they may be put in the same bucket or spread across several
buckets, as the histogram balances the heights of the buckets. Some buckets cover only one or a
few values, because there are many rows with those values. Some buckets cover many values,
because there are few rows with those values.
Height-balanced histograms are more suitable for computing selectivity estimates and are used
by the Oracle9i Server.

Oracle9i: SQL Tuning Workshop 7-20

 Histograms and Selectivity

• Popular and nonpopular values

• Range scan selectivity is determined by the
number of buckets spanned
• Equality search selectivity is determined by:
– The number of buckets (popular values)
– DENSITY (nonpopular values)

Copyright © Oracle Corporation, 2002. All rights reserved.

Popular and Nonpopular Values

When you create histograms, the Oracle9i Server distinguishes between popular and nonpopular
values. Popular values appear more than once as a bucket endpoint value; values that appear only
once or not at all are considered nonpopular.
DENSITY Statistic
When you create a histogram, the DENSITY statistic is calculated. This is calculated by 1 over
NUM_DISTINCT. It is used by the CBO to reduce the bias caused by the repetition of popular
values. When a column has no histograms, this statistic contains a NULL value.
CBO Selectivity Estimates
When the CBO estimates selectivity for a range scan, the number of buckets spanned is divided
by the total number of buckets. The CBO does not consider partial buckets, assuming that the
number of buckets offers enough granularity. For equality search conditions, the Oracle9i
optimizer has two approaches:
• For a popular value, the number of buckets having that endpoint value is divided by the
total number of buckets.
• For a nonpopular value, the DENSITY statistic is used as a selectivity estimate.

Oracle9i: SQL Tuning Workshop 7-21

 Histogram Statistics: Example

• Generate statistics for the PRODUCTS table and the

PROD_LIST_PRICE column; maximum 50 buckets.
SQL> analyze table products compute statistics
2 for table for columns prod_list_price
3 size 50;

• Recompute the statistics on the same column

without specifying the number of buckets.
SQL> analyze table products compute statistics
2 for columns prod_list_price;

Copyright © Oracle Corporation, 2002. All rights reserved.

Example of Histogram Statistics Collection

The column statistics contain the bucket endpoints for the histogram, the number of distinct
values and null values, and the density of those values.
Collection of Statistics
Use the ANALYZE command to collect information for histograms by using the FOR clause (see
also the beginning of this lesson). The SIZE option specifies the number of buckets.
ANALYZE TABLE table {COMPUTE|ESTIMATE} STATISTICS
FOR {TABLE | ALL INDEXES
|ALL [INDEXED] COLUMNS [SIZE n]
|FOR COLUMNS [SIZE n] column [SIZE n]
[,column [SIZE n]]…}
Note: The maximum number of histogram buckets is 255. Note, however, that Oracle will never
create more buckets than the number of distinct column values because that is useless. If you
have one bucket for each distinct value, you can store the exact cardinality of each column value.
Instructor Note
Demonstration script: demo07_08.sql
This demonstration is not very useful yet. However, you can present the histogram data at the
end of this lesson with a query on the data dictionary.
Oracle9i: SQL Tuning Workshop 7-22
 Histogram Statistics: Example

• Generate statistics for the PRODUCTS table and the

PROD_LIST_PRICE column; maximum 50 buckets
using DBMS_STATS.GATHER_TABLE_STATS.
SQL> EXECUTE DBMS_STATS.GATHER_TABLE_STATS
2 (’sh’,’PRODUCTS’, METHOD_OPT =>
3 ’FOR COLUMNS SIZE 50 prod_list_price’);

• The SIZE keyword declares the maximum number

of buckets for the histogram.
• Use the DBMS_STATS package to have the database
automatically decide which columns need
histograms by specifying the SIZE as “auto.”

Copyright © Oracle Corporation, 2002. All rights reserved.

Collection of Statistics (continued)

You can generate histograms by using the DBMS_STATS package. You can generate histograms
for columns of a table or partition. The SIZE keyword declares the maximum number of buckets
for the histogram.
It is recommended that you use the DBMS_STATS package to have the database automatically
decide which columns need histograms. This is done by specifying size as “auto.”

Oracle9i: SQL Tuning Workshop 7-23

 Histogram Tips

• Use the FOR ALL INDEXED COLUMNS option.

• Update histograms frequently if data distribution is
not static.
WHERE clauses with bind variables do not use
histograms.
• Do not use histograms unless they substantially
improve performance.
• Histograms allocate additional storage.

Copyright © Oracle Corporation, 2002. All rights reserved.

Histogram Tips
For many tables, it is appropriate to use the FOR ALL INDEXED COLUMNS clause to collect
statistics on all indexed columns. Be aware, however, that histograms are also created on unique
and primary key constrained columns that are enforced through indexes, which might be a waste
of space.
If the data distribution is not static, the histogram must be frequently updated.
Because the optimizer does not take into consideration the current value of bind variables,
histograms are not helpful (and therefore not worth the expense) when bind variables are used.
Histograms allocate additional storage and should be used only when they can substantially
improve the query plans.

Oracle9i: SQL Tuning Workshop 7-24

 When to Use Histograms

• Use histograms for a column when the data is

highly skewed.
• Avoid using histograms when:
– The column is not used in WHERE clauses
– The column is unique and used only with equality
predicates
– All predicates on the column use bind variables
– The column data is uniformly distributed

Copyright © Oracle Corporation, 2002. All rights reserved.

When to Use Histograms

Histograms are expensive to store; therefore, use them only when they substantially improve
performance. Histograms should be used when the data has a high degree of skew, with some
values occurring much more frequently than others.
Histograms should not be used when:
• The column is not used in the WHERE clause of queries
• The column is unique and is used only with equality queries
• All the predicates on the column use bind variables
• The column data is uniformly distributed
Instructor Note
Histograms, when combined with bind variables, can have undesirable effects

Oracle9i: SQL Tuning Workshop 7-25

 Choosing a Sample Size

• If the data is uniformly distributed, 5% of the rows

will suffice.
• Choose a higher sample size if the number of
distinct values is more than 10% of the number of
rows.
• When using histograms, the number of sampled
rows must be at least 100 times the number of
buckets.

Copyright © Oracle Corporation, 2002. All rights reserved.

Choosing a Sample Size

If the data to be analyzed is almost uniformly distributed, a sample of about 5% of the rows
should be adequate. Do not use histograms in this case, because the benefit will be minimal.
If the number of distinct values is large—such as more than 10% of the number of rows—then a
higher sampling percentage may be necessary.
If you decide to create histograms, the number of rows sampled should be at least 100 times the
number of buckets in the histogram. In other words, the average number of sampled rows per
bucket should be at least 100. You should not combine a very detailed histogram with a minimal
sample size.

Oracle9i: SQL Tuning Workshop 7-26

 Choosing the Number of Buckets

• Start with the default size of 75 buckets.

• Experiment with different sizes to obtain optimum
results.
• Set the number of buckets to be greater than the
number of frequently occurring distinct values, if
there are relatively few.

Copyright © Oracle Corporation, 2002. All rights reserved.

Choosing the Number of Buckets for a Histogram

The default number of buckets is 75. This value provides an appropriate level of detail for most
data distributions. However, because the number of buckets in the histogram, the sample size,
and the data distribution all affect the usefulness of a histogram, you may need to experiment
with different numbers of buckets to obtain the best results.
If the number of frequently occurring distinct values in a column is relatively small, then you
should set the number of buckets to be greater than the number of frequently occurring distinct
values.
Note that the Oracle server will never create more buckets than the number of distinct column
values, and the maximum is 255 buckets. It is recommended that DBMS_STATS package be
used to automatically size the histograms by specifying “auto.”
You can use SQL*Plus AUTOTRACE or EXPLAIN PLAN to see the influence of the number of
buckets on the cost, as estimated by the cost-based optimizer. At a certain moment, raising the
number of buckets will no longer significantly change the cost calculations.
Instructor Note
Be aware of the following: Histograms on NUMBER and DATE columns are always accurate.
However, histograms on alphanumeric columns like VARCHAR2 truncate the values, which
causes collisions if the leading part of the strings are equal. In earlier releases only the first six
characters are considered. This was relieved in release 8.1.6 and later where the first 31
characters are used.
Oracle9i: SQL Tuning Workshop 7-27
 Viewing Histogram Statistics

• Histogram information:
USER/ALL_HISTOGRAMS
SQL> select endpoint_number, endpoint_value
2 from dba_tab_histograms
3 where table_name = ’PRODUCTS’
4 and column_name = ’PROD_LIST_PRICE’;

• The number of buckets in each histogram:

USER/ALL_TAB_COL_STATISTICS

Copyright © Oracle Corporation, 2002. All rights reserved.

Viewing Histogram Statistics

The example query in the slide probably shows the following result:
ENDPOINT_NUMBER ENDPOINT_VALUE
--------------- --------------
1 7.5
2 8.99
3 9.95
...
49 198
50 1200
0 3
You have 67 buckets, including a special zero bucket to store the lowest PROD_LIST_PRICE
values. The last bucket (excluding the 0 bucket) has 1200 as the endpoint value, which is the
maximum PROD_LIST_PRICE value.
Note: If a histogram has popular values, they show up as “gaps” in the endpoint number column.
Multiple buckets with the same endpoint value are stored in the data dictionary as a single row,
to preserve space.

Oracle9i: SQL Tuning Workshop 7-28

 Summary

In this lesson, you should have learned how to:

• Collect table, column, index, and cluster statistics
with the ANALYZE command
• Use the DBMS_STATS package
• Copy statistics between databases
• Determine selectivity for predicates with and
without bind variables
• Utilize histograms
– Popular
– Nonpopular values

Copyright © Oracle Corporation, 2002. All rights reserved.

Summary
This lesson introduced you to table statistics used with the cost-based optimizer. By using the
ANALYZE command or DBMS_STATS package, you can collect statistics on your tables which
are used by the cost-based optimizer to determine optimal execution routes for SQL statements
on analyzed tables.
Selectivity, the expected percentage of rows returned, depends on the type of operation
performed in the WHERE clause. The cost-based approach is more likely to choose an index
access path for a query with good selectivity. Selectivity is influenced by data distribution. When
bind variables are used in a SQL statement, the optimizer must use a built-in default selectivity.
These built-in values cannot be influenced.
Oracle can use a histogram to decide whether or not to use the index. A histogram stores
information about the frequency of various column values. Without histograms, an even
distribution is assumed. Create histograms for columns with skewed data.
Instructor Note (for page 7-28)
Demonstration script: demo07_09.sql. This shows the result of the previous demonstration
when you created histograms on the PRODUCTS table. You can also use the histdemo.sql
script, based on the example histogram in this lesson. That script does not show endpoint
numbers 0 and 1 (see the note above).

Oracle9i: SQL Tuning Workshop 7-29

 Practice Overview

This practice covers the following topics:

• Generating table statistics
• Creating a histogram
• Calculating selectivity
• Querying data dictionary information
• Deleting table statistics

Copyright © Oracle Corporation, 2002. All rights reserved.

Practice Overview
In this practice, you generate table statistics for the PRODUCTS table. Because the
PROD_STATUS column of the PRODUCTS table is a good candidate for a histogram, you create
a histogram for that column. You calculate the selectivity of a certain predicate on the column,
both before and after creating the histogram. Then you query the data dictionary for the last
analyze date and sample size. Lastly, you delete the statistics on the PRODUCTS table from the
data dictionary.

Oracle9i: SQL Tuning Workshop 7-30

Practice 7
1. Turn SQL*Plus AUTOTRACE off.
Compute statistics for the PRODUCTS table and view the table and column statistics with
appropriate select statements.
2. Why is the PROD_STATUS column of the PRODUCTS table a good candidate for creating
a histogram?
Hint: Retrieve the (distinct) PROD_STATUS values and their cardinality.
3. Create a histogram for the PROD_STATUS column with 20 buckets and view the histogram
statistics.
4. Calculate the selectivity of the following select statement, both before and after creating the
histogram:
SQL> select * from products
2 where prod_status = ’obsolete’;
5. Identify the last analyze date and sample size for all tables in your schema.
6. Delete the statistics for the PRODUCTS table and check the data dictionary again.

Oracle9i: SQL Tuning Workshop 7-31

Oracle9i: SQL Tuning Workshop 7-32
 Influencing the Optimizer

Copyright © Oracle Corporation, 2002. All rights reserved.

Schedule: Timing Topic

45 minutes Lecture
10 minutes Practice
55 minutes Total
Instructor Note
This lesson starts with a recapitulation of influencing the optimizer mode at the instance and the
session level. Then hints are introduced: basic access path hints, advanced access path hints, and
additional hints. Finally, hints and views are covered, introducing the concept of mergeable and
nonmergeable views.
 Objectives

After completing this lesson, you should be able to

do the following:
• Influence the optimizer behavior at the following
levels:
– Instance and session level
– Statement level
• Specify access-path hints
• Specify hints on and in views

Copyright © Oracle Corporation, 2002. All rights reserved.

Objectives
After completing this lesson, you should be able to:
• Influence the Oracle9i optimizer behavior at the following levels:
– Instance level, setting initialization parameters
– Session level, using the ALTER SESSION command
– Statement level, using hints
• Specify access path hints
• Specify hints on and in views
Note
• Influencing the optimizer at the instance and session level is also covered in the “Rule-
Based Optimization Versus Cost-Based Optimization” lesson.
• When statistics are not available and cost-based optimization is forced, the optimizer uses
default built-in values instead. This may produce suboptimal execution plans.

Oracle9i: SQL Tuning Workshop 8-2

 Setting the Optimizer Mode

• At the instance level, set the following parameter:

OPTIMIZER_MODE =
{CHOOSE|RULE|FIRST_ROW_n|ALL_ROWS}

• For a session, use the following SQL command:

SQL> alter session set optimizer_mode =
{choose|rule|first_rows_n|all_rows}

Copyright © Oracle Corporation, 2002. All rights reserved.

Setting the Optimizer Mode

Oracle9i has five basic optimizer modes:
CHOOSE Cost-based when statistics are available (ALL_ROWS), and
rule-based when no statistics are available
RULE Forces rule-based optimization regardless the presence of statistics
FIRST_ROWS_N The optimizer uses a cost-based approach, regardless of the presence
of statistics, and optimizes with a goal of best response time to return
first n number of rows. N can equal 1, 10, 100, or 1000.
FIRST_ROWS Optimizes response time until first result comes back (such as for
interactive OLTP applications)
ALL_ROWS Optimizes overall throughput (such as for reports and batch jobs)

Optimizer Approach Levels

At the instance level, you can set the optimizer approach by using the OPTIMIZER_MODE
initialization parameter. When this parameter is not set, it defaults to CHOOSE.
At the session level, the optimizer approach set at instance level can be overruled using the
ALTER SESSION command.
This lesson explains how to overrule the optimizer approach at the statement level by using hints.
Oracle9i: SQL Tuning Workshop 8-3
 Some Additional Parameters

• OPTIMIZER_FEATURES_ENABLE
– Set to current release to enable the latest optimizer
features
– Defaults to current release of database
• OPTIMIZER_INDEX_COST_ADJ

Copyright © Oracle Corporation, 2002. All rights reserved.

Some Additional Parameters

You can use OPTIMIZER_FEATURES_ENABLE to change a set of initialization parameters
which control the optimizer’s behavior. Values are, for example, 8.0.5, 8.0.5, 8.0.6, 8.1., 8.1.6,
8.1.7, and 9.0.1. You can use this parameter to prevent unexpected optimizer behavior when
migrating to a higher version.
Setting the parameter to 9.0.1 enables these new tuning features:
• Peeking of user-defined bind variables
• Complex view merging
• Push-join predicate
• Consideration of bitmap access paths for tables with only B*-tree indexes
• Subquery unnesting
• Index joins
Note: If you upgrade to a new release and you want to enable the features with that release, then
you do not need to explicitly set the OPTIMIZER_FEATURES_ENABLE parameter.

Oracle9i: SQL Tuning Workshop 8-4

Some Additional Parameters (continued)
You can use OPTIMIZER_INDEX_COST_ADJ to tune the optimizer behavior for access path
selection to be index-friendly. This parameter enables you to adjust the costing of index access
paths in the cost-based optimizer and thereby make the optimizer prone to selecting an index
access path over a full table scan.
The default for this parameter is 100 percent, which makes the optimizer estimate index access
paths at the regular cost. Any other value will make the optimizer estimate the access path at that
percentage of the regular cost. For example, setting it to 50 percent will make the index access
path seem half as expensive as normal.
The legal range of values for this parameter is 1 to 10,000 percent. You can use this parameter to
tune the performance of a system where it is felt that the optimizer chooses too few or too many
index access paths.
Note: There are more parameters to influence the optimizer that are specific to joins; those
parameters will be covered in the “Sorting and Joining” lesson.

Oracle9i: SQL Tuning Workshop 8-5

 Optimizer Hint Syntax

SELECT
INSERT /*+ hint */
UPDATE comment
text
DELETE

SELECT
INSERT --+ hint
UPDATE comment
text
DELETE

Copyright © Oracle Corporation, 2002. All rights reserved.

Optimizer Hint Syntax

Enclose hints within the comments of a SQL statement. You can use either style of comment.
The hint delimiter (+) must come immediately after the comment delimiter. If you separate them
by a space, the optimizer will not recognize that the comment contains hints.
For example, the following hints force the tables in the FROM clause to be joined left to right:
select /*+ORDERED */ ...
select /*+ ORDERED */ ...
The following attempt is treated as a normal comment because of the intervening space:
select /* +ORDERED */ ...
The Oracle9i Server ignores incorrectly specified hints. However, be aware of the following:
• You never get an error message.
• Other (correctly) specified hints within the same comment are considered.
• Combinations of conflicting hints are also ignored by the Oracle9i Server.

Oracle9i: SQL Tuning Workshop 8-6

 Rules for Hints

• Place hints immediately after the first SQL keyword

of a statement block.
• Each statement block can have only one hint
comment, but it can contain multiple hints.
• Hints apply only to the statement block in which
they appear.
• If a statement uses aliases, hints must reference
aliases rather than table names.

Copyright © Oracle Corporation, 2002. All rights reserved.

Rules for Hints

• You must place the hint comment immediately after the first keyword (SELECT, INSERT,
DELETE, or UPDATE) of a SQL statement block.
• A statement block can have only one comment containing hints, but it can contain many
hints inside that comment.
• Hints apply only to the statement block in which they appear and override instance- or
session-level parameters.
• A statement block is:
– A simple SELECT, INSERT, UPDATE, or DELETE statement
– A parent statement or a subquery of a complex statement
– A part of a compound query using set operators (UNION, MINUS, INTERSECT)
• If a SQL statement uses aliases, then hints must reference the alias rather than the table
name.
Instructor Note
Optimizer goals set at the session level with the ALTER SESSION command apply only to
statements submitted directly. They seem not to affect SQL that is run from within PL/SQL. You
should use hints to influence the cost-based optimizer for any SQL statements submitted from
within PL/SQL. You do not, however, need to raise this issue in class.

Oracle9i: SQL Tuning Workshop 8-7

 Hint Recommendations

• Use hints carefully because they imply a high

maintenance load.
• Be aware of the performance impact of hard-coded
hints when they become less valid.

Copyright © Oracle Corporation, 2002. All rights reserved.

Hint Recommendations
Use hints as a last remedy when tuning SQL statements. Be aware that the optimizer cannot
change execution plans, even when better alternatives become available.
Hints may become less valid (or even invalid) when the database structure or contents change.
Note: Beware that all hints, except the RULE hint, force cost-based optimization regardless of
your optimizer mode setting at the session or instance level. Hence statistics may be required to
ensure an optimal execution plan.

Oracle9i: SQL Tuning Workshop 8-8

 Optimizer Hint Example

SQL> update --+ INDEX(p PROD_CATAGORY_IDX)

2 products p
3 set p.prod_min_price =
4 (select
5 (pr.prod_list_price*.95)
6 from products pr
7 where p.prod_id = pr.prod_id)
8 where p.prod_category = ’Men’
9 and p.prod_status = ’available, on stock’
10 /

Copyright © Oracle Corporation, 2002. All rights reserved.

Optimizer Hint Syntax Example

The slide shows an example with a hint that advises the cost-based optimizer to use the index.
The execution plan will look like this:
Execution Plan
----------------------------------------------------------
0 UPDATE STATEMENT Optimizer=CHOOSE
(Cost=2 Card=3 Bytes=195)
1 0 UPDATE OF ’PRODUCTS’
2 1 TABLE ACCESS (BY INDEX ROWID) OF ’PRODUCTS’
(Cost=2 Card =3 Bytes=195)
3 2 INDEX (RANGE SCAN) OF ’PROD_CATEGORY_IDX’
(NON-UNIQUE) Cost=1 Card=3)
4 1 TABLE ACCESS (BY INDEX ROWID) OF ’PRODUCTS’
(Cost=2 Card=1 Bytes=26)
5 4 INDEX (UNIQUE SCAN) OF ’PRODUCTS_PK’ (UNIQUE)
(Cost=1 Card=291)
Instructor Note
Demonstration script: demo08_01.sql

Oracle9i: SQL Tuning Workshop 8-9

 Hint Categories

• Hints for optimization mode

– RULE, CHOOSE
– ALL_ROWS, FIRST_ROWS(n)
• Hints for access path methods
• Hints for parallel execution
• Hints for join orders and operations

Copyright © Oracle Corporation, 2002. All rights reserved.

Hint Categories
Parameters to change the optimizer mode at the instance or session level (RULE, CHOOSE,
ALL_ROWS, FIRST_ROWS(n)) were discussed in the “Rule-Based Optimization Versus Cost-
Based Optimization” lesson. You can specify these four values as a hint at the statement level as
well.
Hints for Access Path Methods
This lesson concentrates on this hint category.
Hints for Parallel Execution
This topic is covered in a separate course.
Hints for Join Orders and Operations
Optimizing joins is covered in the “Sorting and Join Techniques” lesson.

Oracle9i: SQL Tuning Workshop 8-10

 Basic Access Path Hints

FULL Performs a full table scan

ROWID Accesses a table by ROWID

INDEX Scans an index in ascending order

INDEX_ASC Scans an index in ascending order

INDEX_DESC Scans an index in descending order

AND_EQUAL Merges single-column indexes

INDEX_FFS Performs a fast full index scan

NO_INDEX Disallows using a set of indexes

Copyright © Oracle Corporation, 2002. All rights reserved.

Basic Access Path Hints

Hint Description
FULL(t) Chooses a full table scan
ROWID(t) Chooses a table scan by ROWID
INDEX(t [idx]…) Chooses an (ascending) index scan
INDEX_ASC(t [idx]…) Chooses an ascending index scan
INDEX_DESC(t[idx]…) Chooses a descending index scan (This hint has no effect
when accessing multiple tables; such statements always
perform ascending range scans.)
AND_EQUAL Chooses an execution plan that uses an access path that
merges the t idx idx [idx]…)scans on several single-column
indexes (minimum 2, maximum 5)
INDEX_FFS(t[idx]…) Causes a fast full index scan to be performed
NO_INDEX(t[idx]…) Explicitly disallows a set of indexes for the specified table

Note: The NO_INDEX hint is useful if you use distributed query optimization. It applies to
function-based, B*-tree, bitmap, cluster, and domain indexes. If this hint does not specify an
index name, the optimizer will not consider a scan on any index on the table.
Oracle9i: SQL Tuning Workshop 8-11
Specifying Index Names
An index hint can optionally specify one or more index names:
• If this hint specifies a single available index, the optimizer performs a scan on this index. The
optimizer does not consider a full table scan or a scan on another index on the table.
• If this hint specifies a list of available indexes, the optimizer considers the cost of a scan on
each index in the list and then performs the index scan with the lowest cost. The optimizer
may also choose to scan multiple indexes from this list and merge the results, if such an access
path has the lowest cost. The optimizer does not consider a full table scan or a scan on an
index that is not listed in the hint.
• If this hint specifies no indexes, the optimizer considers the cost of a scan on each available
index on the table and then performs the index scan with the lowest cost. The optimizer might
also choose to scan multiple indexes and merge the results, if such an access path has the
lowest cost. The optimizer does not consider a full table scan.
Fast Full Index Scans
Fast full index scans are an alternative to full table scans when the index contains all the columns
that are needed for the query. It cannot be used to eliminate a sort operation. It reads the entire
index by using multiblock reads (as in a full table scan).
Note: During a full table scan Oracle can read multiple blocks into memory in a single I/O
operation. This is commonly referred to as multiblock reads. It is controlled by the
DB_FILE_MULTIBLOCK_READ_COUNT initialization parameter. This parameters values is OS
dependent.
Instructor Note (for the next page)
Demonstration script demo08_02.sql.
Purpose: This script first drops all indexes on the CUSTOMERS table, then creates an index on the
COUNTRY_ID column. Several SELECT statements are run, some including hints. After each
SELECT statement, the screen is paused for you to explain the results before moving to the next
statement.

Oracle9i: SQL Tuning Workshop 8-12

 Advanced Access Path Hints

CLUSTER Performs a cluster scan

HASH Performs a hash scan

HASH_AJ Transforms a NOT IN subquery

into a hash anti-join

MERGE_AJ Transforms a NOT IN subquery

into a merge anti-join

MERGE_SJ Transforms a correlated EXISTS

subquery into a merge semijoin

USE_CONCAT Rewrites OR into UNION ALL and turns off

INLIST processing

NO_EXPAND Prevents OR-expansions

Copyright © Oracle Corporation, 2002. All rights reserved.

Advanced Access Path Hints

Hint Description

CLUSTER(t) Chooses a cluster scan (applies to index clusters only)

HASH(t) Chooses a hash scan (applies to hash clusters only)
HASH_AJ(t) Transforms a NOT IN subquery into a hash anti-join
MERGE_AJ(t) Transforms a NOT IN subquery into a merge anti-join
MERGE_SJ(t) Transforms a correlated EXISTS subquery into a merge semijoin
USE_CONCAT Rewrites OR into a UNION ALL; turns off inlist processing and
OR-expands all disjunctions
NO_EXPAND Prevents OR-expansions (This hint is the opposite of
USE_CONCAT.)

Anti-Joins and Semijoins

These two techniques to change subquery statements into equivalent joins are explained in the
“Sorting and Joining” lesson.

Oracle9i: SQL Tuning Workshop 8-13

 Buffer Cache Hints

CACHE Places blocks at the MRU end

of the LRU list (full table scan)

NOCACHE Places blocks at the LRU end of the

LRU list (full table scan) (default)

Copyright © Oracle Corporation, 2002. All rights reserved.

Buffer Cache Hints

The CACHE hint specifies that the blocks retrieved for this table are placed at the most recently
used (MRU) end of the LRU list in the buffer cache when a full table scan is performed. This
option is useful for small lookup tables.
The NOCACHE hint specifies that the blocks retrieved for this table are placed at the least recently
used end of the LRU list in the buffer cache when a full table scan is performed. This is the
normal behavior of the buffer cache, ensuring that a single full table scan of a large table cannot
push all recently used blocks from the LRU list.
Note: You can also specify the CACHE or NOCACHE attribute when creating or altering tables.
Instructor Note
Until release 8.0, you could use the CACHE_SIZE_THRESHOLD parameter to determine what is
considered a small table in this context. This parameter was eliminated in Oracle8i, featuring
multiple buffer pools. Check the “table scans (long tables)” and “table scans (short tables)”
statistics in V$SYSSTAT to see the effects of the CACHE and NOCACHE hints.
The CACHE and NOCACHE hints are being deprecated in favor of using multiple buffer pools.

Oracle9i: SQL Tuning Workshop 8-14

 Hints and Views

• Do not use hints in views.

• Use view optimization techniques:
– Statement transformation
– Access view results as a table
• Hints can be used on mergeable views and
nonmergeable views.

Copyright © Oracle Corporation, 2002. All rights reserved.

Hints and Views

You should not use hints inside or on views. This is because views can be defined in one context
and used in another; such hints can result in unexpected plans. In particular, hints inside views or
on views are handled differently depending on whether or not the view is mergeable into the top-
level query.
View Optimization
To optimize a statement that accesses a view, the optimizer has two alternatives. Normally the
statement is transformed into an equivalent statement that accesses the view’s base tables. The
optimizer can use one of these techniques to transform the statement:
• Merge the view’s query into the referencing query block in the accessing statement
• Push the predicate of the referencing query block inside the view
When these transformations are impossible, the view’s query is executed and the result is
accessed as though it were a table. This appears as a VIEW step in execution plans.

Oracle9i: SQL Tuning Workshop 8-15

Hints and Views (continued)
Mergeable Views
The optimizer can merge a view into a referencing query block, provided the view definition does
not contain the following:
• Set operators (UNION, UNION ALL, INTERSECT, MINUS)
• A CONNECT BY clause
• The ROWNUM pseudocolumn
• Group functions (AVG, COUNT, MAX, MIN, SUM) in the select list
Hints and Mergeable Views
Optimization approach and goal hints can occur in a top-level query or inside views:
• If there is such a hint in the top-level query, that hint is used regardless of any other such hints
inside the views.
• If there is no top-level optimizer-mode hint, then mode hints in referenced views are used as
long as all mode hints in the views are consistent.
• If two or more mode hints in the referenced views conflict, then all mode hints in the views are
discarded and the session mode is used, whether default or user-specified.
Access method and join hints on referenced views are ignored unless the view contains a single
table (or references another view with a single table). For such single-table views, an access-
method hint or a join hint on the view applies to the table inside the view.
Access method and join hints can also appear in a view definition:
• If the view is a subquery (that is, if it appears in the FROM clause of a SELECT statement),
then all access-method and join hints inside the view are preserved when the view is merged
with the top-level query.
• For views that are not subqueries, access-method and join hints in the view are preserved only
if the top-level query references no other tables or views (that is, if the FROM clause of the
SELECT statement contains only the view).
Hints and Nonmergeable Views
With nonmergeable views, optimization mode hints inside the view are ignored. The top-level
query decides the optimization mode.
Because nonmergeable views are optimized separately from the top-level query, access-method
and join hints inside the view are always preserved. For the same reason, access-method hints on
the view in the top-level query are ignored.
However, join hints on the view in the top-level query are preserved because in this case, a
nonmergeable view is similar to a table.

Oracle9i: SQL Tuning Workshop 8-16

 View Processing Hints

MERGE Merges complex views or subqueries with

the surrounding query

NO_MERGE Does not merge mergeable views

Copyright © Oracle Corporation, 2002. All rights reserved.

View Processing Hints

You can cause a view to be merged on a per-query basis by using the MERGE and NO_MERGE
hints. The NO_MERGE hint causes the Oracle9i Server not to merge mergeable views.
Example
SQL> select /*+ MERGE(v) */
2 p.prod_id, p.prod_name, v.sold
3 from products p,
4 (select s.prod_id, sum(amount_sold) as sold
5 from sales s
6 group by s.prod_id) v
7 where p.prod_id = v.prod_id
8 and p.prod_status = ’obsolete’;
Note: Because the group function is used on line 6, the MERGE(v) hint is ignored.
Instructor Note
Demonstration script: demo08_03.sql to show that the first code example cannot be merged.
Starting with version 8.1.6, global hints are available.You can use them to specify hints at the
upper query level that are meant for views accessed by that query, so that you do not need to
specify hints in the view definition. This is done with a special dot notation syntax; see Oracle9i
Database Performance Guide and Reference for details.
The SALES table is a partitioned table. For an explanation of partitioning, refer students to the
documentation.
Oracle9i: SQL Tuning Workshop 8-17
 Summary

In this lesson, you should have learned how to:

• Set the optimizer mode
• Use optimizer hint syntax
• Determine access path hints
• Analyze hints and their impact on views

Copyright © Oracle Corporation, 2002. All rights reserved.

Summary
This lesson introduced you to additional optimizer settings and hints.
You can set the optimizer mode to use either the cost-based method or rule-based method. The
supported optimizer mode values are CHOOSE, RULE, FIRST_ROWS(n), and ALL_ROWS.
By using hints, you can influence the CBO at the statement level. Use hints as a last remedy when
tuning SQL statements. There are several hint categories, one of which is hints for access path
methods.
To specify a hint, use the hint syntax within the SQL statement. Alternatively, you can use the
Oracle SQL Analyze graphical interface to specify a hint within SQL statements.
Be careful when using hints with views. Hints inside views or on views are handled differently
depending on whether or not the view is mergeable into the top-level query.
Instructor Note
After this lesson, students can complete Workshop 2 (sorting, grouping, and set functions). Note
that some theory about sorting is discussed in the beginning of the next lesson.

Oracle9i: SQL Tuning Workshop 8-18

 Practice Overview

This practice covers the following topics:

• Analyzing a query on a table without any statistics
• Using a hint to force a full table scan

Copyright © Oracle Corporation, 2002. All rights reserved.

Practice Overview
In this practice, you investigate RBO behavior for a query on the PROMOTIONS table, which is a
small table with an index. This is accomplished by deleting table statistics and setting the
optimizer mode to CHOOSE. Then you use a hint to force a full table scan on the table. Compare
the results and decide the best approach for this query.

Oracle9i: SQL Tuning Workshop 8-19

Practice 8
1. Delete all statistics for the PROMOTIONS table.
2. Display the contents of the PROMOTIONS table. Note that it is a relatively small table.
There is an index on the PROMO_ID column, to support the primary key constraint.
3. Alter your session and set the optimizer mode to CHOOSE.
4. Analyze the select statement shown.
SQL> select promo_id, promo_name, promo_cost
2 from promotions
3 where promo_id > 450;
5. What optimization method is used?
6. Include a hint to use a full table scan with the select statement used in step 4.
7. View the results of the EXPLAIN PLAN command. What optimization method is used?
With the PROMOTIONS table, which approach is more efficient?

Oracle9i: SQL Tuning Workshop 8-20

 Sorting and Joining

Copyright © Oracle Corporation, 2002. All rights reserved.

Schedule: Timing Topic

90 minutes Lecture
90 minutes Total
Instructor Note
This lesson is long, and includes a lot of detail about join techniques. Probably the best
approach is to teach the basics (until “Join Order Rules”), and have the participants perform
Workshop 3 (“Joins”). Then continue this lesson, and have them continue with Workshop 4
(“Subqueries”) at the end of the lesson.
 Objectives

After completing this lesson, you should be able to do

the following:
• Optimize sort performance
• Describe different join techniques
• Explain join optimization
• Find optimal join execution plans

Copyright © Oracle Corporation, 2002. All rights reserved.

Objectives
After completing this lesson, you should be able to:
• Optimize sort performance by using Top-N SQL
• Describe the techniques used for executing join statements
• Explain the optimization performed on joins
• Find the optimal execution plan for a join
Joins are probably the most frequent cause of performance problems.
Instructor Note
For more details about hash joins, see Oracle9i Performance and Tuning Guide.
Instructor Note (for page 9-4)
Demonstration script: demo09_01.sql
Instructor Note (for page 9-5)
Demonstration script: Run the demo09_02.sql script to demonstrate the code shown.

Oracle9i: SQL Tuning Workshop 9-2

 Tuning Sort Performance

• Sorting large sets can be expensive:

Tune sort parameters.
• Note that DISTINCT, GROUP BY, and most set
operators cause implicit sorts.
• Minimize sorting by one of the following:
– Try to avoid DISTINCT and GROUP BY
– Use UNION ALL instead of UNION
– Enable index access to avoid sorting

Copyright © Oracle Corporation, 2002. All rights reserved.

Tuning Sort Performance

You should contact your database administrator to tune sort parameters such as the following:
• SORT_MULTIBLOCK_READ_COUNT
• SORT_AREA_SIZE
• SORT_AREA_RETAINED_SIZE
For more details about these parameters, check the Oracle9i Reference.
If possible, write SQL statements in such a way that sorting is not needed. Note that several
SQL language components cause implicit sorts, such as DISTINCT, GROUP BY, UNION,
MINUS, and INTERSECT. Make sure that you get only the sorts that are needed for your result.
Sorts can be avoided by creating indexes. Concatenated indexes, with the appropriate DESC or
ASC attributes, can especially help avoid sorts. Note, however, that you loose the advantage of
multiblock I/O, because to avoid a sort the index must be scanned sequentially. An index fast
full scan (using multiblock I/O) does not guarantee that the rows are returned in the right order.
Instructor Note
The students have already investigated some sort behavior in Workshop 2 (sorting, grouping,
and set functions), such as implicit sorts and index access to avoid sorting.

Oracle9i: SQL Tuning Workshop 9-3

 Top-N SQL

SQL> SELECT *
2 FROM (SELECT prod_id
3 , prod_name
4 , prod_list_price
5 , prod_min_price
6 FROM products
7 ORDER BY prod_list_price DESC)
8 WHERE ROWNUM <= 5;

Copyright © Oracle Corporation, 2002. All rights reserved.

Top-N SQL
The idea behind the Top-N SQL feature is that you do not need to sort a full set if, for example,
you are only interested in the four highest (or lowest) values.
If a set is too big to be sorted in memory, the performance is significantly and especially
degrading. This is caused by the I/O to and from temporary storage of intermediate results on
disk.
If you only want to know the four highest values, you only need an array with four slots to scan
the set and keep the four highest values in that array.
The WHERE clause of the statement above is merged into the in-line view to prevent the full
PRODUCTS table from being sorted by PROD_LIST_PRICE. This appears in the execution
plan as follows:
Execution Plan
----------------------------------------------------------
0 SELECT STATEMENT Optimizer=CHOOSE (Cost=248 Card=5 Bytes=660000)
1 0 COUNT (STOPKEY)
2 1 VIEW (Cost=248 Card=10000 Bytes=660000)
3 2 SORT (ORDER BY STOPKEY) (Cost=248 Card=10000 Bytes=370000)
4 3 TABLE ACCESS (FULL) OF ’PRODUCTS’ (Cost=109 Card=10000
Bytes=370000)

Oracle9i: SQL Tuning Workshop 9-4

 Join Terminology

• Join statement
• Join predicate, nonjoin predicate
• Single-row predicate
SQL> SELECT c.cust_last_name, c.cust_first_name,
2 co.country_id, co.country_name
3 FROM customers c , countries co
4 WHERE c.country_id = co.country_id
5 AND co.country_id = ’JP’
6 OR c.cust_id = 205;

Copyright © Oracle Corporation, 2002. All rights reserved.

Join Terminology
A join statement is a select statement with more than one table in the FROM clause.
However, the optimizer uses the techniques usually associated with joins when executing other
kinds of statements where more than one table is referenced, such as in subqueries. Subqueries
are discussed later in this lesson. This is an example of a SQL:1999 compliant join statement
which is available in Oracle9i:
SQL> SELECT c.cust_last_name, c.cust_first_name,
2 co.country_id, co.country_name
3 FROM customers c JOIN countries co
4 ON (c.country_id = co.country_id) –- join predicate
5 AND co.country_id = 'JP' –- nonjoin predicate
6 OR c.cust_id = 205; -- single-row predicate
A join predicate is a predicate in the WHERE clause that combines the columns of two of the
tables in the join.
A nonjoin predicate is a predicate in the WHERE clause that references only one table.
A single-row predicate is an equality predicate on a column with a unique or primary key
constraint, or a column with a unique index without corresponding constraint. The optimizer
knows that these predicates always return one row or no rows at all.

Oracle9i: SQL Tuning Workshop 9-5

 Join Terminology

• Equijoin
SQL> SELECT c.cust_last_name, co.country_name
2 FROM customers c, countries co
3 WHERE c.country_id = co.country_id;

• Nonequijoin
SQL> SELECT c.cust_last_name, c.cust_credit_limit
2 , cr.credit_limit_id
3 FROM customers c, credit_limits cr
4 WHERE c.cust_credit_limit
5 BETWEEN cr.credit_limit_min
6 AND cr.credit_limit_max;

Copyright © Oracle Corporation, 2002. All rights reserved.

Join Terminology (continued)

An equijoin is a join statement where the join predicate shows the equal (=) sign between the
columns. An equijoin is the most common type of join. The statement above is an example of
an equijoin.
A nonequijoin refers to join statements where the join predicate uses one of the other
comparison operators, such as <, >, <=, >=, and <>. Nonequijoins are also called theta-joins.
They occur infrequently and may indicate a poor database design.
Instructor Note
The CREDIT_LIMITS table is not part of the standard SH schema. Run the
demo09_03.sql script to create the CREDIT_LIMITS table, insert rows, and run the
nonequijoin statement shown above.
Note the full table scan on both tables.
In Workshop 3, the students investigate an example of a nonequijoin.

Oracle9i: SQL Tuning Workshop 9-6

 Join Operations

• A join operation combines the output from two row

sources and returns one resulting row source.
• Join operation types include the following:
– Nested loops join
– Sort/merge join
– Hash join
– Cluster join
– Full outer Joins

Copyright © Oracle Corporation, 2002. All rights reserved.

Join Operations
A join operation combines the output from two row sources and returns one resulting row
source. Row sources were introduced in the “EXPLAIN and AUTOTRACE” lesson.
Join Operation Types
The Oracle9i Server supports different join operation types:
• Nested loops join
• Sort/merge join
• Hash join
• Cluster join
• Full outer Join
Instructor Note
Oracle9i also supports index joins as a new access path. The INDEX_JOIN hint instructs the
optimizer to use that path. This hint has an effect if a sufficiently small number of indexes exist
that (together) contain all columns required to resolve the query, so that access to that table can
be avoided. So it is not a different join technique, but rather a smart way to avoid table access
by using multiple indexes.

Oracle9i: SQL Tuning Workshop 9-7

 Nested Loops Joins

• One of the two tables is defined as the outer table or

the driving table.
• The other table is called the inner table.
• For each row from the outer table, all matching rows
in the inner table are retrieved.

For each row in the outer table

For each row in the inner table

Check for a match

Copyright © Oracle Corporation, 2002. All rights reserved.

Basic Execution Plan of Nested Loops Joins

NESTED LOOPS
TABLE ACCESS (…) OF our_outer_table
TABLE ACCESS (…) OF our_inner_table
Because they are often used with indexed table accesses, nested loops are common with an
execution plan that looks like following:
NESTED LOOPS
TABLE ACCESS (BY ROWID) OF our_outer_table
INDEX (... SCAN) OF outer_table_index (…..)
TABLE ACCESS (BY ROWID) OF our_inner_table
INDEX (RANGE SCAN) OF inner_table_index (NON-UNIQUE)
It is common that the outer (driving) table is accessed with a full table scan.
Nonequijoins
If you have a nonequijoin, a hash join is not possible. Depending on the nonjoin predicates,
nonequijoins can result in a nested loop with full table scans on both the outer and the inner
tables. This means that a full table scan is performed on the outer table; and for each row, a full
table scan is performed (again and again) on the inner table. This usually results in poor
performance.

Oracle9i: SQL Tuning Workshop 9-8

 Nested Loops Join Plan

Nested loops

2 3

Table access Table access

(Outer/driving table) (Inner table)

Copyright © Oracle Corporation, 2002. All rights reserved.

Nested Loops Join Plan

For each row from the outer (driving) table, all rows from the inner table that satisfy the join
predicate are retrieved.
Any nonjoin predicates on the inner table are considered after this initial retrieval, unless a
composite index, which combines both the join and the nonjoin predicate, is used.
The nested loops join can solve all kinds of joining requirements, no matter what the join
predicate is. Therefore, you should use the nested loops join when you cannot use sort/merge,
hash, or cluster joins. These join techniques are discussed later in this lesson.

Oracle9i: SQL Tuning Workshop 9-9

 Sort/Merge Joins

A sort/merge join is executed as follows:

• The rows from each row source are sorted
on the join predicate columns
• The two sorted row sources are then merged
and returned as the resulting row source

Copyright © Oracle Corporation, 2002. All rights reserved.

Sort/Merge Joins
In the sort operations, the two row sources are sorted on the values of the columns used in the
join predicate. If a row source has already been sorted in a previous operation, the sort merge
operation skips the sort on that row source.
Sorting could make this join technique expensive, especially if sorting cannot be performed in
memory.
The merge operation combines the two sorted row sources to retrieve every pair of rows that
contain matching values for the columns used in the join predicate.
Basic Execution Plan
MERGE (JOIN)
SORT (JOIN)
TABLE ACCESS (…) OF tableA
SORT (JOIN)
TABLE ACCESS (…) OF tableB

Oracle9i: SQL Tuning Workshop 9-10

 Sort/Merge Join Plan

Merge 1

Sort 2 3 Sort

4 5
Table access Table access

Copyright © Oracle Corporation, 2002. All rights reserved.

Sort/Merge Join Plan

The table accesses (row sources 4 and 5) can be based on index scans if there are nonjoin
predicates that can use an index. However, this join operation most often appears with full table
scans on both tables.
The sequence (which table is sorted first) does not matter; there is no concept of an
outer/driving table. Remember that the sorting can result in the creation of temporary segments
on disk (giving more I/O than just the table scans themselves).
Sort/merge joins tend to perform better than an indexed nested loops join if the number of rows
satisfying the join condition represents a major part of the total number of rows in the two
tables.
Note: The break-even point for a sort/merge join and an indexed nested loops join depends on
the sorting performance of the database. Sort performance can be influenced by defining
temporary tablespaces of tablespace type TEMPORARY, setting the SORT_AREA_SIZE
parameter to define the amount of memory to use for a sort, and setting
SORT_AREA_RETAINED_SIZE to release memory after a sort. This is typically a DBA task.

Oracle9i: SQL Tuning Workshop 9-11

 Hash Joins

A hash join is executed as follows:

• Both tables are split into as many partitions as
required, using a full table scan.
• For each partition pair, a hash table is built in
memory on the smallest partition.
• The other partition is used to probe the hash table.

Copyright © Oracle Corporation, 2002. All rights reserved.

Hash Joins
The Oracle9i Server considers hash joins only when you use the cost-based optimizer. Hash
joins, like sort/merge joins, can be performed only for equijoins. These are the steps performed
for a hash join:
• The Oracle server performs a full table scan on both tables and splits each into as many
partitions as required, based on the available memory.
• The Oracle server builds a hash table on the smallest partition and uses the other partition
to probe the hash table. All partition pairs that do not fit into memory are placed onto disk.
The number of partitions into which the tables are split depends on the amount of available
memory.
Basic Execution Plan
HASH JOIN
TABLE ACCESS (..) OF tableA
TABLE ACCESS (..) OF tableB
Performance Considerations
As a general rule, hash joins outperform sort/merge joins. In some cases, a sort/merge join can
outperform a nested loops join; it is even more likely that a hash join will.

Oracle9i: SQL Tuning Workshop 9-12

 Hash Join Plan

Hash join

2 3

Table access Table access

Copyright © Oracle Corporation, 2002. All rights reserved.

Hash Join Plan

To execute a hash join, the Oracle server performs these steps:
• Steps 2 and 3 perform full table scans of the tables that must be joined.
• Step 1 builds a hash table out of the rows coming from step 2 and probes it with each row
coming from step 3.
Note: The HASH_AREA_SIZE initialization parameter controls the amount of memory used
for hash join operations and the HASH_MULTIBLOCK_IO_COUNT initialization parameter
controls the number of blocks that a hash join operation should read and write concurrently.

Oracle9i: SQL Tuning Workshop 9-13

 Joining Multiple Tables

You can join only two row sources at a time. Joins

with more than two tables are executed as follows:
1. Two tables are joined, resulting in a row source.
2. The next table is joined with the row source
resulting from step 1.
3. Repeat step 2 until all tables are joined.

Copyright © Oracle Corporation, 2002. All rights reserved.

Joining Multiple Tables

The Oracle9i Server can join only two row sources at a time.
Join operations (such as nested loops and sort/merge) are used as building blocks if the join
statement contains more than two tables.
Except for the limitations inherent in each kind of join statement, there are no restrictions on
the combination of join operations involved in a join with more than two tables.
Thus, you may see nested loops operations mixed with hash joins or sort/merge joins in the
same execution plan.

Oracle9i: SQL Tuning Workshop 9-14

 Outer Joins

• Join predicates with a plus (+) sign

• Nonjoin predicates with a plus (+) sign
• Predicates without a plus (+) sign will disable outer
join functionality. initialization parameter
SQL> SELECT s.time_id, t.time_id
2 FROM sales s, times t
3 WHERE s.time_id (+) = t.time_id;

Copyright © Oracle Corporation, 2002. All rights reserved.

Outer Joins
An outer join is a join where the join predicates have a plus (+) sign added to signify that the
join returns a row even if the table with this special predicate (the outer joined table) does not
have a row that satisfies the join condition.
Outer Join Example
SQL> select s.time_id, t.time_id
2 from sales s, times t
3 where s.time_id (+) = t.time_id -- join predicate with +
4 and s.promo_id (+) = 19 -- non-join predicate with +
5 and s.quantity_sold (+) > 45 -- non-join predicate with +
6 /
Note: Predicates without a plus (+) sign on the table that is outer joined will disable the outer
join functionality.
Instructor Note
Demonstration script: Run the demo09_04.sql script to demonstrate the code shown. The
first example shows the outer join disabled, the second join shows the outer join enabled.

Oracle9i: SQL Tuning Workshop 9-15

 SQL: 1999 Outer Joins

• No plus (+) sign

• Keywords OUTER JOIN
SQL> SELECT s.time_id, t.time_id
2 FROM sales s
3 RIGHT OUTER JOIN times t
4 ON (s.time_id = t.time_id);

Copyright © Oracle Corporation, 2002. All rights reserved.

Outer Joins
In Oracle9i SQL:1999 compliant join syntax has been introduced. So now you can specify
RIGHT, LEFT, and FULL OUTER joins to create the outer joins instead of using the (+) sign.
This is not intended to improve performance. The usage of (+) is still supported.
Note: This statement is the equivalent to the statement in the previous slide.

Oracle9i: SQL Tuning Workshop 9-16

 Full Outer Joins

• A full outer join acts like a combination of the left

and right outer joins.
• In addition to the inner join, rows from both tables
that have not been returned in the result of the inner
join are preserved and extended with nulls.
• This syntax has been introduced in Oracle9i.
SQL> SELECT c.cust_id, c.cust_last_name
2 , co.country_name
3 FROM customers c
4 FULL OUTER JOIN countries co
5 ON (c.country_id = co.country_id);

Copyright © Oracle Corporation, 2002. All rights reserved.

Outer Joins
A full outer join acts like a combination of the left and right outer joins. In addition to the
inner join, rows from both tables that have not been returned in the result of the inner join are
preserved and extended with nulls. In other words, with full outer joins you can join tables
together, yet still show rows which do not have corresponding rows in joined-to tables.
For both the cost-based and rule-based approaches, the optimizer first determines whether
joining two or more of the tables definitely results in a row source containing at most one
row. The optimizer recognizes such situations based on UNIQUE and PRIMARY KEY
constraints on the tables. If such a situation exists, then the optimizer places these tables first
in the join order. The optimizer then optimizes the join of the remaining set of tables.
The CBO, generates a set of execution plans based on the possible join orders, join methods,
and available access paths. The optimizer then estimates the cost of each plan and chooses the
one with the lowest cost. The optimizer estimates costs in these ways:
The cost of a nested loops operation is based on the cost of reading each selected row of the
outer table and each of its matching rows of the inner table into memory. The optimizer
estimates these costs using the statistics in the data dictionary. The cost of a sort merge join is
based largely on the cost of reading all the sources into memory and sorting them. The CBO’s
choice of join orders can be overridden with the ORDERED hint. If the ORDERED hint
specifies a join order that violates the rule for an outer join, then the optimizer ignores the hint
and chooses the order. Also, you can override the optimizer’s choice of a join method with
hints.
Oracle9i: SQL Tuning Workshop 9-17
 Execution of Outer Joins

• Indexes can be used for outer join predicates.

• Nonjoin predicates on the outer joined table can use
indexes only if no index is used for the join
predicate.
SQL> SELECT c.cust_id, co.country_name
2 FROM customers c , countries co
3 WHERE c.country_id(+) = co.country_id
4 AND co.country_id = ’IT’;

Copyright © Oracle Corporation, 2002. All rights reserved.

Outer Join Examples

SQL> SELECT c.cust_id, co.country_name
2 FROM customers c , countries co
3 WHERE c.country_id(+) = co.country_id;
The optimizer chooses to use a HASH_JOIN.
SQL> SELECT c.cust_id, co.country_name
2 FROM customers c , countries co
3 WHERE c.country_id(+) = co.country_id
4 AND co.country_id = ’SA’;
The optimizer can use an index on the co.country_id column (a single-row predicate)
COUNTRIES is the driving table.
SQL> SELECT co.country_name, c.cust_id
2 FROM customers c , countries co
3 WHERE c.country_id(+) = co.country_id
4 and c.cust_id (+) =780;
This statement is an enabled outer join; both the index on c.country_id and
co.cust_id are usable. If you remove the (+) in the join predicate, the outer join is
disabled.
Instructor Note
Demonstration script: demo09_05.sql

Oracle9i: SQL Tuning Workshop 9-18

 The Optimizer and Joins

The optimizer decides about the following:

1. The order in which to join the tables
2. Which join operation to apply for each join
3. The access path for each row source

Copyright © Oracle Corporation, 2002. All rights reserved.

The Optimizer and Joins

The Oracle9i optimizer optimizes joins in three phases. First, the optimizer generates a list of
all possible join orders. Then, for each possible join order, the best join operation is selected.
Finally, the Oracle9i optimizer determines the best access path for each row source of the
execution plan.
Instructor Note
Do not go into too much detail here. This is covered in more detail in the “Everything You
Always wanted to Know about the Oracle Optimizer” seminar (65320GC10), by using event
10053.

Oracle9i: SQL Tuning Workshop 9-19

 Join Order Rules

Rule 1
A single-row predicate forces its row source to be
placed first in the join order.

Rule 2
For outer joins, the table with the outer join operator
(+) must come after the other table in the join order
for processing the join.

Copyright © Oracle Corporation, 2002. All rights reserved.

Join Order Rules

• A single-row predicate forces its row source to be placed first in the join order.
• For outer join statements, the table with the outer join operator must come after (in the
join order) the other table in the join predicate.
By making a single-row predicate table the driving table of a join operation, the optimizer
effectively eliminates one join operation. For example, for a nested loops join, the main loop is
not a real loop anymore. The single-row predicate results in one row (or no rows at all), so the
inner loop is executed only once (or not at all).
The reason for the second rule (about outer joins) is that you would otherwise be unable to
produce the correct outer join results. For example, if you join the COUNTRIES table with the
CUSTOMERS table and you also want to see countries that have no corresponding customers,
you must start with the COUNTRIES table.
After considering these two rules, determine the join order and join operations based on the
optimizer goal that is used.

Oracle9i: SQL Tuning Workshop 9-20

 RBO Join Optimization

• As a first step, all possible join orders are

generated.
• This potentially gives the following:
Number of Tables Join Orders
---------------- -----------
2 2! = 2
3 3! = 6
4 4! = 24

• Parse time grows exponentially when adding tables

to a join.

Copyright © Oracle Corporation, 2002. All rights reserved.

RBO Join Optimization

The number of possible execution plans is determined by the number of tables in the join and
the specified join predicates. Theoretically, the number of possible execution plans is:
Number of Tables Possible Plans

2 2
3 3! = 3 * 2 = 6
4 4! = 4 * 3 * 2 = 24
5 5! = 5 * 4 * 3 * 2 = 120
... ...

Because only the execution plans that follow a join predicate are considered, the number of
possible execution plans usually does not reach this value when the number of tables is greater
than 3. However, joining 20 tables still results in many possible execution plans. Remember
that all tables involved in a join, including those that are hidden in views, count as separate
tables.
Instructor Note
The exclamation mark used on the slide is the factorial operator from mathematics.

Oracle9i: SQL Tuning Workshop 9-21

 RBO Join Optimization

The following rules are applied in order:

1. Minimize nested loops operations in which the
inner table is accessed with a full table scan.
2. Tie: Minimize sort/merge operations.
3. Still a tie: Choose a driving table
with the best-ranked access path.
4. Still a tie: Choose a driving table
that appears later in the FROM clause.

Copyright © Oracle Corporation, 2002. All rights reserved.

RBO Join Optimization (continued)

These are the rules applied by the rule-based optimizer, in the specified order:
1. Try to minimize nested loops operations in which the inner table is accessed with a full
table scan.
2. If there is a tie, try to minimize sort/merge operations.
3. If there is still a tie, choose a driving table with the best-ranked access path.
4. If there is still a tie, choose a driving table that appears later in the FROM clause
(the right-to-left rule).

Oracle9i: SQL Tuning Workshop 9-22

 CBO Join Optimization

• A set of possible execution plans is generated,

limited by two parameters:
– OPTIMIZER_SEARCH_LIMIT
– OPTIMIZER_MAX_PERMUTATIONS
• The optimizer then estimates the cost of each plan
and chooses the one with the lowest cost.

Copyright © Oracle Corporation, 2002. All rights reserved.

CBO Join Optimization

The number of different execution plans considered is limited by the value of the
OPTIMIZER_SEARCH_LIMIT parameter, which specifies the maximum number of tables for
which the optimizer considers a Cartesian product. The default value is 5. As soon as you join
more tables, the optimizer considers only regular joins; that is, it considers only join orders
where the tables are connected by equijoin predicates.
The OPTIMIZER_MAX_PERMUTATIONS initialization parameter directly limits the number
of permutations of the join order that the optimizer considers. The default value is 80,000. By
reducing this value, you can ensure that parse times for queries joining lots of tables stay within
acceptable limits; the disadvantage is that the optimizer may miss the optimal plan.

Oracle9i: SQL Tuning Workshop 9-23

 Estimating Join Costs

• Nested loops join:

Cost » read(A) + card(A) * read(B)
– OPTIMIZER_INDEX_CACHING parameter
• Sort/merge join:
Cost » read(A) + sort(A) + read(B) + sort(B)

Copyright © Oracle Corporation, 2002. All rights reserved.

Estimating Join Costs

The cost of a nested loops join is based on the cost of reading each selected row of the outer
table and each of its matching rows of the inner table.
The cost of a sort/merge join is based on the cost of reading and sorting both row sources. The
cost-based optimizer also considers other factors when determining the cost of each operation.
For example:
• A smaller sort area size is likely to increase the cost for a sort/merge join because sorting
takes more CPU time and I/O in a smaller sort area.
• A larger multiblock read count is likely to decrease the cost for a sort/merge join in
relation to a nested loops join. If a large number of sequential blocks can be read from
disk in a single I/O, an index on the inner table for the nested loops join is less likely to
improve performance over a full table scan.
Another important parameter that influences the CBO cost estimates for nested loops joins is
OPTIMIZER_INDEX_CACHING. The range of values is 0–100, and the default value is 0. By
setting this parameter to a higher value, the CBO will assume that any indexes on the inner
table will be in the buffer cache. This means that a nested loops join will be considered to be
more attractive.
Instructor Note
The curly equal sign (») is used in the slide to denote “is approximately equal to.” Only the
main cost components are mentioned in the formulas.
Oracle9i: SQL Tuning Workshop 9-24
 Star Joins

PRODUCTS CUSTOMERS

SALES

CHANNELS PROMOTIONS TIMES

Copyright © Oracle Corporation, 2002. All rights reserved.

Star Joins
One type of data warehouse design centers on what is known as a star schema, which is
characterized by one or more very large fact tables that contain the primary information in the
data warehouse and a number of much smaller dimension tables (or lookup tables), each of
which contains information about the entries for a particular attribute in the fact table.
A star join is a join between a fact table and a number of lookup tables. Each lookup table is
joined to the fact table by a primary-key to foreign-key join, but the lookup tables are not joined
to each other. The fact table normally has a concatenated index on the key columns to facilitate
this type of join.
A star join is executed using the normal join operations, but using a join order that does not
correspond to the join predicates: the smaller dimension tables are joined together first (like a
Cartesian product). The resulting row source is then used for a concatenated index access into
the fact table using a indexed nested loops join.
Note: The star transformation, which is discussed in the “Advanced Indexes” lesson, should
not be confused with the star joins. Whereas star joins work well for schemas with a small
number of dimensions and dense fact tables, star transformation works well for schemas with a
large number of dimensions and sparse fact tables.

Oracle9i: SQL Tuning Workshop 9-25

Star Join Example

SQL> SELECT amount_sold,prod_name, promo_name,

2 FROM sales s, products p, promotions o
3 WHERE s.prod_id =p.prod_id
4 AND s. promo_id = o.promo_id;
In this example, you must construct a concatenated index on the columns PROD_ID and
PROMO_ID of the SALES table to enable star query optimization.
Note: Star query optimization is considered only when using the cost-based optimizer (CBO).
Star joins usually outperform the corresponding normal multitable join because the optimizer
can start with a small dimension table as the driving table, but must join that table with the
(large) fact table. This produces a large intermediate result that must be joined with all
remaining dimension tables. The advantage of a star join is that the fact table is the last table in
the join order.
Instructor Note
It seems that the concatenated index is no longer needed since release 8.0.5.

Oracle9i: SQL Tuning Workshop 9-26

 Hints for Join Orders

SQL> select /*+ ORDERED */...

2 from T1, T2, ...
3 where ...

SQL> select /*+ LEADING(T1) */...

2 from T1, T2, ...
3 where ...

SQL> select /*+ STAR */...

2 from T1, T2, ...
3 where ...

Copyright © Oracle Corporation, 2002. All rights reserved.

Hints for Join Orders

Specifying a hint forces the CBO to be used. Note that specifying a hint such as an index hint
for one of the tables in the join results in cost-based optimization to be used for the other tables
in the join, as well.
ORDERED()
The ORDERED() hint joins the tables in the order in which they appear in the FROM clause,
from left to right. You can experiment with different join orders by changing the table order in
the FROM clause.
LEADING()
The LEADING() hint causes the optimizer to use the specified table as the first table in the
join order. The difference with the ORDERED hint is that LEADING determines only the driving
table.
STAR
The STAR hint forces a star join plan to be used if possible. A star join plan has the largest table
in the query last in the join order and joins it with a nested loops join on a concatenated index.
The STAR hint applies when there are at least three tables, the large table’s concatenated index
has at least three columns, and there are no conflicting access or join method hints. The
optimizer also considers different permutations of the small tables.
Instructor Note
The LEADING hint was introduced in Oracle8i release 8.1.6.

Oracle9i: SQL Tuning Workshop 9-27

 Hints for Join Operations

SQL> select /*+ USE_NL(T2) */...

2 from T1, T2, ...
3 where ...

SQL> select /*+ USE_MERGE(T2) */...

2 from T1, T2, ...
3 where ...

SQL> select /*+ USE_HASH(T2) */...

2 from T1, T2, ...
3 where ...

Copyright © Oracle Corporation, 2002. All rights reserved.

Hints for Join Operations

In the slide, the three hints to influence the join operation accept one or more table names (or
aliases). It is important to know that the table name (or alias) you specify should be the inner
table, not the driving table, for the join operation you want to be performed. See also the last
note on the next page.
Also note that the USE_HASH and USE_MERGE hints are always ignored for nonequijoins,
because nonequijoins can be resolved only with a nested loops join operation.
Examples
The USE_NL() hint causes each specified table to be joined with another row source with a
nested loops join using the specified table as the inner table:
SQL> select /*+ USE_NL(T2 T3) ORDERED */ ...
2 from T1, T2, T3
3 where ...
This should give the following execution plan structure:
NESTED LOOPS
TABLE ACCESS (…) OF T1
NESTED LOOPS
TABLE ACCESS (…) OF T2
TABLE ACCESS (…) OF T3
Oracle9i: SQL Tuning Workshop 9-28
Hints for Join Operations (Continued)
Examples (continued)
The USE_MERGE hint causes each specified table to be joined with the previous row source in
the join order with a sort-merge join:
SQL> select /*+ USE_MERGE(TAB2) */ ...
2 from TAB1, TAB2
3 where ...
The USE_HASH hint causes each specified table to be joined with the previous row source in the
join order with a hash join:
SQL> select /*+ USE_HASH(TAB2) */ ...
2 from TAB1, TAB2
3 where ...
Using the USE_NL(), USE_MERGE(), and USE_HASH() Hints
The Oracle9i Server uses these three join operation hints when the referenced table (or alias) is
forced to be the inner table of a join, and they are ignored if the referenced table (or alias) is the
outer table. That is why these hints are often combined with the ORDERED hint.
If you have the idea that the optimizer is ignoring your USE_NL, USE_MERGE, or USE_HASH
hints, the Oracle9i optimizer probably found a better plan for a different join order.
To eliminate this possibility, you can add the ORDERED hint and change the order of the tables
in your FROM clause, to force a certain join order.
It is possible to specify two arguments in the USE_HASH() and USE_MERGE() hints when
both row sources are tables instead of a table being joined to a result set.

Oracle9i: SQL Tuning Workshop 9-29

 Other Join Hints

SQL> select /*+ STAR_TRANSFORMATION */...

2 from T1, T2, ...
3 where ...

SQL> select /*+ DRIVING_SITE(T1) */...

2 from T1, T2, ...
3 where ...

Copyright © Oracle Corporation, 2002. All rights reserved.

STAR_TRANSFORMATION
The hint in the slide is not really a join hint, but rather a hint to optimize data access for the
facts table of a star schema. Refer to the “Advanced Indexes” lesson for a further description of
this hint.
DRIVING_SITE
The DRIVING_SITE hint forces query execution to be done at a different site than that
selected by the Oracle9i Server. This hint can be used with either rule-based or cost-based
optimization.
Example
SQL> SELECT /*+ DRIVING_SITE(co) */
2 c.cust_last_name
3 , c.cust_first_name
4 , co.country_name
5 FROM customers c
6 , countries@rsite co
7 WHERE c.country_id = co.country_id;
Instructor Note
Check whether your students are familiar with the countries@rsite notation for remote
tables; rsite is a database link reference.
Oracle9i: SQL Tuning Workshop 9-30
 Subqueries and Joins

• Subqueries, like joins, are statements that reference

multiple tables.
• Subquery types:
– Noncorrelated subqueries
– Correlated subqueries
– NOT IN subqueries (anti-joins)
– EXISTS subqueries (semijoins)

Copyright © Oracle Corporation, 2002. All rights reserved.

Noncorrelated Subqueries
A noncorrelated subquery does not contain any references to the outer (main) query and can be
executed independently. For example:
SQL> SELECT c.*
2 FROM customers c
3 WHERE c. country_id IN
4 (SELECT co.country_id
5 FROM countries co
6 WHERE co.country_subregion = ’Asia’);
This statement is executed as a hash join with the subquery as the outer table. It retrieves all
customers in location Asia. This is the execution plan structure:
Execution Plan
---------------------------------------------------------
0 SELECT STATEMENT Optimizer=CHOOSE (Cost=363 Card=6250
Bytes= 918750)
1 0 HASH JOIN (Cost=363 Card=6250 Bytes=918750)
2 1 TABLE ACCESS (FULL) OF ’COUNTRIES’ (Cost=1 Card=2
Bytes=28)
3 1 TABLE ACCESS (FULL) OF ’CUSTOMERS’ (Cost=360
Card=50000 Bytes=6650000)

Oracle9i: SQL Tuning Workshop 9-31

Correlated Subqueries
This example of a correlated subquery finds the latest promotion date for every promotion:
SQL> SELECT p1.*
2 FROM promotions p1
3 WHERE p1.promo_begin_date =
4 (SELECT MAX(p2.promo_begin_date)
5 FROM promotions p2
6 WHERE p1.promo_id = p2.promo_id);
The above query has the following execution plan:
Execution Plan
----------------------------------------------------------
0 SELECT STATEMENT Optimizer=CHOOSE (Cost=13 Card=3 Bytes=294)
1 0 HASH JOIN (Cost=13 Card=3 Bytes=294)
2 1 VIEW OF ’VW_SQ_1’ (Cost=9 Card=501 Bytes=11022)
3 2 SORT (GROUP BY) (Cost=9 Card=501 Bytes=5010)
4 3 TABLE ACCESS (FULL) OF ’PROMOTIONS’ (Cost=3 Card=501
Bytes=5010)
5 1 TABLE ACCESS (FULL) OF ’PROMOTIONS’ (Cost=3 Card=501
Bytes=38076)
The hash join operation in this case can be considered as access to p1 as the outer table and the
access to p2 as the inner table.
Anti-Joins
An anti-join is a select statement with a subquery with a NOT IN predicate. For example:
SQL> SELECT c.*
2 FROM customers c
3 WHERE c.cust_income_level = ’F: 110,000 - 129,999’
4 AND c.country_id NOT IN
5 (SELECT co.country_id
6 FROM countries co
7 WHERE co.country_subregion = ’Europe’);
An anti-join returns rows from the left side of the predicate for which there are no corresponding
rows on the right side of the predicate.
Semijoins
A semijoin returns rows that match an EXISTS subquery. For example:
SQL> SELECT c.*
2 FROM customers c
3 WHERE EXISTS (SELECT ’x’
4 FROM countries co
5 WHERE c.country_id = co.country_id);

Oracle9i: SQL Tuning Workshop 9-32

 Initialization Parameters that Influence
Joins

• Hash join parameters:

– HASH_JOIN_ENABLED
– HASH_AREA_SIZE
– HASH_MULTIBLOCK_IO_COUNT
• Sort parameters:
– SORT_AREA_SIZE
– SORT_AREA_RETAINED_SIZE
– SORT_MULTIBLOCK_READ_COUNT

Copyright © Oracle Corporation, 2002. All rights reserved.

Hash Join Parameters

HASH_JOIN_ENABLED, HASH_AREA_SIZE, and HASH_MULTIBLOCK_IO_COUNT all
control the way hash joins are performed.
HASH_JOIN_ENABLED
This parameter specifies whether the optimizer should consider using a hash join as a join
method. (When set to FALSE, hash joining is turned off; that is, it is not available as a join
method that the optimizer can consider. However, hash joining is still available on statement
level—using a hint—even if the HASH_JOIN_ENABLED parameter is set to FALSE.)
HASH_AREA_SIZE
This parameter specifies the maximum amount of memory, in bytes, to be used for hash joins.
HASH_MULTIBLOCK_IO_COUNT
This parameter specifies how many sequential blocks a hash join reads and writes in one I/O.
Sort Parameters
Several sort parameters influence how the Oracle9i Server performs a sort operation. Sort
performance, especially when dealing with large amounts of data, varies a great deal based on
different values of these parameters. Sort performance directly influences sort/merge join
performance.

Oracle9i: SQL Tuning Workshop 9-33

 Throwaway of Rows

• Rows that are retrieved but not used

• Identify throwaway by comparing the number of
rows of the following:
– The join operation
– The input row sources

Copyright © Oracle Corporation, 2002. All rights reserved.

Throwaway of Rows
Compare the number of rows from the two input row sources with the number of rows from the
join operation. If both input row sources have more rows than the join operation, you have
identified a throwaway of rows.
Examples
Rows Operation
100 NESTED LOOPS
45 TABLE ACCESS (…) OF our_outer_table
4530 TABLE ACCESS (…) OF our_inner_table
In this example, approximately 98% of the retrieved rows (4,430 out of 4,530 rows from the
inner table) are thrown away, so there should be room for performance improvement.
Rows Operation
100 MERGE (JOIN)
23 SORT (JOIN)
23 TABLE ACCESS (…) OF table_A
150 SORT (JOIN)
150 TABLE ACCESS (…) OF table_B
In this example, only a small fraction of the retrieved rows are thrown away, so you cannot
expect significant performance improvements by reducing throwaway. TKPROF is required to
determine the throwaway rows because AUTOTRACE does not show this .

Oracle9i: SQL Tuning Workshop 9-34

 Minimize Throwaway of Rows

• Compare available indexes with join and nonjoin

predicates.
• Consider hinting for a nested loop instead of
sort/merge operations.

Copyright © Oracle Corporation, 2002. All rights reserved.

Minimize Throwaway or Rows

Comparing available indexes with predicates means returning to the task of finding the optimal
access path for a single table access. In several situations, the data model itself is the core of a
throwaway problem. Consider the following two examples.
N-Way Joins
An N-way join is a join with more than two tables, where all tables reference one another. If the
references are of such a type that the fully specified access to any of the three tables requires
the foreign keys from both the other tables, then the join will have a built-in throwaway of
rows. To improve performance, either change the data model or experiment to find the join
order with the least amount of throwaway.
Sick Join Predicates
SQL> SELECT …
2 FROM …
3 WHERE a.key1 = b.key1
4 AND a.key2 = DECODE(a.key2, 0, a.key2, b.key2)
A sick join condition only enables the join to use the first (clean) join predicate for the join
operation and then throw away the row pairs that do not satisfy the second join predicate.
To improve performance, change the second predicate to:
4 AND (a.key2 = b.key2 OR a.key2 = 0)
This will not remove the throwaway of rows, but should limit it significantly.

Oracle9i: SQL Tuning Workshop 9-35

 Minimize Processing

• Changing nested loops joins into sort/merge joins

can do the following:
– Remove index access
– Add sort overhead
• Theoretically, hash joins are the most efficient join
operation.
• Cluster joins need less I/O than the corresponding
nested loops join.

Copyright © Oracle Corporation, 2002. All rights reserved.

Minimize Processing
Changing nested loops joins into sort/merge joins can remove index access but might add
significant sorting overhead.
Hash joins require less processing than the same sort/merge operation; hash joins are
theoretically the fastest algorithm to join tables. However, they need appropriate tuning.
Cluster joins need less I/O than the corresponding nested loops join because the child rows that
belong to one parent row are stored together in the same logical cluster block.

Oracle9i: SQL Tuning Workshop 9-36

 Summary

In this lesson, you should have learned how to:

• Use the Top-N SQL feature
• Describe available join operations
• Optimize join performance against different
requirements
• Influence the join order
• Discover that tuning joins is more complicated than
tuning single table statements

Copyright © Oracle Corporation, 2002. All rights reserved.

Summary
This lesson introduced you to Top-N SQL, performance issues with sorting, and how to
optimize join statements.
There are several different join operations that Oracle9i supports:
• Nested loops join
• Sort/merge join
• Hash join
• Cluster join
• Full outer Join
The Oracle9i optimizer optimizes joins in three phases. First, the optimizer generates a list of
all possible join orders. Then, for each possible join order, the best join operation is selected.
Finally, the Oracle9i optimizer determines the best access path for each row source of the
execution plan. RBO differs from CBO. Specifying a hint forces the CBO to be used.
Throwaway of rows occurs when both input row sources have more rows than the result of the
join operation. Minimize throwaway of rows and improve performance.
Instructor Note
Workshops 3 (Joins) and 4 (Subqueries) can be done at this point, if not completed already.
Oracle9i: SQL Tuning Workshop 9-37
Oracle9i: SQL Tuning Workshop 9-38
 Optimizer Plan Stability

Copyright © Oracle Corporation, 2002. All rights reserved.

Schedule: Timing Topic

40 minutes Lecture
30 minutes Practice
70 minutes Total
 Objectives

After completing this lesson, you should be able to do

the following:
• Identify the purpose and benefits of optimizer plan
stability
• Create stored outlines
• Use stored outlines
• Edit stored outlines
• Maintain stored outlines

Copyright © Oracle Corporation, 2002. All rights reserved.

Objectives
After completing this lesson, you should be able to:
• Identify the purpose and benefits of optimizer plan stability
• Create stored outlines
• Use stored outlines
• Edit stored outlines
• Maintain stored outlines

Oracle9i: SQL Tuning Workshop 10-2

 Optimizer Plan Stability

• Enables well-tuned applications to force the use of

the desired SQL access path
• Maintains consistent execution plans through
database changes
• Is implemented using stored outlines consisting of
hints
• Groups stored outlines in categories

Copyright © Oracle Corporation, 2002. All rights reserved.

Optimizer Plan Stability

For every query, the optimizer prepares a tree of operations called an execution plan that
defines the order and methods of operation that the Oracle server follows to execute the
query.
Because the optimizer may have incomplete information, sometimes the best possible plan is
not chosen. In these cases, you may find it worthwhile to influence the optimizer’s plan
selection by rewriting the SQL statement, by using hints, or by using other tuning techniques.
When satisfied, you may want to ensure that the same tuned plan is generated whenever the
same query is recompiled, even when factors that affect optimization have changed.
Oracle9i provides you with a means of stabilizing execution plans across Oracle releases,
database changes, or other factors that can cause an execution plan to change. You can create
a stored outline containing a set of hints used by the optimizer to create an execution plan.
Stored outlines can be grouped in named categories. Different categories are useful for
creating groups of execution paths for different situations. The same SQL statement may
have a stored outline in more than one category. For example, you may want to have an
OLTP category and a DSS category.
Instructor note
Use EXPLAIN or SQL Trace and TKPROF to get the correct execution plan. In the next two
pages, the CREATE_STORED_OUTLINES and USE_STORED_OUTLINES parameters
work with ALTER SYSTEM and ALTER SESSION, but not in the init.ora file. This
was done within the initiative to reduce the overall number of Oracle8i initialization
parameters.

Oracle9i: SQL Tuning Workshop 10-3

 Plan Equivalence

• SQL statement text must match.

• Plans are maintained through:
– New Oracle versions
– New statistics on objects
– Initialization parameter changes
– Database reorganizations
– Schema changes
• Plan equivalence can control execution plans for
third-party applications.

Copyright © Oracle Corporation, 2002. All rights reserved.

SQL Statement Equivalence

Plan stability relies on exact textual matching of queries when determining whether a query
has a stored outline. This is the same criterion that is used to determine whether an execution
plan in the shared pool can be reused.
Stored outlines rely partially on hints that the optimizer uses to achieve stable execution
plans. Therefore, the degree to which a plan remains equivalent depends on the capabilities
of the hints that the plan uses. The execution steps included in a stored outline include row
access methods, join order, join methods, distributed accesses, and view/subquery merging.
Distributed access does not include the execution plan on the remote node.
Plan Stability
These plans are maintained through many types of database and instance changes. Therefore,
if you develop applications for mass distribution, you can use stored outlines to ensure that
all your customers access the same execution plans. For example, if the schema is changed
by adding an index, the execution plans do not change. You must generate new stored
outlines, or disable the plan stability feature, to benefit from the new index.

Oracle9i: SQL Tuning Workshop 10-4

 Creating Stored Outlines

• For all statements during a session:

SQL> ALTER SESSION
2 SET CREATE_STORED_OUTLINES = TRAIN;
SQL> SELECT ... ;
SQL> SELECT ... ;
• For a specific statement:
SQL> CREATE OR REPLACE OUTLINE CU_CO_JOIN
2 FOR CATEGORY TRAIN ON
3 SELECT co.country_name,
4 cu.cust_city, cu.cust_last_name
5 FROM countries co , customers cu
6 WHERE co.country_id = cu.country_id
...

Copyright © Oracle Corporation, 2002. All rights reserved.

Creating Stored Outlines

You can create outlines for a session, or you can create them for specific SQL statements.
Outlines can derive their input from both the RBO and CBO. However, the Oracle9i
optimizer can only use outlines when using the CBO, because outlines rely on hints.
CREATE_STORED_OUTLINES Parameter
The Oracle server creates stored outlines automatically when you set this parameter to TRUE
or to a category name. When the parameter is set to TRUE, the DEFAULT category is used.
When the parameter is activated, the Oracle server creates outlines for all executed SQL
statements. You can deactivate this by setting the parameter to FALSE. When you use this
parameter, the Oracle server generates the names for each stored outline.
CREATE OUTLINE Command
You can also create stored outlines for specific statements by using the CREATE OUTLINE
command.
One advantage of using this command is that you can name stored outlines. In the example,
the CREATE OUTLINE command assigns the category TRAIN to the stored outline.
If you omit a category name when generating outlines, they are placed in the DEFAULT
category.
Instructor Note
Note the comments on the previous page state that it is not a good idea to enable this feature
at the instance level. You usually create stored outlines at the statement or session level only.
Demonstration script: demo10_01.sql

Oracle9i: SQL Tuning Workshop 10-5

 Using Stored Outlines

• Set USE_STORED_OUTLINES to TRUE

or to a category name.
SQL> ALTER SESSION
2 SET USE_STORED_OUTLINES = TRAIN;
SQL> SELECT ...

• You can set CREATE_STORED_OUTLINES and

USE_STORED_OUTLINES at two levels:
– ALTER SYSTEM
– ALTER SESSION

Copyright © Oracle Corporation, 2002. All rights reserved.

Using Stored Outlines

To use a stored outline for a particular SQL statement:
• USE_STORED_OUTLINES must be set to TRUE or a category name:
– If it is set to TRUE, then outlines from the DEFAULT category are used.
– If it is set to a category name, then the outlines from that category are used. If
there is no matching outline in that category, but there is one in the DEFAULT
category, then that outline is used.
• The statement must match the text of the statement in the outline. They are compared
using the same method for comparing cursors in the shared pool. This means that hints
in the outline must be used in the statement text to cause a match. Values in bind
variables do not need to match.
Instructor Note
Demonstration script: demo10_02.sql
Instructor Note (for page 10-7)
The USER_OUTLINE_HINTS view also shows a STAGE column, indicating the stage
(during compilation) at which the hint was applied.
Demonstration script: demo10_03.sql

Oracle9i: SQL Tuning Workshop 10-6

 Data Dictionary Information

SQL> SELECT NAME, CATEGORY, USED

2 , SQL_TEXT
3 FROM USER_OUTLINES;

SQL> SELECT NODE, HINT

2 FROM USER_OUTLINE_HINTS
3 WHERE NAME = ...;

SQL> SELECT SQL_TEXT, OUTLINE_CATEGORY

2 FROM V$SQL
3 WHERE ...;

Copyright © Oracle Corporation, 2002. All rights reserved.

Data Dictionary Information

Information about stored outlines can be obtained from the USER_OUTLINES and
USER_OUTLINE_HINTS data dictionary views. The column descriptions of these two
views are:
USER_OUTLINES
NAME Name of the outline
CATEGORY User-defined name of the category to which this outline belongs
USED Outline usage indication (USED, UNUSED, or UNDEFINED)
TIMESTAMP Timestamp of outline creation
VERSION Oracle version that created the outline
SQL_TEXT SQL text of the query, including any hints that were part of the original
statement

USER_OUTLINE_HINTS
NAME Name of the outline
NODE ID of the query or subquery to which the hint applies (The top-level query is
labeled 1; subqueries start with 2.)
JOIN_POS Position of the table in the join order (The value is 0 for all hints except access
method hints, which identify a table to which the hint and the join position apply.)
HINT Text of the hint

Oracle9i: SQL Tuning Workshop 10-7

 Execution Plan Logic

In Same
shared outline Execute
y y
pool? category? outline plan

n n

Query DD for
matching outline

Regular Integrate outline

n Found? y
execute and generate plan

Copyright © Oracle Corporation, 2002. All rights reserved.

Execution Plan Logic

To determine a SQL statement’s execution plan, Oracle9i uses the following logic:
• The statement is compared to statements in the shared pool for matching text and
outline category.
• If no matching statement is found, the data dictionary is queried for a matching outline.
• If a matching outline is found, Oracle9i integrates the outline into the statement and
creates the execution plan.
• If no outline is found, the statement is executed using normal (nonoutline) methods.
If an outline specifies the use of an object that cannot be used, the statement will not use the
hint. For example, it references an index that no longer exists. To verify that a stored outline
is being used, the explain plan for a statement must be compared when running with and
without the USE_STORED_OUTLINES set.

Oracle9i: SQL Tuning Workshop 10-8

 Maintaining Stored Outlines

• Use OUTLN_PKG to:

– Drop unused outlines
– Drop categories of outlines
– Rename a category
• Use ALTER OUTLINE to:
– Rename an outline
– Rebuild an outline
– Change the category of an outline
• Outlines are stored in schema OUTLN

Copyright © Oracle Corporation, 2002. All rights reserved.

Maintaining Stored Outlines

Use procedures in the OUTLN_PKG package to manage stored outlines and their categories:

Procedure Description
DROP_UNUSED Drops outlines that have not been used since they were
created
DROP_BY_CAT Drops outlines assigned to the specified category name
UPDATE_BY_CAT Reassigns outlines from one category to another
You can alter individual outlines by using the ALTER OUTLINE command and drop them
by using the DROP OUTLINE command.
You can export and import plans by exporting the schema OUTLN, where all outlines are
stored. Outlines can be queried from tables in the OUTLN schema:
• OL$ contains the outline name, category, create timestamp, and the text of the
statement.
• OL$HINTS contains the hints for the outlines in OL$.
Also, there are equivalent data dictionary views: DBA_OUTLINES and
DBA_OUTLINE_HINTS.
Note: Because the user OUTLN is automatically created at database creation with password
OUTLN, this password should be changed for security reasons.
Oracle9i: SQL Tuning Workshop 10-9
 Maintaining Stored Outlines

SQL> begin
2 outln_pkg.DROP_UNUSED;
3 outln_pkg.UPDATE_BY_CAT
4 (’DEFAULT’,’TRAIN’);
5 outln_pkg.DROP_BY_CAT(’TRAIN’);
6 end;

Copyright © Oracle Corporation, 2002. All rights reserved.

Maintaining Stored Outlines (continued)

The DROP_UNUSED procedure does not accept any arguments and automatically drops all
outlines that have never been used since they were created.
The UPDATE_BY_CAT procedure example moves all outlines from the DEFAULT category
to the TRAIN category. For example, this procedure is useful to move a set of outlines from
an experimental (test environment) category into a production category.
The DROP_BY_CAT procedure example purges all outlines that belong to the TRAIN
category in a single call.
Export and Import Stored Outlines
By exporting and importing the OUTLN schema, you can distribute outlines to other
databases, such as to all departments that use a certain application. This approach ensures
execution plan consistency at all sites.
Instructor Note
Demonstration script: demo10_04.sql, then run demo10_03.sql again.

Oracle9i: SQL Tuning Workshop 10-10

 Outline Editing Overview

• Stored outline editing is new in Oracle9i .

• Tune execution plans without having to change the
application.
• This is possible by editing the content of the saved
plan.

Copyright © Oracle Corporation, 2002. All rights reserved.

Outline Editing Overview

In Oracle9i, outline editing extends the usefulness of stored outlines by introducing a user-
editing interface, which enables users and third-party vendors to tune their execution plans
by editing the stored outlines that are used to influence the optimizer. This can be useful
when using the same application in different environments (small versus big databases).
This feature benefits both application developers and customer support personnel. Although
the optimizer usually chooses optimal plans for queries, there are times when the user knows
something about the execution environment that is inconsistent with the heuristics the
optimizer follows. Sometimes an execution plan is acceptable in one environment but not in
another. By editing the outline directly, the user can tune the query without having to change
the application.
The application developer might generate outlines in a staging area and notice that some plan
did not take advantage of an index that could improve performance. It might be easier to edit
the outline to use the index rather than searching through the application code and tuning the
SQL until it eventually yields the desired result.
Customer support might be at a customer site investigating a problem query. By creating an
outline and then editing it, it is likely that with some simple edits, such as changing join
order, the problem can be solved quickly. In this way, the customer’s problem is solved
immediately without having to go through the process of debugging and updating the
application itself.

Oracle9i: SQL Tuning Workshop 10-11

 Outline Editing Overview

• The outline is cloned in a staging area

• The outline is edited in the user’s session
• When satisfied with the result, the editor can
publicize the result to the user community.

Copyright © Oracle Corporation, 2002. All rights reserved.

Outline Editing Overview (continued)

For the customer whose environment has unique characteristics that might cause an outline to
yield a less than optimal execution plan, the ability to make minor adjustments to the outline
enhances the ability to support specific customer needs. In this sense, stored outlines are
made more adaptive as users can make finely tuned adjustments to the saved plan.
Stored outline metadata is maintained in the OUTLN schema and maintained by the server.
You are advised not to update these tables directly in the same way you are advised not to
update system tables. Therefore, Oracle must provide users with a way to safely edit an
outline without compromising the integrity of the outline for the rest of the user community.
To accomplish this, Oracle proposes that the outline be cloned into the user’s schema at the
onset of the outline editing session. All subsequent editing operations are performed on that
clone until the user is satisfied with the edits and chooses to publicize them. In this way, any
editing done by the user does not impact the rest of the user community, which continues to
use the public version of the outline until the edits are explicitly saved.

Oracle9i: SQL Tuning Workshop 10-12

 Editable Attributes

• Join order
• Join methods
• Access methods
• Distributed execution plans
• Distribution methods for parallel query execution
• Query rewrite
• View and subquery merging

Copyright © Oracle Corporation, 2002. All rights reserved.

Editable Attributes
Join order: Join order defines the sequence in which tables are joined during query
execution. This includes tables produced by evaluating subqueries and views as well as
tables appearing in the FROM clauses of subqueries and views.
Join methods: Join methods define the methods used to join tables during query execution.
Examples are nested loops join and sort/merge join.
Access methods: Access methods define the methods used to retrieve table data from the
database. Examples are indexed access and full table scan.
Distributed execution plans: Distributed queries have execution plans that are generated for
each site at which some portion of the query is executed. The execution plan for the local site
at which the query is submitted can be controlled by plan stability and equivalent plans must
be produced at that site. In addition, driving site selection can be controlled centrally even
though it might normally change when certain schema changes are made.
Distribution methods: For parallel query execution, distribution methods define how the
inputs to execution nodes are partitioned.
View and subquery merging and summary rewrite: View and subquery merging and
summary rewrite is meant to include all transformations in which objects or operations that
occur in one subquery of the original SQL statement are caused to migrate to a different
subquery for execution. Summary rewrite can also cause one set of objects or operations to
be replaced by another.
Oracle9i: SQL Tuning Workshop 10-13
 Outline Cloning

• Public outlines:
– Default setting when creating outlines
– Stored in the OUTLN schema
– Used when USE_STORED_OUTLINES is set to TRUE or a
category
• Private outlines:
– Stored in the user’s schema for the duration of the
session
– Can be edited
– Used when USE_PRIVATE_OUTLINES is set to TRUE or
a category
– Changes can be saved as public outlines

Copyright © Oracle Corporation, 2002. All rights reserved.

Outline Cloning
Public Outlines
In Oracle9i, all outlines are public objects indirectly available to all users on the system for
whom the USE_STORED_OUTLINES configuration parameter setting applies. Outline data
resides in the OUTLN schema that can be thought of as an extension to the SYS schema, in
the sense that it is maintained by the system only. You are discouraged from manipulating
this data directly to avoid security and integrity issues associated with outline data. Outlines
will continue to be public by default and only public outlines are generally available to the
user community.
Private Outlines
In Oracle9i, the notion of a private outline to aid in outline editing is introduced. A private
outline is an outline seen only in the current session and whose data resides in the current
parsing schema. By storing the outline data for a private outline directly in the user’s schema,
users are given the opportunity to manipulate the outline data directly through DML in
whichever way they choose. Any changes made to such an outline are not seen by any other
session on the system and applying a private outline to the compilation of a statement can be
done only in the current session through a new session parameter. Only when a user
explicitly chooses to save edits back to the public area do the rest of the users see them.
An outline clone is a private outline that has been created by copying data from an existing
outline.
Oracle9i: SQL Tuning Workshop 10-14
 Outline: Administration and Security

• Privileges required for cloning outlines:

– SELECT_CATALOG_ROLE
– CREATE ANY OUTLINE
• DBMS_OUTLN_EDIT.CREATE_EDIT_TABLES
– Creates required temporary tables in user’s schema
for cloning and editing outlines
– Requires EXECUTE privilege on DBMS_OUTLN_EDIT

Copyright © Oracle Corporation, 2002. All rights reserved.

Outline: Administration and Security

SELECT_CATALOG_ROLE
This role is required for the CREATE OUTLINE FROM command unless the issuer of the
command is also the owner of the outline. Any CREATE OUTLINE statement requires the
CREATE ANY OUTLINE privilege. Specification of the FROM clause will require the
additional SELECT_CATALOG_ROLE role because such a command exposes SQL text to
different users who might otherwise not be privileged to read the text.
DBMS_OUTLN_EDIT.CREATE_EDIT_TABLES
A supporting command procedure that creates the metadata tables in the invoker’s schema.
This procedure is callable by anyone with the EXECUTE privilege on the
DBMS_OUTLN_EDIT package. Refer to the Supplied PL/SQL Packages Reference for more
information. Note also that the DBMS_OUTLN package is synonymous with OUTLN_PKG.
Private outlines are private to the session editing them. To accommodate the possibility of
multiple users editing the same outline at the same time (not a good practice, but possible)
the data is partitioned by session. This is really scratch data not intended for long-term use
and this removes the need for users to worry about cleaning up when they are done because
the temporary data goes away when the session is closed.

Oracle9i: SQL Tuning Workshop 10-15

 Outline: Administration and Security

• V$SQL
• OUTLINE_SID column added
• Identifies the session ID from which the outline was
retrieved

Copyright © Oracle Corporation, 2002. All rights reserved.

Outline: Administration and Security (continued)

V$SQL
A column is added to the V$SQL fixed view to help users distinguish whether a shared cursor
was compiled while using a private outline or a public outline. OUTLINE_SID is the name
of this new column and it identifies the session ID from which the outline was retrieved. The
default is 0 which implies a lookup in the OUTLN schema.

Oracle9i: SQL Tuning Workshop 10-16

 Configuration Parameters

USE_PRIVATE_OUTLINES is a session parameter that

controls the use of private outlines instead of public
outlines.
ALTER SESSION SET USE_PRIVATE_OUTLINES =
TRUE | FALSE | category_name ;
• TRUE enables the use of private outlines in the
DEFAULT category.
• FALSE disables use of private outlines.
• category_name enables use of private outlines in
the named category.

Copyright © Oracle Corporation, 2002. All rights reserved.

Configuration Parameters
The USE_PRIVATE_OUTLINES session parameter is added to control the use of private
outlines instead of public outlines. When an outlined SQL command is issued, this parameter
causes outline retrieval to come from the session private area rather than the public area that
is usually consulted as per the setting of USE_STORED_OUTLINES. If no outline exists in
the session private area, no outline is used for the compilation of the command.
You can specify a value for this session parameter by using the following syntax:
ALTER SESSION SET USE_PRIVATE_OUTLINES =
TRUE | FALSE | category_name ;
Where :
• TRUE enables use of private outlines and defaults to the DEFAULT category
• FALSE disables use of private outlines
• category_name enables use of private outlines in the named category
When a user begins an outline editing session, the parameter should be set to the category to
which the outline being edited belongs. This enables the feedback mechanism in that it
allows the private outline to be applied to the compilation process.
Upon completion of outline editing, this parameter should be set to FALSE to restore the
session to normal outline lookup as dictated through the USE_STORED_OUTLINES
parameter.

Oracle9i: SQL Tuning Workshop 10-17

 Create Outline Syntax

The CREATE OUTLINE command has been extended to

clone outlines:
CREATE [OR REPLACE]
[PUBLIC | PRIVATE] OUTLINE [outline_name]
[FROM [PUBLIC | PRIVATE] source_outline_name]
[FOR CATEGORY category_name] [ON statement]

Copyright © Oracle Corporation, 2002. All rights reserved.

Create Outline Syntax Changes

The new syntax elements in the above slide are:
PUBLIC: The outline is to be created for use by PUBLIC. This is the default because outline
creation is intended for systemwide use.
PRIVATE: The outline is to be created for private use by the current session only and its data
is stored in the current parsing schema. When specified, the prerequisite outline tables and
indices must exist in the local schema.
FROM: This construct provides a way to create an outline by copying an existing one.
source_outline_name: This is the name of the outline being cloned. By default, it is
found in the public area, but if preceded by the PRIVATE keyword it will be found in the
local schema.
The addition of the PRIVATE and FROM keywords enable outline cloning. When you want
to edit an outline, you do so on a private copy which is created by specifying the PRIVATE
keyword. In the FROM clause, the source outline to be edited is named and is found in the
public area unless preceded by the PRIVATE keyword, in which case the you would be
copying a private version of the named outline.
When specifying the FROM clause, existing semantics apply to the outline name and
category, so if unspecified, an outline name is generated under the DEFAULT category.
When a PRIVATE outline is being created, if the prerequisite outline tables that hold the
outline data do not exist in the local schema, an error is returned.

Oracle9i: SQL Tuning Workshop 10-18

 Outline Cloning Examples

• Clone a public outline to the session private area:

CREATE PRIVATE OUTLINE private_outline FROM
public_outline;

• Clone a private copy to another private copy:

CREATE PRIVATE OUTLINE private_outline2 FROM
PRIVATE private_outline1 FOR CATEGORY cat2;

• Copy a private outline back to the public area:

CREATE OR REPLACE OUTLINE public_outline FROM
PRIVATE private_outline

Copyright © Oracle Corporation, 2002. All rights reserved.

Outline Cloning Examples

CREATE PRIVATE OUTLINE private_outline
FROM public_outline;
This example shows how to clone a public outline to the session private area for editing
purposes. In this case, PUBLIC_OUTLINE is cloned into the private area and given the
name PRIVATE_OUTLINE.
CREATE PRIVATE OUTLINE private_outline2
FROM PRIVATE private_outline1 FOR CATEGORY cat2;
This example shows how to clone a private copy of an outline to another private copy. In this
case, PRIVATE_OUTLINE2 is being created in the CAT2 category from the
PRIVATE_OUTLINE1 outline, which is local to the current session. This might be done if
the user wants to save several versions of the outline. Specification of the category is
necessary here to avoid namespace collision.
CREATE OR REPLACE OUTLINE public_outline
FROM PRIVATE private_outline;
This example shows how to copy a private outline back to the public area for general use.
This is a way to save local edits back to the master copy of the outline. This operation is also
called publicize.

Oracle9i: SQL Tuning Workshop 10-19

 Outline Cloning Examples

• Create a private outline independently:

CREATE PRIVATE OUTLINE private_outline
ON select count(*) from customers;

• Copy one public outline to another:

CREATE OR REPLACE OUTLINE public_outline2
FROM public_outline1 FOR CATEGORY cat2;

Copyright © Oracle Corporation, 2002. All rights reserved.

Outline Cloning Examples (continued)

CREATE PRIVATE OUTLINE private_outline

ON select count(*)
from customers;
This example demonstrates how you might create a private outline independently without
copying from a preexisting outline. PRIVATE_OUTLINE is created in the session private
area. This is useful for users who want to experiment with an outline before creating it for
public consumption.
CREATE OR REPLACE OUTLINE public_outline2
FROM public_outline1 FOR CATEGORY cat2;
This example shows how to copy one public outline to another. Here the
PUBLIC_OUTLINE1 public outline is copied to PUBLIC_OUTLINE2 and placed in the
CAT2 category. Specification of the category is necessary here to avoid namespace collision.

Oracle9i: SQL Tuning Workshop 10-20

 Outline Cloning Examples

Replace an outline with itself:

CREATE OR REPLACE PRIVATE OUTLINE priv_outl1
FROM PRIVATE priv_outl1;

dbms_outln_edit.refresh_private_outline(
’private_outline1’)

Copyright © Oracle Corporation, 2002. All rights reserved.

Outline Cloning Examples (continued)

CREATE OR REPLACE PRIVATE OUTLINE private_outline1

FROM PRIVATE private_outline1;
This example shows how to replace an outline with itself. This might seem like a useless
operation, but it actually serves the dual purpose of invalidating dependent shared cursors
and bringing the shared pool view of the object in line with what is on disk. The feedback
mechanism relies on this aspect of the command. You can also use
DBMS_OUTLN_EDIT.REFRESH_PRIVATE_OUTLINE or ALTER SYSTEM FLUSH
SHARED_POOL to accomplish this.

Oracle9i: SQL Tuning Workshop 10-21

 Summary

In this lesson, you should have learned how to:

• Use stored outlines to ensure execution plan
consistency
• Create outlines for a session or a single statement
• Organize outlines in categories
• Enable or disable using outlines or categories of
outlines
• Maintain outlines with the OUTLN_PKG package or
the ALTER OUTLINE command

Copyright © Oracle Corporation, 2002. All rights reserved.

Summary
This lesson introduced you to the use of stored outlines and its influence on optimizer plan
stability, and use of the package OUTLN_PKG on stored outlines.
Oracle9i provides you with a means of stabilizing execution plans across Oracle releases,
database changes, or other factors that can cause an execution plan to change. You can create
a stored outline containing a set of hints used by the optimizer to create an execution plan.
Stored outlines rely partially on hints that the optimizer uses to achieve stable execution
plans. These plans are maintained through many types of database and instance changes. You
can use stored outlines to ensure that all your customers access the same execution plans.
You can create outlines for a session, or you can create them for specific SQL statements.
To use stored outlines, the USE_STORED_OUTLINES option must be set to TRUE or a
category name. Use procedures in the OUTLN_PKG package to manage stored outlines and
their categories. There are data dictionary views that store information on the outlines.
Note: For more details about plan stability and using stored outlines, refer to Oracle9i
Tuning, “Optimizer Modes, Plan Stability, and Hints.”

Oracle9i: SQL Tuning Workshop 10-22

 Practice Overview

This practice covers the following topics:

• Creating a stored outline
• Altering your session to use a stored outline
• Using a stored outline
• Querying the data dictionary
• Removing a stored outline

Copyright © Oracle Corporation, 2002. All rights reserved.

Practice Overview
In this practice, you create, use, and then remove a stored outline. You must alter your
session in order for the outline to be used. Use the USER_OUTLINES and
USER_OUTLINE_HINTS data dictionary views to view the status of your outline.

Oracle9i: SQL Tuning Workshop 10-23

Practice 10
1. Turn SQL*Plus AUTOTRACE off.
Make sure that the SUPPLIER_ID column of the PRODUCTS table is not indexed by
running the dai.sql script from the demo labs directory, then create a stored outline for
the following SQL statement:
SQL> @dai.sql
SQL> select p.prod_id, p.prod_name, p.prod_min_price
2 from products p
3 where p. supplier_id = 260;
2. Check the USER_OUTLINES data dictionary view to find the outline name and category
name for the outline you just created; also, query the USED column to see that the outline
is not yet used.
3. Check the USER_OUTLINE_HINTS data dictionary view to see which hints are stored
with your outline.
4. Issue an ALTER SESSION command to enable your session using the stored outline.
5. Execute the SQL statement in step 1, create an index on the PROD_STATUS column in
the PRODUCTS table, and execute the SQL statement again to see whether the execution
plan has changed.
6. Check the USER_OUTLINES data dictionary view again to see whether the outline has
been used.
7. Drop the outline, and issue the SQL statement again to see that the index will now be
used.

Oracle9i: SQL Tuning Workshop 10-24

 Advanced Indexes

Copyright © Oracle Corporation, 2002. All rights reserved.

Schedule: Timing Topic

45 minutes Lecture
45 minutes Total
Instructor Note
• Note that index hints are revisited, adding the INDEX_COMBINE hint to the hints covered
in the “Influencing the Optimizer” lesson. Star transformation is covered in this lesson
because it is a feature based on bitmapped indexes. Make a clear distinction between star
transformation (as discussed in this lesson) and star joins (as discussed in the “Sorting and
Joining” lesson). Finally, function-based indexes are introduced.
• To speed the demonstrations against the SALES table, create a copy of it using a subset of
rows from the original SALES table.
To create the copy of the SALES table, execute the command (this command is also
provided in the demo11_01.sql) script:
CREATE TABLE copy_sales AS
SELECT *
FROM sales
WHERE rownum < 200000;
 Objectives

After completing this lesson, you should be able to do

the following:
• Create bitmapped indexes
• Identify bitmapped index operations
• Specify bitmapped index hints
• Use star transformations
• Create function-based indexes
• View data dictionary information

Copyright © Oracle Corporation, 2002. All rights reserved.

Objectives
After completing this lesson, you should be able to:
• Create bitmapped indexes, identify bitmapped index operations, and use bitmapped index
hints.
• Describe, identify and use star transformations.
• Create and use function-based indexes.
• View data in the data dictionary on indexes.

Oracle9i: SQL Tuning Workshop 11-2

 Bitmapped Indexes

• Compared with regular B*-tree indexes, bitmapped

indexes are faster and use less space for low
cardinality columns.
• Each bitmap index comprises storage pieces called
bitmaps.
• Each bitmap contains information about a particular
value for each of the indexed columns.
• Bitmaps are compressed and stored in a B*-tree
structure.

Copyright © Oracle Corporation, 2002. All rights reserved.

Bitmapped Indexes
Bitmap indexing provides both substantial performance improvements and space savings over
usual (B*-tree) indexes when there are few distinct values in the indexed columns.
If the number of distinct keys in a column is less than or equal to 1% of the total number of
rows for a table, then the column has low cardinality. For example, in a table with 10,000 rows,
if there are fewer than 100 different values in a particular column, then that column has low
cardinality, and a bitmap index may provide the best performance for queries based on that
column.
The maximum number of columns in a single bitmapped index is 30.
The Oracle9i Server compresses bitmap storage, making bitmapped indexes very
storageefficient. The bitmaps are stored in a B*-tree structure internally to maximize access
performance.
Note: Rule-based optimization does not consider using bitmapped indexes.

Oracle9i: SQL Tuning Workshop 11-3

 Bitmapped Index Structure

Each position in a bitmap stores information about a

particular row.
0 Row #1

1 Row #2

1 Row #3
0

1
1
0

Copyright © Oracle Corporation, 2002. All rights reserved.

Bitmapped Index Structure

Any type of index is designed to help find rows that contain certain key values. A regular (B*-
tree) index does this by pairing the ROWID for a row with its key value, then sorts those pairs
into a B*-tree structure. The ROWID serves as a pointer to a row with that value.
A bitmapped index has a very different structure; no ROWIDs are stored. Because each
different key value has its own bitmap, you benefit the most from bitmapped indexes when you
only have a few distinct index values.
Each bitmap header stores start and end ROWIDs. Based on these values, the Oracle9i Server
uses an internal algorithm to map bitmaps onto ROWIDs.
Each position in a bitmap maps to a potential row in the table. The contents of that position in
the bitmap for a particular value indicate whether that row has that value in the bitmapped
columns. The value stored is a 1 if the row values match the bitmap condition and 0 otherwise.

Oracle9i: SQL Tuning Workshop 11-4

 Creating Bitmapped Indexes
SQL> CREATE BITMAP INDEX prod_supplier_id
2 ON products (supplier_id);
Supplier Supplier Supplier Supplier
Row ID ID ID ID
value 1 2 3 4 ...
’1’ 1 0 0 0
’2’ 0 1 0 0
’3’ 0 0 1 0
’4’ 0 0 0 1
’2’ 0 1 0 0
’3’ 0 0 1 0
. . . . .
. . . . .

Copyright © Oracle Corporation, 2002. All rights reserved.

Creating Bitmapped Indexes

To create a bitmapped index, use the CREATE INDEX command and add the keyword
BITMAP. In the above example, a bitmap index is being created on the SUPPLIER_ID column
of the PRODUCTS table.
The Oracle9i Server then creates a series of bitmaps, each one corresponding to a particular
value of the columns involved. In the example, if the SUPPLIER_ID column has 289 distinct
values, then 289 bitmaps are built. If another SUPPLIER_ID value were later added, a new
bitmap for that value is created.
In the bitmap for SUPPLIER_ID ’1’, the first position contains a one, showing that the first
row has that value. In the bitmap for SUPPLIER_ID ’2’, the first position contains a zero
because the first row does not have that value. Each row maps to a position in the bitmaps, and
the one matching its conditions will contain the digit one (1) in that position.
If a bitmap index involves more than one column, then there will be a bitmap for every possible
combination.
Note: Bitmapped indexes cannot be declared as unique (B*-tree indexes are more appropriate
in such cases anyway), and neither can the NOSORT option be used when creating bitmapped
indexes.

Oracle9i: SQL Tuning Workshop 11-5

Creating Bitmapped Indexes
If the table is partitioned, you must use the LOCAL option to create the index:
CREATE BITMAP INDEX index_name
ON table_name(column_name)
LOCAL [NOLOGGING|LOGGING COMPUTE STATISTICS];

Option Description
LOCAL Must be a locally partitioned bitmap index
LOGGING|NOLOGGING Determines whether the creation of the index will be
logged (LOGGING) or not logged (NOLOGGING) in
the redo log file
COMPUTE STATISTICS Collects statistics at relatively little cost during the
creation of an index

Oracle9i: SQL Tuning Workshop 11-6

 Using Bitmapped Indexes for Queries

Supplier
ID
SQL> SELECT * 3
2 FROM products
3 WHERE supplier_id = 3; 0
0

Rows 1
returned 0
0
1
.
.

Copyright © Oracle Corporation, 2002. All rights reserved.

Example of Querying on a Bitmapped Index

A query that includes a WHERE condition involving the bitmapped column can use the
bitmapped index. In the above example, a search is being conducted for SUPPLIER_ID of 3.
This query may scan the bitmap for SUPPLIER_ID ’3’ for positions with a 1 in them.
Positions with a 1 will have their corresponding rows returned for the query. In some cases,
such as a query counting the number of rows with SUPPLIER_ID ’3’, the query might
simply use the bitmap itself and count the number of 1s, not needing the actual rows.
Instructor Note
You can discuss the benefit of a bitmapped index in this single predicate statement. Note that
this example is meant to illustrate bitmapped indexes only. Usually, bitmapped indexes are
efficient with more complex WHERE clauses using AND/OR/NOT.

Oracle9i: SQL Tuning Workshop 11-7

 Combining Bitmapped Indexes

Due to fast bit-and, bit-minus, bit-or operations,

bitmapped indexes are efficient:
• When using IN (value_list)
• When predicates are combined with AND/OR

A B A and B A or B not A
1 1 1 1 0
0 1 0 1 1
1 0 0 1 0
0 0 0 0 1
0 1 0 1 1

Copyright © Oracle Corporation, 2002. All rights reserved.

Combining Bitmapped Indexes

You can use bitmapped indexes efficiently when a query combines several possible values for a
column or when two separately indexed columns are used.
Query Combining Multiple Values for a Column
In some cases, a WHERE clause might allow several different values for a column:
SQL> SELECT * FROM products
2 WHERE supplier_id IN (1, 3);
In this case, the bitmaps for both 1 and 3 would be used. For each position, if either of the
bitmaps has a 1 in that position, then the row is returned. A bit-or operation is performed first,
followed by a translation into ROWIDs.
Query Combining Two Different Bitmapped Indexes
In some cases, a WHERE clause might reference several separately indexed columns.
SQL> SELECT * FROM products
2 WHERE supplier_id = 3
3 AND prod_pack_size = ’card box’;
If both the SUPPLIER_ID and PROD_PACK_SIZE columns have bitmapped indexes, a bit-
and operation on the two bitmaps would quickly locate the rows that you are looking for. The
more complex the compound WHERE clauses get, the more you benefit from bitmapped
indexing.

Oracle9i: SQL Tuning Workshop 11-8

 When to Use Bitmapped Indexes

• Columns with low cardinality

• Columns that are frequently used in:
– Complex WHERE clause conditions
– Group functions (such as COUNT, SUM)
• Very large tables
• DSS systems with many ad hoc queries and few
concurrent DML changes

Copyright © Oracle Corporation, 2002. All rights reserved.

When to Use Bitmapped Indexes

Use bitmapped indexes in the following circumstances:
• The column has a low cardinality, meaning that there are few possible values for the
column. A slightly higher cardinality would also be appropriate if the column is frequently
used in complex conditions in the WHERE clauses of queries.
• The WHERE clauses of queries contain multiple predicates of these low- or medium-
cardinality columns. Bitmapped indexes are especially helpful for complex ad hoc queries
with lengthy WHERE clauses or aggregate queries (containing SUM, COUNT, or other
aggregate functions).
• The table has many rows. On a table with 1 million rows, even 10,000 distinct possible
values would be acceptable.
• There are frequent, possibly ad hoc, queries on the table.
• The environment is oriented toward data warehousing (decision support systems).
Bitmapped indexes are not ideal for OLTP environments because of their locking behavior. It is
not possible to lock a single bitmap position. The smallest amount of a bitmap that can be
locked is a bitmap segment, which can be up to half a data block in size. Changing the value of
a row results in a bitmap segment becoming locked, in effect blocking changes on a number of
rows. This is a serious disadvantage when there are many UPDATE, INSERT, or DELETE
statements being issued by users. It is not a problem when data is loaded or updated in bulk
actions, as in data warehousing environments.
Oracle9i: SQL Tuning Workshop 11-9
 Advantages of Bitmapped Indexes

When used appropriately, bitmapped indexes provide:

• Reduced response time for many ad hoc queries
• Substantial reduction of space usage compared to
other indexing techniques
• Dramatic performance gains, even on low-end
hardware

Copyright © Oracle Corporation, 2002. All rights reserved.

Advantages of Bitmapped Indexes

When used appropriately, bitmap indexes can provide substantial advantages:
• They provide reduced response time for many ad hoc queries. The cost-based optimizer is
able to consider the use of bitmapped indexes and decide when their use would be the
most efficient way to execute a given query, incorporating parallelism, if appropriate.
• They provide a substantial reduction of space usage compared to regular B*-tree indexes.
If only regular (B*-tree) indexes are used, then the index designer must predict which
columns will be used together and create a multicolumn (concatenated) index for these
columns. This regular index would require a large amount of space and be sorted, making
it useless for many queries that involve only a subset of the columns. The index designer
could also create indexes on the other permutations of these columns; this would use a
great deal of storage space. Bitmapped indexes, however, can be created individually and
then combined efficiently at run time. If the column to be indexed is a unique key column,
then the space required for the bitmapped index will exceed that of the B*-tree index. Use
bitmapped indexes on low- or medium-cardinality columns only.
• You can obtain dramatic performance gains, even on low-end hardware. You do not need
parallel processing capability to achieve performance gains with bitmapped indexes.

Oracle9i: SQL Tuning Workshop 11-10

 Bitmapped Index Guidelines

• Reduce bitmap storage by:

– Declaring columns NOT NULL when possible
– Using fixed-length data types when feasible
– Using the ALTER TABLE … MINIMIZE
RECORDS_PER_BLOCK command
• Improve bitmap performance by increasing the
following parameters:
– CREATE_BITMAP_AREA_SIZE (default 8 MB)
– BITMAP_MERGE_AREA_SIZE (default 1 MB)

Copyright © Oracle Corporation, 2002. All rights reserved.

Bitmapped Index Guidelines

You can do several things to improve performance and storage of bitmapped indexes:
• Declare NOT NULL constraints on all possible columns
• Use fixed length data types
• Use the ALTER TABLE...MINIMIZE RECORDS_PER_BLOCK command
The ALTER TABLE command determines the number of rows that are stored per block, and
limits the number of rows per block accordingly. Issue this statement before creating any
bitmapped indexes.
The CREATE_BITMAP_AREA_SIZE parameter determines the amount of memory allocated
for bitmap creation. The default value is 8 MB, which is enough in most cases. A larger value
can result in larger continuous bitmaps and thus faster query processing. The
SORT_AREA_SIZE parameter also influences bitmap creation performance.
The BITMAP_MERGE_AREA_SIZE parameter determines the amount of memory used to
merge bitmaps retrieved from an index range scan. The default value is 1 MB.
Instructor Note
When you create a bitmapped index, the theoretical maximum number of rows per block is
calculated, based on data dictionary information. When columns are NOT NULL and fixed in
width, the row size is more predictable. This means that the bitmaps do not contain many
unnecessary (zero) bits. Note that ALTER TABLE...MINIMIZE RECORDS_PER_BLOCK
can waste space, because it limits the number of rows stored per data block, regardless of the
block space they occupy.
Oracle9i: SQL Tuning Workshop 11-11
 What Is a Bitmap Join Index?
1.2.3
Sales Customers
CREATE BITMAP INDEX cust_sales_bji
ON sales(c.cust_city)
FROM sales s, customers c
WHERE c.cust_id = s.cust_id;

start end
key bitmap
ROWID ROWID
10.8000.3 <Rognes, 1.2.3, 10.8000.3, 1000100100010010100…>
<Aix-en-Provence, 1.2.3, 10.8000.3, 0001010000100100000…>
<Marseille, 1.2.3, 10.8000.3, 0100000011000001001…>
……………………………………
SELECT SUM(s.amount_sold) Only the index and the
FROM sales s, customers c SALES table are used to
WHERE s.cust_id = evaluate the query. No
c.cust_id join with the CUSTOMERS
AND c.cust_city = ’Marseille’; table is needed.

Copyright © Oracle Corporation, 2002. All rights reserved.

What Is a Bitmap Join Index?

In addition to a bitmap index on a single table, you can create a bitmap join index in Oracle9i.
A bitmap join index is a bitmap index for the join of two or more tables. A bitmap join index is
a space-efficient way of reducing the volume of data that must be joined by performing
restrictions in advance.
As you can see in the above slide, Oracle9i introduces a new CREATE BITMAP INDEX
syntax that you can use to specify a FROM and a WHERE clause. Here, you create a new bitmap
join index named CUST_SALES_BJI on the SALES table. The key of this index is the
CUST_CITY column of the CUSTOMERS table.
This example assumes that there is an enforced primary key-foreign key relationship between
SALES and CUSTOMERS to ensure that what is stored in the bitmap reflects the reality of the
data in the tables. The CUST_ID column is the primary key of CUSTOMERS and also the
foreign key from the SALES table to the CUSTOMERS table. The FROM and WHERE clause in
the CREATE statement allow Oracle9i to make the link between the two tables. They represent
the natural join condition between the two tables.

Oracle9i: SQL Tuning Workshop 11-12

What Is a Bitmap Join Index? (continued)
The middle part of the previous graphic shows you a theoretical implementation of this
bitmap join index. Each entry or key in the index represents a possible city found in the
CUSTOMERS table. A bitmap is then associated to one particular key. The meaning of the
bitmap is quite obvious because it is the same representation as for traditional bitmap
indexes. Each bit in a bitmap corresponds to one row in the SALES table. If you take the
first key above (Rognes), you can see that the first row in the SALES table corresponds to a
product sold to a Rognes customer, whereas the second bit is not a product sold to a Rognes
customer.
The interesting aspect of this structure becomes clear with the last part of the slide. Indeed,
when the user tries to find what the total cost of all sales for the Aix-en-Provence customers
is, Oracle9i can use the above bitmap join index and the SALES table to answer the
question. In this case, there is no need to compute the join between the two tables
explicitly. Using the bitmap join index in this case is much faster than computing the join at
query time.

Oracle9i: SQL Tuning Workshop 11-13

 Bitmap Join Index:
Advantages and Disadvantages

• Advantages:
– Good performance for join queries and space efficient
– Is especially useful for large dimension tables in star
schemas
• Disadvantages:
– More indexes are required: Up to one index per
dimension table column rather than one index per
dimension table.
– Maintenance costs are higher: Building or refreshing a
bitmap join index requires a join.

Copyright © Oracle Corporation, 2002. All rights reserved.

Advantages and Disadvantages

A bitmap join index is an index on one table that involves columns of one or more different
tables through a join.
The volume of data that must be joined can be reduced if bitmap join indexes are used as joins
that have already been precalculated. In addition, bitmap join indexes that can contain multiple
dimension tables can eliminate bitwise operations that are necessary in the star transformation’s
use of bitmap indexes.
An alternative to a bitmap join index is a materialized join view, which is the materialization of
a join in a table. Compared to a materialized join view, a bitmap join index is much more space
efficient because it compresses ROWIDs of the fact tables.
Also, queries using bitmap join indexes can be sped up by means of bitwise operations.
On the other hand, you may need to create more bitmap join indexes on the fact table to satisfy
the maximum number of different queries. This mean that you may have to create one bitmap
join index for each column of the corresponding dimension tables. Of course, the implication of
having many indexes on one table is that maintenance costs are higher, especially when the fact
table is updated.
Note: Dimensions are covered in detail in lesson 12

Oracle9i: SQL Tuning Workshop 11-14

Advantages and Disadvantages (Continued)
Because of the necessity of storing the results of join results, bitmap join indexes have the
following restrictions:
• Parallel DML is currently supported only on the fact table. Parallel DML on one of
the participating dimension tables will mark the index as unusable.
• Only one table can be updated concurrently by different transactions when using the
bitmap join index.
• No table can appear twice in the join.
• You cannot create a bitmap join index on an index-organized table or a temporary
table.
• The columns in the index must all be columns of the dimension tables.
• The dimension table join columns must be either primary key columns or have unique
constraints.
• If a dimension table has composite primary key, each column in the primary key must
be part of the join.
• Bitmap join indexes cannot be built or rebuilt online.

Oracle9i: SQL Tuning Workshop 11-15

 Indexes and Row-Access Methods

• Access methods discussed in an earlier lesson:

– B*-tree index access
– Table access by ROWID
– B*-tree index merge
– Fast full index scans
• Index hints
• Star transformation

Copyright © Oracle Corporation, 2002. All rights reserved.

Indexes and Row Access Methods

In the “Indexes and Basic Access Methods” lesson, you identified the following basic row
access methods:
• B*-tree index access
• Table access by ROWID
• B*-tree index merge
• Fast full index scan
In the “Influencing the Optimizer” lesson, you learned how to influence the optimizer by
specifying basic access path hints, such as INDEX, NO_INDEX, INDEX_ASC,
INDEX_DESC, AND_EQUAL, and INDEX_FFS.
After completing this section of the chapter, you should know how to influence the cost-based
optimizer by specifying the INDEX_COMBINE hint for bitmapped indexes, and investigate
how star queries can be executed more efficiently by using the star transformation technique.

Oracle9i: SQL Tuning Workshop 11-16

 Index Hints

INDEX Scans an index in ascending order

INDEX_ASC Scans an index in ascending order

INDEX_DESC Scans an index in descending order

AND_EQUAL Merges single-column indexes

INDEX_COMBINE Uses bitmapped indexes

INDEX_FFS Performs a fast full index scan

NO_INDEX Disallow using a set of indexes

Copyright © Oracle Corporation, 2002. All rights reserved.

Index Hints
Most of the following index hints are covered in the “Influencing the Optimizer” lesson:

INDEX(t [idx]...) Chooses an (ascending) index scan for the specified table
INDEX_ASC(t [idx]...) Chooses an ascending index scan for the specified table
INDEX_DESC(t [idx]...) Chooses a descending index scan for the specified table
AND_EQUAL Chooses an execution plan that uses an access path,
(t idx idx [idx]…) merging the scans on several single-column
indexes (minimum 2, maximum 5)
INDEX_COMBINE(t [idx]...) Uses a Boolean combination of bitmapped indexes
INDEX_FFS(t [idx]...) Causes a fast full index scan to be performed
NO_INDEX (t [idx]…) Explicitly disallows a set of indexes for the specified table

Oracle9i: SQL Tuning Workshop 11-17

 INDEX_COMBINE Hint Example

SQL> SELECT --+INDEX_COMBINE(SALES)

2 *
3 FROM SALES
4 WHERE (promo_id = 9999 AND
5 cust_id <> 50)
6 OR prod_id BETWEEN 1000 AND 10;

Copyright © Oracle Corporation, 2002. All rights reserved.

The INDEX_COMBINE Hint

The INDEX_COMBINE hint is designed for bitmapped index operations. Remember the
following:
• If certain indexes are given as arguments for the hint, the optimizer tries to use some
combination of those particular bitmap indexes.
• If no indexes are named in the hint, all indexes are considered hinted.
• The optimizer always tries to use hinted indexes, whether or not it considers them cost
effective.
INDEX_COMBINE Hint Example
Suppose all three columns referenced in the WHERE predicate of the statement above
(PROMO_ID, CUST_ID, and PROD_ID) have a bitmapped index. When you enable
AUTOTRACE, the execution plan of the statement above might look as shown in the slide on the
next page.
Instructor Note
Demonstration script: demo11_01.sql
Make sure that you create bitmapped indexes on the PROMO_ID, CUST_ID, and PROD_ID
columns of the SALES table first. You can run the cbi.sql script to create the bitmapped
indexes. Note that the execution plan is displayed in the next slide.
Oracle9i: SQL Tuning Workshop 11-18
 INDEX_COMBINE Hint Example

Execution Plan
--------------------------------------------------------
0 SELECT STATEMENT Optimizer=CHOOSE (Cost=189
Card=240 Bytes=18480)
1 0 TABLE ACCESS (BY INDEX ROWID) OF
’COPY_SALES’ (Cost=189 Card=240 Bytes=18480)
2 1 BITMAP CONVERSION (TO ROWIDS)
3 2 BITMAP OR
4 3 BITMAP MINUS
5 4 BITMAP INDEX (SINGLE VALUE) OF
’CS_PROMO_ID_BIX’
6 4 BITMAP INDEX (SINGLE VALUE) OF
’CS_CUST_ID_BIX’
7 3 BITMAP MERGE
8 7 BITMAP INDEX (RANGE SCAN) OF
’CS_PROD_ID_BIX’

Copyright © Oracle Corporation, 2002. All rights reserved.

INDEX_COMBINE Hint Example (continued)

In the example, the following bitmap row sources are used:

Bitmap Row Source Description

BITMAP CONVERSION Converts bitmaps into ROWIDs to access a table
TO ROWIDS: COUNT: Returns the number of entries if the actual values are
not needed
BITMAP OR Computes the bit-wise OR of two bitmaps
BITMAP MINUS Subtracts the bits of one bitmap from another (This operator is
used for negated predicates.)
BITMAP INDEX SINGLE VALUE: Looks up the bitmap for a single key value
RANGE SCAN: Retrieves bitmaps for a value range
BITMAP MERGE Merges several bitmaps resulting from a range scan into one
(using a bit-wise AND operator)

Oracle9i: SQL Tuning Workshop 11-19

 Star Transformation

Dimension
tables PRODUCTS CUSTOMERS

SALES
Facts
table

CHANNELS PROMOTIONS TIMES

Copyright © Oracle Corporation, 2002. All rights reserved.

Star Transformation
One type of data warehouse design centers around what is known as a star schema, which is
characterized by one or more large fact tables that contain the primary information in the data
warehouse and a number of much smaller dimension tables (or lookup tables), each of which
contains information about the entries for a particular attribute in the fact table.
A star query is a join between a fact table and a number of lookup tables. Each lookup table is
joined to the fact table using a primary-key to foreign-key join, but the lookup tables are not
joined to each other.
Star queries are quite challenging to tune. There are two basic approaches to this problem:
• Do as smart a job as possible. The “Sorting and Joining” lesson discussed how to use star
joins to optimize star queries.
• Transform the statement into an easier one. This lesson discusses this approach, using
bitmapped indexes. This approach is also known as star transformation.

Oracle9i: SQL Tuning Workshop 11-20

 Star Transformation

• Execute star queries efficiently, especially in the

following cases:
– Many dimension tables
– A sparse fact table
– When not all dimensions have
constraining predicates
• Set the STAR_TRANSFORMATION_ENABLED
initialization parameter
• Use the STAR_TRANSFORMATION hint

Copyright © Oracle Corporation, 2002. All rights reserved.

Star Transformation (continued)

Star transformation is a cost-based query transformation aimed at executing star queries
efficiently.
The star transformation is especially effective if any of the following conditions exist:
• The number of dimension tables is large; this makes the Cartesian product of the
dimension tables too expensive.
• The fact table is sparse.
• There are queries where not all dimension tables have constraining predicates (no
additional nonjoin predicates).
The transformation can choose to combine bitmapped indexes corresponding precisely to the
constrained dimensions and does not need to produce the Cartesian product of the dimension
tables, which is the basis of the other approach to tuning star queries.
The STAR_TRANSFORMATION_ENABLED parameter is dynamic and can be set to TRUE at
the session level with the ALTER SESSION command:
SQL> alter session set star_transformation_enabled = true;
Alternatively, you can specify the STAR_TRANSFORMATION hint at the statement level.

Oracle9i: SQL Tuning Workshop 11-21

 Star Transformation Example

SQL> SELECT s.amount_sold, p.prod_name

2 , ch.channel_desc
3 FROM sales s, products p
4 , channels ch, customers c
5 WHERE s.prod_id= p.prod_id
6 AND s.channel_id = ch.channel_id
7 AND s.cust_id = c.cust_id
8 AND ch.channel_id in (’I’,’P’,’S’)
9 AND c.cust_city = ’Asten’
10 AND p.prod_id > 40000;

Copyright © Oracle Corporation, 2002. All rights reserved.

Star Transformation Example

In this example, the SALES table is the fact table, and CHANNELS, PRODUCTS, and
CUSTOMERS are the dimension tables.
Because there are bitmapped indexes on the CHANNEL_ID, PROD_ID, and CUST_ID
columns of the SALES table, the query can be rewritten so that these bitmapped indexes can be
used.
Note that you are joining four tables but selecting only column values of three of them; the
CUSTOMERS table is used only to select customers that are in Asten. Also note that there are no
nonjoin predicates on the facts table.
Note: It is not possible to force star transformation, and there are several constraints. Often the
optimizer does not consider a star transformation because a better execution plan is available, or
one of the constraints for star transformation is violated. For more information about star
transformation, see Oracle9i Concepts, “The Optimizer.”
Instructor Note
Do not try to demonstrate star transformation unless you are certain that it will work. Note that
the STAR_TRANSFORMATION hint works only if the optimizer thinks it is feasible.
The SQL statement is available in the demo11_02.sql script. demo11_02a.sql contains
the statement shown on the next page. The results are the same for both queries.
The next page shows how this statement can be transformed by the optimizer to benefit from
bitmapped indexes. Note that this is not how it exactly works internally, but is a way to explain
and visualize star transformation.
Oracle9i: SQL Tuning Workshop 11-22
 Star Transformation Example

SQL> SELECT s.amount_sold, p.prod_name ,ch.channel_desc

2 FROM sales s, products p, channels ch
3 WHERE s.prod_id= p.prod_id
4 AND s.channel_id = ch.channel_id
5 AND s.cust_id in
6 (SELECT cust_id
7 FROM customers)
8 AND s.channel_id in
9 (SELECT channel_id
10 FROM channels
11 WHERE channel_id in (’I’,’P’,’S’))
12 AND s.cust_id in
13 (SELECT cust_id
14 FROM customers
15 WHERE cust_city = ’Asten’)
16 AND s.prod_id in
17 (SELECT prod_id
18 FROM products
19 WHERE prod_id > 40000);

Copyright © Oracle Corporation, 2002. All rights reserved.

Star Transformation Example (continued)

Comparing this statement with the original one, you see that:
• Four subqueries are added
• The CUSTOMERS table disappeared in the FROM clause (line 2), together with its
corresponding join predicate (line 6 on the previous page).
Note that you do not need the CUSTOMERS table in the main query, because the SELECT
clause does not contain any CUSTOMERS column expression.
The four subqueries will be executed first, generating lists of values. You can use each of these
lists to retrieve bitmaps. These bitmaps can be merged, resulting in one bitmap for those rows in
the SALES table that match the condition in the subquery WHERE clause.
The resulting four bitmaps (one for each subquery) can be ANDed, resulting in one bitmap for
those rows in the SALES table that match all four conditions. Note that this final bitmap can be
very selective on the facts table without the need to perform any join operation yet. This means
that the join is now easier to perform and can be driven from the facts table, without the need to
produce a Cartesian product of the dimension tables, which is the basis of star join optimization,
as discussed in the “Sorting and Joining” lesson.
Instructor Note
Note that the join operation itself is not optimized yet, but the join is reduced to three tables,
and the final bitmap reduces the cardinality of the facts table. Conversely, three additional
tables must be accessed for the subqueries.
Oracle9i: SQL Tuning Workshop 11-23
 Function-Based Indexes

SQL> CREATE INDEX FBI_UPPER_LASTNAME

2 ON CUSTOMERS(upper(cust_last_name));

SQL> ALTER SESSION

2 SET QUERY_REWRITE_ENABLED = TRUE;

SQL> SELECT *
2 FROM customers
3 WHERE upper(cust_last_name) = ’SMITH’;

Copyright © Oracle Corporation, 2002. All rights reserved.

Function-Based Indexes
You can use Oracle9i to create indexes based on column expressions (virtual columns). Function-
based indexes provide an efficient mechanism for evaluating statements that contain expressions
in their WHERE clauses. For example, you can use function-based indexes to create case-
insensitive indexes, as the example in the slide shows.
You can create a function-based index to materialize computational-intensive expressions in the
index, so that the Oracle server does not need to compute the value of the expression when
processing SQL statements.
Function-based index expressions can use any function that is deterministic; that is, the returned
value must not change for a given set of input values. PL/SQL functions used in defining
function-based indexes must be DETERMINISTIC. The index owner needs the EXECUTE
privilege on the defining function. If the EXECUTE privilege is revoked, then the function-based
index is marked DISABLED. A function-based index can be created as a bitmap index.
Enabling Function-Based Indexes
Use the following command to enable function-based indexes:
SQL> alter session set query_rewrite_enabled = true;
Query rewrites are discussed in more detail in the “Materialized Views and Temporary Tables”
lesson.
Instructor Note
The Oracle server does not need to compute the value of the expression when processing SELECT
and DELETE statements. For INSERT and UPDATE statements the Oracle server must evaluate
the expression to update the index. Demonstration script: demo11_03.sql.
Oracle9i: SQL Tuning Workshop 11-24
 Function-Based Indexes: Usage

• Materialize computational-intensive expressions

• Facilitate case-insensitive searches
• Provide simple form of data compression
• Can be used for an NLS sort index

Copyright © Oracle Corporation, 2002. All rights reserved.

Data Dictionary Information

You can query the data dictionary to see specific information about function-based indexes:
SQL> select table_name, index_name, index_type
2 from user_indexes
3 where index_name = ’FBI_UPPER_LASTNAME’;

TABLE_NAME INDEX_NAME INDEX_TYPE

---------- -------------------- ---------------------
CUSTOMERS FBI_UPPER_LASTNAME FUNCTION-BASED NORMAL

SQL> select column_expression

2 from user_ind_expressions
3 where index_name = ’FBI_UPPER_LASTNAME’;

COLUMN_EXPRESSION
-------------------------------------------------
UPPER("CUST_LAST_NAME")
Instructor Note
Demonstration script: demo11_04.sql

Oracle9i: SQL Tuning Workshop 11-25

 Data Dictionary Information

SQL> SELECT i.index_name, i.index_type

2 , ic.column_name, i.status
3 FROM user_indexes i
4 , user_ind_columns ic
5 WHERE i.index_name = ic.index_name
6 AND i.table_name=’SALES’;

Copyright © Oracle Corporation, 2002. All rights reserved.

Data Dictionary Information (continued)

The result of the above query is:

INDEX_NAME INDEX_TYPE COLUMN_NAME STATUS

----------------- --------------- -------------------- --------
SALES_PROD_BIX BITMAP PROD_ID N/A
SALES_CUST_BIX BITMAP CUST_ID N/A
SALES_TIME_BIX BITMAP TIME_ID N/A
SALES_CHANNEL_BIX BITMAP CHANNEL_ID N/A
SALES_PROMO_BIX BITMAP PROMO_ID N/A
Instructor Note
Demonstration script: demo11_05.sql
Purpose: To demonstrate the code shown.
Note: The results shown from running the script table are based on the COPY_SALES table.

Oracle9i: SQL Tuning Workshop 11-26

 Summary

In this lesson, you should have learned about:

• Bitmapped indexes
• Indexes and row-access methods
• Index hints
• Star transformation
• Function-based indexes

Copyright © Oracle Corporation, 2002. All rights reserved.

Summary
This lesson introduced you to bitmapped indexes, star transformations, and function-based
indexes.
• Bitmap indexing provides both substantial performance improvements and space savings
over usual (B*-tree) indexes when there are few distinct values in the indexed columns.
• The star transformation is a cost-based query transformation aimed at executing star
queries efficiently. Star transformations are used in data warehousing. One type of data
warehouse design centers around what is known as a star schema, which is characterized
by one or more large fact tables that contain the primary information in the data
warehouse and a number of much smaller dimension tables (or lookup tables), each of
which contains information about the entries for a particular attribute in the fact table.
• With function-based indexes you can create indexes based on column expressions (virtual
columns). Function-based indexes provide an efficient mechanism for evaluating
statements that contain expressions in their WHERE clauses, or for supporting linguistic
sorting.
You can use the data dictionary views to retrieve information about your indexes.

Oracle9i: SQL Tuning Workshop 11-27

Oracle9i: SQL Tuning Workshop 11-28
 Materialized Views
and Temporary Tables

Copyright © Oracle Corporation, 2002. All rights reserved.

Schedule: Timing Topic

60 minutes Lecture
60 minutes Total
 Objectives

After completing this lesson, you should be able to do

the following:
• Identify the purpose and benefits of materialized
views
• Create materialized views
• Enable query rewrites
• Create dimensions
• Identify the benefits of temporary tables

Copyright © Oracle Corporation, 2002. All rights reserved.

Objectives
After completing this lesson, you should be able to:
• Identify the purpose and benefits of materialized views
• Create materialized views
• Enable query rewrites
• Create dimensions
• Identify the benefits of temporary tables

Oracle9i: SQL Tuning Workshop 12-2

 Materialized Views

A materialized view:
• Is an “instantiation” of a SQL statement
• Has its own data segment and offers:
– Space management options
– Use of its own indexes
• Is useful for:
– Expensive and complex joins
– Summary and aggregate data

Copyright © Oracle Corporation, 2002. All rights reserved.

Materialized Views
A materialized view stores both the definition of a view plus the rows resulting from the
execution of the view. Like a view, it uses a query as the basis, but the query is executed at the
time the view is created, and the results are stored in a table. You can define the table with the
same storage parameters as any other table and place it in the tablespace of your choice. You
can also index the materialized view table like other tables to improve the performance of
queries executed against them.
When a query can be satisfied with data in a materialized view, the Oracle9i Server transforms
the query to reference the view rather than the base tables. By using a materialized view,
expensive operations such as joins and aggregations do not need to be re-executed.
Instructor Note
Materialized views (MVs) are replacing read-only snapshots. However, MVs provide greater
functionality, such as the ability to allow query rewrites directly or through dimensions, as
covered in this lesson. The SNAPSHOT keyword is supported in place of MATERIALIZED
VIEW for backward compatibility.
There were several issues and restrictions with MVs and query rewrites in 8.1.5. For example,
as soon as you have a non-join predicate in the MV definition, only exact matches will result in
a query rewrite. Make sure to test any MV demonstration, and don’t improvise.

Oracle9i: SQL Tuning Workshop 12-3

 Create Materialized Views

SQL> CREATE MATERIALIZED VIEW

2 fweek_pscat_costs_mv
3 AS
4 SELECT t.week_ending_day
5 , t.calendar_year
6 , p.prod_subcategory
7 , sum(c.unit_cost) AS dollars
8 FROM costs c
9 , times t
10 , products p
11 WHERE c.time_id = t.time_id
12 AND c.prod_id = p.prod_id
13 GROUP BY t.week_ending_day
14 , t.calendar_year
15 , p.prod_subcategory;

Materialized view created.

Copyright © Oracle Corporation, 2002. All rights reserved.

Create Materialized Views

If a query involving summaries, large or multiple joins, or even both, is likely to be used
multiple times, it can be more efficient to create a materialized view of the query results. This
requires a single execution of the query and the storage space to preserve the results.
If the queries are likely to be reused over time, you may also need a mechanism to update the
materialized view as the base tables change. The performance and storage costs of maintaining
the materialized view must be compared to the costs of re-executing the original query
whenever it is needed. This is discussed in more detail later in this lesson.
The process of modifying a query to use the view rather than the base table is called a query
rewrite. This is also discussed in more detail later in this lesson.
Note: This is just an example of the CREATE MATERIALIZED VIEW command; more
details about optional command clauses will be discussed later during this lesson.
Instructor Note
Demonstration script: demo12_01.sql.
Purpose: The script creates the MV with the ENABLE QUERY REWRITE clause; this can also
be done later with the ALTER MATERIALIZED VIEW command. The script also creates
some indexes on the MV and analyzes the MV to collect statistics.

Oracle9i: SQL Tuning Workshop 12-4

 Refresh Materialized Views

• Refresh types:
– COMPLETE
– FAST
– FORCE
– NEVER
• Create materialized view logs for FAST refreshes
SQL> CREATE MATERIALIZED VIEW LOG ON …

Copyright © Oracle Corporation, 2002. All rights reserved.

Refresh Materialized Views

Depending on the activity on the base tables and the accuracy of the information required,
refreshing of materialized views can be done accordingly. The refresh mechanism is similar to
that used by snapshots: COMPLETE, FAST, FORCE, and NEVER.
A COMPLETE refresh involves truncating existing data and reinserting all the data by re-
executing the materialized view query definition.
FAST refreshes apply only the changes made since the last refresh. Two types are available:
• Using materialized view logs: In this case, all changes to the base tables are captured in a
log and then applied to the materialized view.
• Using ROWID ranges: A materialized view can be refreshed fast after direct path loads,
based on the ROWIDs of the new rows. Direct loader logs are required for this refresh
type.
Not all materialized views can use a FAST refresh; the Oracle9i SQL Reference manual
contains descriptions of the various view types and their refresh options.
A materialized view defined with a refresh type of FORCE refreshes with the fast mechanism if
possible, or else uses a complete refresh. Force is the default refresh type.
The NEVER option suppresses all refreshes of the materialized view.
Note: Materialized view logs are created using the CREATE MATERIALIZED VIEW LOG
command.

Oracle9i: SQL Tuning Workshop 12-5

 Refresh Materialized Views

• Manual refresh
– By using the DBMS_MVIEW package
• Automatic refresh
– Synchronous: Upon commit of changes made to the
underlying tables—but independent of the committing
transaction
– Asynchronous: Define a refresh interval
for the materialized view

Copyright © Oracle Corporation, 2002. All rights reserved.

Refresh Materialized Views (continued)

As well as choosing the refresh type, you can decide what mode to use to keep materialized
views synchronized with their base tables.
Manual Refresh
Manual refreshes are performed using the DBMS_MVIEW package. The DBMS_MVIEW
package provides a number of procedures and functions to manage materialized views,
including the REFRESH, REFRESH_DEPENDENT, and REFRESH_ALL_MVIEWS
procedures.
Automatic Refresh
Automatic refresh can be performed:
• ON COMMIT: When this option is specified for a materialized view, it is updated
whenever changes to one of the base tables are committed.
Note: The refresh of the MV is done as part of the transaction so the commit will take
longer.
• At a specified time: Refresh on a materialized view can be scheduled to occur at a
specified time. For example, it can be refreshed every Monday at 9:00 a.m. by using the
START WITH and NEXT clauses. In order for such refreshes to occur, the instance must
initiate job processes with the JOB_QUEUE_PROCESSES parameter.
As with refresh types, refresh modes are not all available to every materialized view. The
Oracle9i Supplied PL/SQL Packages and Types Reference manual describes the modes and
when each can be used.
Refer to the Oracle9i Data Warehousing Guide for more information on Materialized views.

Oracle9i: SQL Tuning Workshop 12-6

 Materialized Views: Manual Refresh

• Specific materialized views

DBMS_MVIEW.REFRESH
(’FWEEK_PSCAT_COSTS_MV’
, PARALLELISM => 10);
• Materialized views based on one or more tables
DBMS_MVIEW.REFRESH_DEPENDENT
(’CAL_MONTH_SALES_MV’);

• All materialized views due for refresh

DBMS_MVIEW.REFRESH_ALL_MVIEWS(n);
– Returns a binary integer value as an output parameter

Copyright © Oracle Corporation, 2002. All rights reserved.

Manual Refresh (Continued)

The following is a list of possible refresh scenarios for materialized views:
• Refresh specific materialized views using the REFRESH procedure
• Refresh all materialized views that depend on a given set of base tables using the
REFRESH_DEPENDENT procedure
• Refresh all materialized views that have not been refreshed since the last bulk load to one
or more detail tables using the REFRESH_ALL_MVIEWS procedure
The procedures in the package use a number of parameters to specify the following:
• Refresh method
• Whether to proceed if an error is encountered
• Whether to use a single transaction (consistent refresh)
• The rollback segment to use
You need server job queues to run the refresh job. Therefore the appropriate initialization
parameters, JOB_QUEUE_PROCESSES and JOB_QUEUE_INTERVAL, must be set to enable
job queue refresh processes.

Oracle9i: SQL Tuning Workshop 12-7

 Query Rewrites

• To use a materialized view instead of the base

tables, a query must be rewritten.
• Query rewrites are transparent to applications.
• Query rewrites do not require special privileges on
the materialized view.
• A materialized view can be enabled or disabled for
query rewrites.

Copyright © Oracle Corporation, 2002. All rights reserved.

Query Rewrites
Because accessing a materialized view may be significantly faster than accessing the
underlying base tables, the optimizer rewrites a query to access the view when the query allows
it. The query rewrite activity is transparent to applications. In this respect, their use is similar to
the use of an index.
Users do not need explicit privileges on materialized views to use them. Queries executed by
any user with privileges on the underlying tables can be rewritten to access the materialized
view.
A materialized view can be enabled or disabled. A materialized view that is enabled is
available for query rewrites.

Oracle9i: SQL Tuning Workshop 12-8

 Query Rewrites

The optimizer rewrites queries in order to use

materialized views:
• The QUERY REWRITE privilege allows users to enable
materialized views.
• The DBMS_OLAP package has options to use
materialized views.

Copyright © Oracle Corporation, 2002. All rights reserved.

Query Rewrites (continued)

To enable or disable individual materialized views for query rewrites, you must have the
GLOBAL QUERY REWRITE or the QUERY REWRITE system privilege. Both versions of the
privilege allow you to enable materialized views in your own schema. The GLOBAL version
allows you to enable any materialized views you own, whereas the simple QUERY REWRITE
privilege requires that the base tables, as well as the views, be in your own schema.
The DBMS_OLAP supplied package provides a collection of materialized view analysis and
advisory functions. For more details, see the Oracle9i Supplied PL/SQL Packages and Types
Reference.

Oracle9i: SQL Tuning Workshop 12-9

 Query Rewrites

• Use EXPLAIN PLAN or AUTOTRACE to verify that

query rewrites occur.
• Check the query response:
– Fewer blocks are accessed
– Response time should be significantly better

Copyright © Oracle Corporation, 2002. All rights reserved.

Query Rewrites (continued)

The best method to detect whether query rewrites occur is to use the EXPLAIN PLAN
command or the AUTOTRACE setting in SQL*Plus. You should also notice improved response
time if a materialized view is used by the optimizer.
Note: There are several system privileges that control whether you are allowed to create
materialized views and modify them, and whether query rewrites are enabled for you. Contact
your local database administrator to set this up properly, or refer to the Oracle9i
Administrator’s Guide for details.
There are also many data dictionary views that contain information about materialized views;
check the Oracle9i Performance Guide and Reference manual and the Oracle9i Reference for
details.
Instructor Note
Note that the main audience for this class is developers, not database administrators; try to
avoid DBA-related topics. Also note that there are some bugs with SQL*Plus AUTOTRACE not
showing the query rewrite, because it uses an implicit second database session that does not
inherit all settings from your main session. Using EXPLAIN is safer.
Instructor Note (for page 12-12)
Demonstration script: demo12_02.sql

Oracle9i: SQL Tuning Workshop 12-10

 Create Materialized Views:
Syntax Options

CREATE MATERIALIZED VIEW mview_name

[TABLESPACE ts_name]
[PARALLEL (DEGREE n)]
[BUILD {IMMEDIATE|DEFERRED}]
[REFRESH {FAST|COMPLETE|FORCE
|NEVER|ON COMMIT}]
[{ENABLE|DISABLE} QUERY REWRITE]

AS SELECT … FROM …

Copyright © Oracle Corporation, 2002. All rights reserved.

Create Materialized Views: Syntax Options

The CREATE MATERIALIZED VIEW syntax is similar to the CREATE SNAPSHOT
command, which it replaces. There are some additional options:
• The BUILD IMMEDIATE option causes the materialized view to be populated when the
CREATE command is executed. This is the default behavior. You can choose the BUILD
DEFERRED option, which creates the structure but does not populate it until the first
refresh occurs.
• Instead of the BUILD option, you can also specify ON PREBUILT TABLE when you
want an existing summary table to be the source of a materialized view.
• The ENABLE/DISABLE QUERY REWRITE clause determines whether query rewrites
are automatically enabled for the materialized view.
Note: The default options in the syntax are underlined. This is not the complete syntax; refer to
the Oracle9i SQL Reference for more details.
Instructor Note
You can use ON PREBUILT TABLE to convert an existing table to a preinitialized
materialized view. There is an additional clause for this option, WITH/WITHOUT REDUCED
PRECISION. This clause, which defaults to WITHOUT REDUCED PRECISION, determines
whether or not the columns in the subquery must have the same precision as the corresponding
columns in the table.

Oracle9i: SQL Tuning Workshop 12-11

 Enabling and Controlling
Query Rewrites

• Query rewrites are available only with cost-based

optimization.
QUERY_REWRITE_ENABLED = {true|false}
QUERY_REWRITE_INTEGRITY =
{enforced|trusted|stale_tolerated}

• Optimizer hints to influence query rewrites are

REWRITE and NOREWRITE.

Copyright © Oracle Corporation, 2002. All rights reserved.

Enabling and Controlling Query Rewrites

OPTIMIZER_MODE: Query rewrites are available only under cost-based optimization.
QUERY_REWRITE_ENABLED: This parameter can be set to FALSE to suppress query
rewrites.
This is a dynamic instance/session parameter.
QUERY_REWRITE_INTEGRITY: This is also a dynamic instance/session parameter.
It accepts the following values:
• ENFORCED (the default) enables query rewrites only if the optimizer can guarantee
consistency. Only updated materialized views and enabled validated constraints are used
for query rewrites.
• TRUSTED allows query rewrites based on declared (not necessarily enforced)
relationships. All updated materialized views and dimensions (discussed later in this
lesson) and constraints with the RELY flag are used for query rewrites.
• STALE_TOLERATED allows query rewrites to use materialized views that have not been
refreshed since the last DML operation and relationships that have been declared.
Note: There are no object privileges associated with query rewrites. Users with access to the
base tables implicitly benefit from query rewrites.
You can use the hint REWRITE to restrict the materialized views that are considered for query
rewrites. The NOREWRITE hint is available to suppress query rewrites.

Oracle9i: SQL Tuning Workshop 12-12

 Query Rewrite Example

SQL> explain plan for

2 SELECT t.week_ending_day
3 , t.calendar_year
4 , p.prod_subcategory
5 , sum(c.unit_cost) AS dollars
6 FROM costs c
7 , times t
8 , products p
9 WHERE c.time_id = t.time_id
...

OPERATION NAME
------------------------ --------------------
SELECT STATEMENT
TABLE ACCESS FULL fweek_pscat_costs_mv

Copyright © Oracle Corporation, 2002. All rights reserved.

Query Rewrite Example

The execution plan shows that the materialized view is accessed instead of joining all four base
tables to produce the result.
A REWRITE/NOREWRITE hint overrides a materialized view’s definition, set in the CREATE
or ALTER MATERIALIZED VIEW command with the ENABLE QUERY REWRITE clause.
This example shows a transparent query rewrite where the query exactly matches the
materialized view definition. The next page shows an example of a query that does not match
the materialized view definition.
Instructor Note
Tell the students to analyze the materialized view, as one of the previous demonstrations did.
Demonstration script: demo12_03.sql.

Oracle9i: SQL Tuning Workshop 12-13

 Query Rewrite Example
SQL> SELECT t.week_ending_day
2 , t.calendar_year
3 , p.prod_subcategory
4 , sum(c.unit_cost) AS dollars
5 FROM costs c, times t, products p
6 WHERE c.time_id = t.time_id
7 AND c.prod_id = p.prod_id
8 AND t.calendar_year = ’1999’
9 GROUP BY t.week_ending_day, t.calendar_year
10 , p.prod_subcategory
11 HAVING sum(c.unit_cost) > 10000;

SQL> SELECT week_ending_day

2 , prod_subcategory
3 , dollars
4 from fweek_pscat_costs_mv
5 where calendar_year = ’1999’
6 and dollars > 10000;

Copyright © Oracle Corporation, 2002. All rights reserved.

Query Rewrite Example (continued)

The optimizer can use the materialized view created earlier to satisfy the query.
Note: This query does not exactly match the materialized view definition. You have added a
nonjoin predicate on line 8 and a HAVING clause on the last line. The nonjoin predicate is
merged into the rewritten query against the materialized view and the HAVING clause is
translated into a second component of the WHERE clause.
Instructor Note
Demonstration script: demo12_04.sql

Oracle9i: SQL Tuning Workshop 12-14

 Dimensions: Overview

• Dimensions are data dictionary structures that

define hierarchies based on existing columns.
• Dimensions are optional but recommended because
they:
– Enable additional query rewrites
– Help document hierarchies
– Can be used by OLAP tools

Copyright © Oracle Corporation, 2002. All rights reserved.

Dimensions
Dimensions are data dictionary structures that define hierarchies based on columns in existing
database tables. Although they are optional, they are recommended because they:
• Enable additional rewrite possibilities without the use of constraints (Implementation of
constraints may not be desirable in a data warehouse for performance reasons).
• Help document dimensions and hierarchies explicitly.
• Can be used by OLAP tools.
A dimension is an object stored in the data dictionary that enables a query to leverage the
results of an existing summary to produce another summary much cheaper. For example if the
original query had 1700 I/Os, using the materialized view would take about 170 I/Os and the
query on using a dimension would take approximately 17 I/Os.
Instructor Note
If you are running out of time, you can say that dimensions enable additional query rewrite
capabilities, specifically targeted at data warehouse environments; then you can refer them to
the excellent chapters in the Oracle9i Data Warehousing Guide manual for more details. You
can then continue to discuss temporary tables, the last topic of this lesson.

Oracle9i: SQL Tuning Workshop 12-15

 Dimensions and Hierarchies

All

Fiscal year Year

Hierarchy

Fiscal quarter Quarter Attributes:

Day name
Month name
Fiscal month Month Quarter
description
Days in quarter

Fiscal week

Day

Copyright © Oracle Corporation, 2002. All rights reserved.

Dimensions and Hierarchies

Dimensions describe business entities such as products, departments, and time in a
hierarchical, categorized manner. A dimension can consist of one or more hierarchies. In the
example shown in the slide, the time dimension consists of three hierarchies.
Each hierarchy comprises multiple levels. Each value at a lower level in the hierarchy is the
child of one and only one parent at the next higher level. A hierarchy consists of 1-to-n
relationships between levels, with the parent level representing a level of aggregation of the
child level. In the example, the Fiscal Time hierarchy consists of the fiscal week, fiscal
month, fiscal quarter, and fiscal year.
A level in the hierarchy has a level key that can identify one or more dimensional attributes. In
the example, month is a level where the level key MONTH identifies the MONTH_NAME
attribute.
You can use level keys and their related attribute names interchangeably in a query.

Oracle9i: SQL Tuning Workshop 12-16

 Dimensions: Example Table

SQL> SELECT time_id day

2 , calendar_month_desc month
3 , calendar_quarter_desc quarter
4 , calendar_year year
5 FROM times;

DAY MONTH QUARTER YEAR

--------- -------- ------- ----------
01-JAN-98 1998-01 1998-Q1 1998
02-JAN-98 1998-01 1998-Q1 1998
03-JAN-98 1998-01 1998-Q1 1998
04-JAN-98 1998-01 1998-Q1 1998
05-JAN-98 1998-01 1998-Q1 1998
06-JAN-98 1998-01 1998-Q1 1998
...
30-DEC-01 2001-12 2001-Q4 2001
31-DEC-01 2001-12 2001-Q4 2001

Copyright © Oracle Corporation, 2002. All rights reserved.

Dimensions: Example Table

You have a TIMES table containing time information in several separate columns.
On the next two pages you will create a dimension, TIMES_DIM, on this table.

Oracle9i: SQL Tuning Workshop 12-17

 Dimensions and Hierarchies

TIMES table columns TIMES_DIM dimension

- CALENDAR_YEAR - YEAR
- CALENDAR_QUARTER_DESC - QUARTER
- CALENDAR_MONTH_DESC - MONTH
- TIME_ID - DAY

Attributes
- DAY_NAME
- CALENDAR_MONTH_NAME
- DAYS_IN_CAL_QUARTER

Copyright © Oracle Corporation, 2002. All rights reserved.

Dimensions and Hierarchies

Dimensions can be based on columns in a single table, as shown in this first example. Later
during this lesson you see that you can also base dimensions on columns from multiple tables.
When creating dimensions, you must define and name hierarchy levels based on the TIMES
table column names.
In this case, the TIMES_DIM dimension has four levels:
• The highest level in the hierarchy consists of the CALENDAR_YEAR column.
• The next level is derived from the CALENDAR _QUARTER_DESC column.
• The next level is derived from the CALENDAR _MONTH_DESC column.
• The lowest level is based on the TIME_ID column.
• There are three attributes. The second level has the CALENDAR _QUARTER_DESC
column as the key, and DAYS_IN_CAL_QUARTER as the attribute. The third level has
the CALENDAR_MONTH_DESC column as the key and CALENDAR_MONTH_NAME as
the attribute. The fourth level has the TIME_ID column as the key, and DAY_NAME as
the attribute.
Note that each attribute value maps to a single key value and vice versa.

Oracle9i: SQL Tuning Workshop 12-18

 Create Dimensions and Hierarchies

SQL> CREATE DIMENSION times_dim

2 LEVEL day IS TIMES.TIME_ID
3 LEVEL month IS TIMES.CALENDAR_MONTH_DESC
4 LEVEL quarter IS TIMES.CALENDAR_QUARTER_DESC
5 LEVEL year IS TIMES.CALENDAR_YEAR
6 HIERARCHY cal_rollup (
7 day CHILD OF
8 month CHILD OF
9 quarter CHILD OF
10 year
11 )
12 ATTRIBUTE day DETERMINES (day_name)
13 ATTRIBUTE month DETERMINES (calendar_month_name)
14 ATTRIBUTE quarter DETERMINES
15 (days_in_cal_quarter);

Copyright © Oracle Corporation, 2002. All rights reserved.

Defining Dimensions and Hierarchies

Hierarchical relationships can be declared between two columns in a dimension table, or
between two columns in different tables in the case of normalized or snowflake schemas to
support such rewrites. A new schema object type, dimension, is provided for this purpose.
The parent-child relationship along a dimension is stronger than referential integrity alone
because it requires that the child join key be mandatory (that is, defined as NOT NULL).
The creator of a dimension must ensure (optionally, with database constraints in the case of a
normalized dimension) that the columns of each hierarchy level are non-NULL and that
hierarchical integrity is maintained.
Note that the key and attributes for a given level must be based on columns from the same
table.
The example shows a single hierarchy within the time dimension, but it is possible to have
multiple hierarchies. For example, another hierarchy can be created to link sales date with a
fiscal calendar or season.
The system privilege, CREATE DIMENSION, is required to create a dimension in your own
schema based on tables that are within the same schema. Another privilege, CREATE ANY
DIMENSION, enables you to create dimensions in any schema.
In the example shown, the TIMES_DIM dimension is based on the TIMES table.

Oracle9i: SQL Tuning Workshop 12-19

 Dimensions Based on Multiple Tables

• Example: A GEOGRAPHY dimension is based on

CITY, STATE, and REGION tables.
• Define foreign key constraints on the child tables
to maintain dimension validity.
• All attributes and keys for one level must belong
to the same table.

Copyright © Oracle Corporation, 2002. All rights reserved.

Dimensions Based on Multiple Tables

A dimension can use columns from several tables. For example, a GEOGRAPHY dimension can
use the following tables and columns:
• CITY table containing CITY_NAME , OFFICE_ADDRESS, POPULATION, and
STATE_CODE
• STATE table consisting of STATE_CODE, STATE_NAME, and REGION_ID
• REGION table comprising REGION_ID, REGION_NAME, and COUNTRY
In this case, foreign key constraints should be defined on a child table to reference the primary
key of the parent table. An example may be CITY.STATE_CODE referencing
STATE.STATE_CODE. This helps maintain dimension validity.
Note that all columns for a given level are stored in the same table.

Oracle9i: SQL Tuning Workshop 12-20

 Dimensions with Multiple Hierarchies

All
Hierarchy Hierarchy
FIS_ROLLUP Fiscal year Year CAL_ROLLUP

Fiscal quarter Quarter

Fiscal month Month

Fiscal week

Day

Copyright © Oracle Corporation, 2002. All rights reserved.

Dimensions with Multiple Hierarchies

The pair of hierarchies shown in the slide can be created within a single dimension. Following
is the statement to do this:
SQL> CREATE DIMENSION times_dim
2 LEVEL day IS TIMES.TIME_ID
3 LEVEL month IS TIMES.CALENDAR_MONTH_DESC
4 LEVEL quarter IS TIMES.CALENDAR_QUARTER_DESC
5 LEVEL year IS TIMES.CALENDAR_YEAR
6 LEVEL fis_week IS TIMES.WEEK_ENDING_DAY
7 LEVEL fis_month IS TIMES.FISCAL_MONTH_DESC
8 LEVEL fis_quarter IS TIMES.FISCAL_QUARTER_DESC
9 LEVEL fis_year IS TIMES.FISCAL_YEAR
10 HIERARCHY cal_rollup (
11 day CHILD OF
12 month CHILD OF
13 quarter CHILD OF year )
14 HIERARCHY fis_rollup (
15 day CHILD OF
16 fis_week CHILD OF
17 fis_month CHILD OF
18 fis_quarter CHILD OF fis_year );

Oracle9i: SQL Tuning Workshop 12-21

 Temporary Tables

• Temporary tables retain data for the duration of a

transaction or session.
• Definitions persist in the data dictionary.
• Data is only visible to the session.
• Temporary tables:
– Use sort space in memory
– Allocate temporary extents, if needed

Copyright © Oracle Corporation, 2002. All rights reserved.

Temporary Tables
With the Oracle8i Server and later, you can create temporary tables. Temporary tables can
improve performance significantly by holding temporary data for reuse within your transaction
or session.
A temporary table has the following properties:
• Temporary table data is visible only within its defined scope; the scope can be defined to
be a session or a transaction.
• The definition of a global temporary table is visible to all sessions. In contrast, the
definition of a local temporary table does not persist at the end of the session that creates
it.
• Temporary table data is stored within the sort space used by the session. If sort space is
not sufficient to accommodate the data, space is allocated in the user’s temporary
tablespace.
• Indexes on temporary tables have the same scope and duration as the table they
correspond to.
• Triggers and views can be defined on temporary tables. However, a view cannot be
defined joining a temporary and a permanent table.
• The CREATE GLOBAL TEMPORARY TABLE AS SELECT command can be used to
create a temporary table and insert data into it.
• Definitions of temporary tables can be exported and imported.

Oracle9i: SQL Tuning Workshop 12-22

 Temporary Tables

SQL> CREATE OR REPLACE VIEW sales_detail

2 AS
3 SELECT cu.cust_last_name, cu.cust_email
4 , cu.cust_income_level
5 , pr.prod_name, ch.channel_desc
6 , pm.promo_name, sa.amount_sold
7 FROM customers cu, products pr
8 , channels ch, promotions pm, sales sa
9 WHERE sa.cust_id = cu.cust_id
10 AND sa.prod_id = pr.prod_id
11 AND sa.channel_id = ch.channel_id
12 AND sa.promo_id = pm.promo_id
13 AND sa.time_id between
14 ’01-DEC-1999’ and ’31-DEC-1999’

View created.

Copyright © Oracle Corporation, 2002. All rights reserved.

Creating Temporary Tables

You can use temporary tables to improve performance when you run complex queries.
Running multiple such queries is relatively slow because the tables are accessed multiple times
for each returned row. It is faster to cache the values from a complex query in a temporary
table, then run the queries against the temporary table.
For example, even with a view defined to simplify further queries, as shown in the slide, the
queries against the view may be slow because the contents of the view are recalculated each
time.
You can use a temporary table to run the computation once, and cache the result in later SQL
queries and joins.

Oracle9i: SQL Tuning Workshop 12-23

 Creating Temporary Tables

SQL> CREATE GLOBAL TEMPORARY TABLE sales_detail_temp

2 ( cust_last_name VARCHAR2(50)
3 , cust_income_level VARCHAR2(30)
4 , cust_email VARCHAR2(30)
5 , prod_name VARCHAR2(50)
6 , channel_desc VARCHAR2(20)
7 , promo_name VARCHAR2(20)
8 , amount_sold NUMBER)
9 ON COMMIT PRESERVE ROWS;

Table created.

SQL> INSERT INTO sales_detail_temp

2 SELECT * FROM sales_detail;

37830 rows created.

Copyright © Oracle Corporation, 2002. All rights reserved.

Creating Temporary Tables (Continued)

The example in the slide shows the command to create a temporary table where rows are
visible only within the session. The temporary table is then populated with rows from the
previously defined view. The computation is performed once and the cached results are
available for future queries.
The clauses that control the duration of the rows are:
• ON COMMIT DELETE ROWS to specify that rows are visible only within the
transaction (This is the default behavior).
• ON COMMIT PRESERVE ROWS to specify that rows are visible for the entire session.
Although the command does not create extents for the table, a session can query the table
without receiving a error.
Instructor Note
There is a bug in 8.1.5 that shows the DURATION to be SYS$TRANSACTION even if the table
is defined with ON COMMIT PRESERVE ROWS. The data in the table, however, is visible
throughout the session. Demonstration script: demo12_05.sql.

Oracle9i: SQL Tuning Workshop 12-24

 Creating Temporary Tables

SQL> SELECT table_name, temporary, duration

2 FROM dba_tables
3 WHERE table_name = ’SALES_DETAIL_TEMP’;

TABLE_NAME T DURATION
------------------------------ - ------------
SALES_DETAIL_TEMP Y SYS$SESSION

Copyright © Oracle Corporation, 2002. All rights reserved.

Creating Temporary Tables (Continued)

The view DBA_TABLES contains columns to indicate that the table is a temporary table. The
column DURATION shows whether data in the table persists for a whole session or only within
the transaction.

Oracle9i: SQL Tuning Workshop 12-25

 Summary

In this lesson, you should have learned how to:

• Create materialized views
• Enable query rewrites using materialized views
• Enable additional query rewrite capabilities with
dimensions and hierarchies
• Create temporary tables to hold data within the
scope of a session or transaction

Copyright © Oracle Corporation, 2002. All rights reserved.

Summary
This lesson introduced you to materialized views, query rewrites, dimensions and hierarchies,
and temporary tables:
• A materialized view stores both the definition of a view plus the rows resulting from the
execution of the view. Like a view, it uses a query as the basis, but the query is executed
at the time the view is created and the results are stored in a table.
• Because accessing a materialized view can be significantly faster than accessing the
underlying base tables, the optimizer rewrites a query to access the view when the query
allows it. The query rewrite activity is transparent to applications.
• Dimensions are data dictionary structures that define hierarchies based on columns in
existing database tables. Dimensions describe business entities such as products,
departments, and time in a hierarchical, categorized manner. A dimension can consist of
one or more hierarchies. Each hierarchy comprises multiple levels.
• The Oracle9i Server enables you to create temporary tables. Temporary tables can
improve performance significantly by holding temporary data for reuse within your
transaction or session.

Oracle9i: SQL Tuning Workshop 12-26

 Alternative Storage Techniques

Copyright © Oracle Corporation, 2002. All rights reserved.

Schedule: Timing Topic

45 minutes Lecture
45 minutes Total
 Objectives

After completing this lesson, you should be able to

identify the characteristics and benefits of:
• Index-organized tables
• External tables

Copyright © Oracle Corporation, 2002. All rights reserved.

Objectives
After completing this lesson, you should be able to identify and describe the benefits of:
• Index-organized tables
• External tables

Oracle9i: SQL Tuning Workshop 13-2

 Storing User Data

Regular
table

External
table OS file

Index-organized
table

Copyright © Oracle Corporation, 2002. All rights reserved.

Storing User Data

The Oracle9i Server offers a wide range of techniques to store user data: regular tables, index-
organized tables, partitioned tables, and external tables.
Partitioning is not covered in this course. This lesson concentrates on the other, nonregular table
storage techniques: index-organized tables and external tables.
Changing the physical storage characteristics of tables sometimes results in significant
performance improvements for certain SQL statements. However, be aware that there are usually
negative side effects.
Data maintenance is more complicated for nonregular tables.
Change regular tables into index-organized tables or external tables when the advantages
outweigh the disadvantages significantly and when these disadvantages do not violate any
minimum performance constraints.
Instructor Note
Partitioning is covered in Oracle9i: Implementing Scalable Systems.

Oracle9i: SQL Tuning Workshop 13-3

 Index-Organized Tables

Indexed access Accessing index-

on table organized table

ROWID

Non-key columns
Key column
Row header

Copyright © Oracle Corporation, 2002. All rights reserved.

Index-Organized Tables: Overview

An indexed-organized table (IOT) is like a regular table with an index on one or more of its
columns, but instead of maintaining two separate segments for the table and the B*-tree index,
the Oracle server maintains one single B*-tree structure which contains:
• The primary key value
• Other (non-key) column values for the row
The B*-tree structure, which is based on the primary key of the table, is organized like an index.
The leaf blocks in this structure contain the rows instead of the ROWIDs. This means that the rows
in the index-organized table are always maintained in the order of the primary key.
Because index-organized tables do not have real ROWIDs, the concept of logical ROWIDs is
introduced in Oracle8i. This enables you to create additional indexes on index-organized tables;
this was not possible in Oracle8 release 8.0.
Because large rows of an index-organized table can destroy the dense and efficient storage of the
B*-tree structure, in Oracle8i you can store part of the row in another segment, called an overflow
area. This is discussed on the following pages.

Oracle9i: SQL Tuning Workshop 13-4

 IOT Performance Characteristics

• B*-tree stores full rows

• Sorted rows
• Logical ROWIDs
• Fast, key-based
access to table data

Copyright © Oracle Corporation, 2002. All rights reserved.

IOT Performance Characteristics

You can access an index-organized table by using either the primary key or a combination of
columns that constitute the leading part of the primary key.
IOTs provide fast, key-based access for queries involving exact match (equality operator) or
range searches on the primary key.
Before Oracle8i, any other access to rows in an index-organized table required a full scan of the
B*-tree structure. That is why Oracle8 introduced the concept of logical ROWIDs, which you can
use to create additional indexes on an index-organized table. Logical ROWIDs internally have the
UROWID data type.
Because the rows are ordered in the IOT, full scans on an index-organized table return rows in a
primary key sequence.
Because there is no duplication of primary key values (compared with regular tables: index
segment and data segment), IOTs use less storage. Index organization is useful for a table that is
frequently accessed using the primary key and has only a few, relatively short nonkey columns.
Instructor Note
The statement in the fourth paragraph about returning the rows in primary key sequence is true
only for serial execution; it is no longer true with parallel execution. Also, the first bullet in the
slide is not always true: the concept of row overflow is introduced on the following pages.

Oracle9i: SQL Tuning Workshop 13-5

 IOT Limitations

• Must have a primary key

• Cannot be part of an index cluster or hash cluster
• Cannot contain LONG columns (although LOB
columns are allowed)

Copyright © Oracle Corporation, 2002. All rights reserved.

IOT Limitations
• Index-organized tables must have a primary key. This is the unique identifier and is used as
the basis for ordering; there is no ROWID to act as a unique identifier in the B*-tree
structure.
• Index-organized tables cannot be part of an index cluster or a hash cluster.
• Index-organized tables cannot include LONG columns, but they can contain LOB columns.
Because index-organized tables are B*-tree structures, they are subject to fragmentation as a
result of incremental updating. You can use the ALTER TABLE … MOVE command to rebuild
the index-organized table:
ALTER TABLE iot_tablename MOVE [OVERFLOW...];
Specifying the optional OVERFLOW clause causes the overflow segment to be rebuilt as well.
Overflow segments are explained on the following pages.
Instructor Note
Oracle8i removed the IOT restrictions regarding secondary indexes and additional UNIQUE
constraints. You can mention the absence of real ROWIDs as an IOT limitation, although the
implementation of logical ROWIDs (with the UROWID data type) resolved all related
shortcomings.
Note that the MOVE option for the ALTER TABLE statement was new in Oracle8i.
Oracle9i: SQL Tuning Workshop 13-6
 When to Use Index-Organized Tables

• Note that IOTs are transparent to users and

applications
• Content-based information applications: text,
images, and sound storage or retrieval

Copyright © Oracle Corporation, 2002. All rights reserved.

When to Use Index-Organized Tables

Applications that use complex data objects have many different indexing requirements. For
example, information-retrieval applications may require a keyword search or a search for similar
words.
Index-organized tables are particularly useful for applications accessing and manipulating
complex data. Examples are information-retrieval, spatial, and OLAP applications.
No special considerations exist for using most SQL statements against an index-organized table.
The structure of the table should be completely transparent to the user.
The Oracle utilities also support index-organized tables. SQL*Loader, using direct path mode,
can load data directly into an index-organized table. This kind of loading can be quicker than
loading into a standard table and then building the indexes afterwards.
Note: IOTs are supported by Oracle8i and later utilities.

Oracle9i: SQL Tuning Workshop 13-7

 Creating Index-Organized Tables

CREATE TABLE table-name

( column_definitions
[,constraint_definitions] )
ORGANIZATION INDEX
[ block_util_parameters ]
[ storage_clause ]
[ TABLESPACE tablespace ]
[ PCTTHRESHOLD integer
[ INCLUDING column_name ] ]
[ OVERFLOW segment_attr_clause ]

Copyright © Oracle Corporation, 2002. All rights reserved.

Creating Index-Organized Tables

The ORGANIZATION INDEX clause of the CREATE TABLE statement specifies that you create
an index-organized table. You must specify a primary key constraint when creating index-
organized tables. If you try to create an index-organized table without a primary key, the
following error is generated:
ORA-25175: no PRIMARY KEY constraint found
PCTTHRESHOLD specifies the percentage of space reserved in a block for a single row in an
index-organized table . If a row exceeds the size calculated based on this value, all columns after
the column named in the INCLUDING clause are moved to the overflow segment. If OVERFLOW
is not specified, then rows exceeding the threshold are rejected. PCTTHRESHOLD defaults to 50
and must be a value from 0 to 50.
INCLUDING column_name specifies a column at which to divide an index-organized table
row into index and overflow portions. All columns that follow column_name are stored in the
overflow data segment. If this is not specified and a row size exceeds PCTTHRESHOLD, all
columns except the primary key columns are moved to the overflow area. The column is either
the name of the last primary key column or any nonprimary key column.
OVERFLOW specifies that index-organized table rows exceeding the specified threshold be placed
in the data segment defined by segment_attr_clause, which specifies the tablespace,
storage, and block utilization parameters.

Oracle9i: SQL Tuning Workshop 13-8

 IOT Row Overflow

IOT tablespace Overflow tablespace

Segment <PK name> Segment SYS_IOT_OVER_n
type: index type: table

Block Block

Rows within Overflow

PCTTHRESHOLD row piece

Copyright © Oracle Corporation, 2002. All rights reserved.

IOT Row Overflow

Large rows in an index-organized table can destroy the dense B*-tree storage. You can overcome
this problem by using an overflow area.
Note that you need additional I/Os to retrieve these large rows because the rows are stored in two
pieces. This results in a decrease in performance compared to an IOT with shorter records that are
mostly stored entirely in the IOT segment alone.
When an index-organized table is created by specifying an OVERFLOW clause, the following
three segments are created in the database:
• A logical table with the name defined in the CREATE TABLE clause
• An index with the same name as the primary key constraint
• A table to accommodate the overflow row pieces (The name of this table is
SYS_IOT_OVER_n, where n is the OBJECT_ID of the index-organized table as seen from
USER_OBJECTS.)
Note: If you create an index-organized table without specifying an OVERFLOW clause, only the
first two segments are created. Give the primary key constraint a name, so that the index segment
receives a meaningful name (as opposed to a system-generated one.)

Oracle9i: SQL Tuning Workshop 13-9

 Retrieving IOT Information

USER_TABLES USER_INDEXES

TABLE_NAME TABLE_NAME
IOT_TYPE INDEX_NAME
IOT_NAME INDEX_TYPE
TABLESPACE_NAME PCT_THRESHOLD
INCLUDE_COLUMN

Copyright © Oracle Corporation, 2002. All rights reserved.

Retrieving IOT Information from the Data Dictionary

Use the following query to list the index-organized tables and information related to their
structure:
SQL> SELECT t.table_name as "IOT"
2 , o.table_name as "Overflow"
3 , i.index_name as "Index"
4 , o.tablespace_name as "Overflow TS"
5 , i.tablespace_name as "Index TS"
6 , i.pct_threshold as "Threshold"
7 FROM user_tables t
8 , user_tables o
9 , user_indexes i
10 WHERE t.table_name = o.iot_name
11 AND t.table_name = i.table_name;

Oracle9i: SQL Tuning Workshop 13-10

 External Tables

SELECT *
FROM ex_table;

External
table OS file

Copyright © Oracle Corporation, 2002. All rights reserved.

External Tables: Overview

Oracle9i provides a way to access data in external sources as if it were in a table in the database.
Oracle allows you read-only access to data in external tables. External tables are defined as tables
that do not reside in the database and can be in any format for which an access driver is provided.
By providing Oracle with metadata describing an external table, Oracle is able to expose the data
in the external table as if it were data residing in a regular database table. The external data can be
queried directly and in parallel using SQL.
You can select, join, or sort external table data. You can also create views and synonyms for
external tables. However, no DML operations (UPDATE, INSERT, or DELETE) are possible, and
no indexes can be created on external tables.

Oracle9i: SQL Tuning Workshop 13-11

 External Tables Performance
Characteristics

• External tables are read-only tables in which the

data is stored outside the database in flat files.
• The metadata for an external table is created
using a CREATE TABLE statement.
• With the help of external tables, Oracle data can
be stored outside the database as flat files.
• Specifying a PARALLEL clause is of value only when
dealing with large amounts of data. Otherwise, it is
not advisable to specify a PARALLEL clause, and
doing so can be detrimental.

Copyright © Oracle Corporation, 2002. All rights reserved.

External Tables Performance Characteristics

When you access the external table through a SQL statement, the fields of the external table can
be used just like any other field in a “normal” table. In particular, you can use the fields as
arguments for any SQL built-in function, PL/SQL function, or Java function. Thus you can
manipulate data from the external source. For data warehousing, you can do more sophisticated
transformations in this way than you can with simple data type conversions. You can also use this
mechanism in data warehousing to do data cleansing.
Parallel Access to External Tables
After the metadata for an external table is created, you can query the external data directly and in
parallel using SQL. As a result, the external table acts as a view, which you use to run any SQL
query against external data without loading the external data into the database. The degree of
parallel access to an external table is specified using standard parallel hints and with the parallel
clause. Using parallelism on an external table allows for concurrent access to the data files that
comprise an external table. Whether a single file is accessed concurrently or not is dependent
upon the access driver implementation and attributes of the data file(s) being accessed (for
example, record formats or seekable media).

Oracle9i: SQL Tuning Workshop 13-12

 External Tables Limitations

• The data can be queried using SQL but you cannot

use DML.
• Indexes cannot be created on external tables.
• The ANALYZE statement is not supported for
gathering statistics for external tables. The
DBMS_STATS package should be used for gathering
statistics for external tables.

Copyright © Oracle Corporation, 2002. All rights reserved.

External Table Limitations

• Whereas you can use external tables to access data stored in a file outside of the database,
you can not perform DML on that external data.
• Indexes are not supported on external tables.
• To gather statistics for an external table, use the DBMS_STATS package. The ANALYZE
command is not supported.

Oracle9i: SQL Tuning Workshop 13-13

 Why Use External Tables?

• Provides a valuable means for performing basic

extraction, transformation, and transportation (ETT)
tasks that are common for data warehousing
• Are transparent to users and applications
• Are a complement to the existing SQL*Loader
functionality
• Are especially useful for environments where:
– The complete external source must be joined with
existing database objects and transformed in a
complex manner.
– The external data volume is large and used only once.

Copyright © Oracle Corporation, 2002. All rights reserved.

When to Use External Tables

External tables provides a valuable means for performing basic extraction, transformation, and
transportation (ETT) tasks that are common for data warehousing. They are especially useful for
environments where the complete external source must be joined with existing database objects
and transformed in a complex manner, or the external data volume is large and used only once.

Oracle9i: SQL Tuning Workshop 13-14

 Creating External Tables

CREATE TABLE table-name

(column_definitions)
ORGANIZATION EXTERNAL
(TYPE ORACLE_LOADER
DEFAULT DIRECTORY dir_object_name
ACCESS PARAMETERS
(RECORDS DELIMITED BY newline
BADFILE bad_dir:’bad_file_name.log’
LOGFILE log_dir:’log_file_name.log’
FIELDS TERMINATED BY ’,’
MISSING FIELD VALUES ARE null
(column_names_in_file))
LOCATION (’file_name.dat’))
PARALLEL
REJECT LIMIT UNLIMITED;

Copyright © Oracle Corporation, 2002. All rights reserved.

Creating External Tables

The ORGANIZATION EXTERNAL clause of the CREATE TABLE statement specifies that you
create an external table. There are several clauses that identify where the external file is located
and how the external file is structured.
A directory object is required prior to using the CREATE TABLE ... ORGANIZATION
EXTERNAL statement. This directory object identifies the location of the external file on the
operating system.
Note: For details about how to create an external table, see the Oracle9i Database
Administrator’s Guide manual.
Instructor Note
The details of creating an external table are covered in the Introduction to Oracle9i: SQL course.
That course is a prerequisite, so do not get into the specifics here.

Oracle9i: SQL Tuning Workshop 13-15

 Retrieving External Tables Information

USER_EXTERNAL_TABLES USER_EXTERNAL_LOCATIONS

TABLE_NAME TABLE_NAME
TYPE_OWNER LOCATION
TYPE_NAME DIRECTORY_OWNER
DEFAULT_DIRECTORY_OWNER DIRECTORY_NAME
DEFAULT_DIRECTORY_NAME
REJECT_LIMIT
ACCESS_TYPE
ACCESS_PARAMETERS

Copyright © Oracle Corporation, 2002. All rights reserved.

Retrieving External Tables Information from the Data Dictionary

Use the following query to list the external tables and information related to their structure:

SQL> SELECT t.table_name

2 , t.type_name
3 , t.default_directory_name
4 , t.reject_limit
5 , t.access_type
6 , t.access_parameters
7 , l.location
8 , l.directory_name
9 FROM user_external_tables t
11 , user_external_locations l
10 WHERE t.table_name = l.table_name;

Oracle9i: SQL Tuning Workshop 13-16

 Summary

In this lesson, you should have learned how to:

• Create index-organized tables (IOTs)
– Identify IOT performance characteristics
– Identify IOT Limitations
– Control IOT row overflow
• Create external tables
– Identify external tables performance characteristics
– Identify external tables limitations

Copyright © Oracle Corporation, 2002. All rights reserved.

Summary
This lesson introduced you to alternative storage techniques including index-organized tables,
index clusters and hash clusters.
• An indexed-organized table (IOT) is like a regular table with an index on one or more of its
columns, but instead of maintaining two separate segments for the table and the B*-tree
index, the Oracle server maintains one single B*-tree structure which contains the primary
key value and other (non-key) column values for the row.
• An external table is a data source that is located outside of the Oracle database. After the
external table is set up using the CREATE TABLE ... ORGANIZATION EXTERNAL
syntax, you can query this data source using a SELECT statement.

Oracle9i: SQL Tuning Workshop 13-17

Oracle9i: SQL Tuning Workshop 13-18
 Data Warehousing Considerations

Copyright © Oracle Corporation, 2002. All rights reserved.

Schedule: Timing Topic

40 minutes Lecture
40 minutes Total
 Objectives

After completing this lesson, you should be able to do

the following:
• Create improved aggregation statements
• Simultaneously insert data into multiple tables
• Use the MERGE statement to conditionally update or
insert data

Copyright © Oracle Corporation, 2002. All rights reserved.

Objectives
Data warehousing applications often have requirements to extract data from multiple
sources, transform it to different formats, transport it to different platforms, and load it into
database tables. The Oracle9i database has a number of new features that streamline these
processes.
For more information, see Oracle9i Data Warehousing Guide and Oracle9i SQL Reference.

Oracle9i: SQL Tuning Workshop 14-2

 The WITH Clause: Overview

• Enables you to use the same query block in a

SELECT statement when it occurs more than once
within a complex query
• Retrieves the results of a query block and stores
them in the user’s temporary tablespace
• Can improve performance

Copyright © Oracle Corporation, 2002. All rights reserved.

The WITH Clause: Overview

Using the WITH clause, you can reuse the same query through materialization when it is
costly to evaluate the query block and it occurs more than once within a complex query.

Oracle9i: SQL Tuning Workshop 14-3

 The WITH Clause

• Name a query block in a SELECT statement to

reference it more than once within a query.
• Can hold multiple query blocks, separated by
commas
• Is resolved as in-line views or temporary tables

Copyright © Oracle Corporation, 2002. All rights reserved.

The WITH Clause

Use the WITH clause to define a query block before using it in a query. This can improve
performance, and it certainly improves readability and maintainability of your SQL
statements. You can also use the WITH clause to isolate the business question from the data
gathering.
A query name is visible to all WITH element query blocks (including their subquery blocks)
defined after it and the main query block itself (including its subquery blocks).
The WITH clause is internally resolved as an in-line view or a temporary table; the cost-
based optimizer chooses the appropriate resolution.
Note: The new WITH clause does not always offer a significant improvement in
performance. However it is an ease-of-use feature.

Oracle9i: SQL Tuning Workshop 14-4

 The WITH Clause: Example

Find all product subcategories whose average product

list price is above one-half the average product list
price of the whole company.
SQL> WITH
2 summary AS (
3 select prod_subcategory
4 , avg(prod_list_price) AS avg_subcat
5 from products
6 group by prod_subcategory)
7 SELECT prod_subcategory, avg_subcat
8 FROM summary
9 WHERE avg_subcat > (
10 SELECT AVG(avg_subcat) * 0.5
11 FROM summary)
12* ORDER BY avg_subcat DESC;

Copyright © Oracle Corporation, 2002. All rights reserved.

The WITH Clause: Example

The output from the query identifies product subcategories as having list prices one-half
above the average list price for the company.
The Autotrace output for the above query is as follows:

Statistics
-----------------------------------------------------
4 recursive calls
22 db block gets
720 consistent gets
20 physical reads
520 redo size
2250 bytes sent via SQL*Net to client
634 bytes received via SQL*Net from client
3 SQL*Net roundtrips to/from client
2 sorts (memory)
0 sorts (disk)
25 rows processed
Instructor Note
Demonstration script: demo14_01.sql and demo14_02.sql
Purpose: These scripts run the statements shown on pages 5 and 6 to demonstrate the WITH
clause statistics.

Oracle9i: SQL Tuning Workshop 14-5

 The WITH Clause: Implementation

Equivalent query, less powerful syntax:

SQL> SELECT prod_subcategory
2 , AVG(prod_list_price) AS avg_subcat
3 FROM products
4 GROUP BY prod_subcategory
5 HAVING AVG(prod_list_price) >
6 (SELECT AVG(prod_list_price) * 0.5
7 FROM products)
8 ORDER BY avg(prod_list_price) DESC;

Copyright © Oracle Corporation, 2002. All rights reserved.

The WITH Clause: Implementation

The query tries to retrieve all product subcategories whose average product list price is
above one-half for the average product list price of the products. In the query, there are two
blocks, one main query block and one subquery block. Both blocks must do some
aggregations.
Using the WITH clause, you can materialize the query block that does the GROUP BY
clause, and thus avoid summarizing the department total cost more than once.
The Autotrace output for the above query is as follows:
Statistics
-----------------------------------------------------
0 recursive calls
8 db block gets
1428 consistent gets
40 physical reads
0 redo size
2204 bytes sent via SQL*Net to client
634 bytes received via SQL*Net from client
3 SQL*Net roundtrips to/from client
2 sorts (memory)
0 sorts (disk)
24 rows processed

Oracle9i: SQL Tuning Workshop 14-6

 The WITH Clause: Usage Notes

• Use it only for SELECT statements.

• A query name is visible to all WITH element query
blocks defined after it and the main query block
itself.
• When the query name is the same as an existing
table name, the parser searches from the inside out.
The query block name takes precedence over the
table name.
• The WITH clause can hold more than one query.
Each query is separated by a comma.

Copyright © Oracle Corporation, 2002. All rights reserved.

The WITH Clause: Usage Notes

A query name is visible to all WITH element query blocks (including their subquery blocks)
defined after it and the main query block itself (including its subquery blocks).
A query name can be the same as some persistent table name or query name in the WITH list
of another query block. If this happens, the parser searches for the right definition inside out
(in terms of query block nesting). The innermost query name definition is used for
resolution. Users must make sure that it does not conflict with permanent tables if that is not
their intention. This is similar to the C language, in which a variable has priority when its
name is the same as some global variable.

Oracle9i: SQL Tuning Workshop 14-7

 The WITH Clause: Benefits

Using the WITH clause adds business value because:

• The business question is isolated from data
gathering.
• The query is easy to read.
• A clause is evaluated only once, even if it appears
multiple times in the query.
• Performance is improved because the instantiated
subquery clause must be calculated only once.

Copyright © Oracle Corporation, 2002. All rights reserved.

Oracle9i: SQL Tuning Workshop 14-8

 Multitable INSERT Statements

• Allow the INSERT...SELECT statement to insert

rows into multiple tables as part of a single DML
statement
• Can be used in data warehousing systems to
transfer data from one or more operational data
sources to a set of target tables
• Can be used for refreshing materialized views

Copyright © Oracle Corporation, 2002. All rights reserved.

Multitable INSERT Statements

The INSERT… SELECT statement with the new syntax can be parallelized and used with
the direct load mechanism. The multitable INSERT statement inserts computed rows
derived from the rows returned from the evaluation of a subquery. There are two forms of
the multitable INSERT statement: unconditional and conditional. For the unconditional
form, an INTO clause list is executed once for each row returned by the subquery. For the
conditional form, INTO clause lists are guarded by WHEN clauses that determine whether the
corresponding INTO clause list is executed.
An INTO clause list consists of one or more INTO clauses. The execution of an INTO
clause list causes the insertion of one row for each INTO clause in the list.
An INTO clause specifies the target into which a computed row is inserted. The target
specified can be any table expression that is legal for an INSERT... SELECT statement.
However aliases cannot be used. The same table can be specified as the target for more than
one INTO clause.
An INTO clause also provides the value of the row to be inserted using a VALUES clause.
An expression used in the VALUES clause can be any legal expression, but can refer only to
columns returned by the select list of the subquery. If the VALUES clause is omitted, the
select list of the subquery provides the values to be inserted. If a column list is given, each
column in the list is assigned a corresponding value from the VALUES clause or the
subquery. If no column list is given, the computed row must provide values for all columns
in the target table.

Oracle9i: SQL Tuning Workshop 14-9

 Multitable INSERT Statement

INSERT ALL INTO product_activity

VALUES(today, product_id, quantity)
INTO prod_sales
(prod_id, time_id, amount_sold,
quantity_sold)
VALUES(product_id, today, total, quantity)
SELECT order_date today, product_id,
SUM(unit_price) total,
SUM(quantity) quantity
FROM orders, order_items
WHERE orders.order_id = order_items.order_id
GROUP BY order_date, product_id;

1330 rows created.

Copyright © Oracle Corporation, 2002. All rights reserved.

Multitable INSERT Statement

The example in the slide inserts the values of order date, product_id and sum(quantity)
into the product_activity table and order date, product_id and sum(unit_price)
into the prod_sales table using a single statement. The values are obtained through a subquery
selecting from a join between the orders and order_items tables.
This is the alternative syntax using two individual INSERT statements.
INSERT INTO product_activity
SELECT order_date today, product_id, SUM(quantity) quantity
FROM orders, order_items
WHERE orders.order_id = order_items.order_id
GROUP BY order_date, product_id;
665 rows created.
INSERT INTO prod_sales
SELECT product_id, order_date today,
SUM(unit_price) total, SUM(quantity) quantity
FROM orders, order_items
WHERE orders.order_id = order_items.order_id
GROUP BY order_date, product_id;
665 rows created.

Oracle9i: SQL Tuning Workshop 14-10

Conditional Multitable Inserts
Conditional multitable inserts use the WHEN clause to specify conditions. This can further
be classified into:
• Conditional FIRST: The first true condition will be inserted.
• Conditional ALL: All true conditions will be inserted.
Conditional FIRST has the following signature:
INSERT FIRST
WHEN <condition> into <table>(col, col,…)
WHEN <condition> into <table>(col, col,…)
ELSE <table>(col, col,...)
SELECT...;
Conditional ALL has the following signature:
INSERT ALL
WHEN <condition> into <table>(col, col,…)
WHEN <condition> into <table>(col, col,…)
ELSE <table>(col, col,...)
SELECT...;

Oracle9i: SQL Tuning Workshop 14-11

 Advantages of Multitable INSERTs

• Eliminates the need for multiple INSERT...

SELECT statements to populate multiple tables
• Eliminates the need for a procedure to do multiple
inserts using conditional logic
• Improves performance significantly over the above
two methods due to the elimination of the cost of
materialization and repeated scans on the source
data

Copyright © Oracle Corporation, 2002. All rights reserved.

Oracle9i: SQL Tuning Workshop 14-12

 MERGE Statements

• Provide the ability to “upsert,” that is, conditionally

update or insert into the database
• Do an update if the row exists, and an insert if it is a
new row
• Avoid multiple updates
• Can be used in data warehousing applications

Copyright © Oracle Corporation, 2002. All rights reserved.

Applications of MERGE statements

• MERGE statements use a single SQL statement to complete an UPDATE or INSERT or
both.
• The statement can be parallelized transparently.
• The array interface can be used for bulk operations.
• Performance is improved because fewer statements require fewer scans of the source
tables.

Oracle9i: SQL Tuning Workshop 14-13

 Merge Syntax

Copyright © Oracle Corporation, 2002. All rights reserved.

Merge Syntax
INTO Clause
Use the INTO clause to specify the target table you are updating or inserting into.
USING Clause
Use the USING clause to specify the source of the data to be updated or inserted. The source
can be a table, view, or the result of a subquery.
ON Clause
Use the ON clause to specify the condition upon which the MERGE operation either updates
or inserts. For each row in the target table for which the search condition is true, Oracle
updates the row based with corresponding data from the source table. If the condition is not
true for any rows,
Oracle inserts into the target table based on the corresponding source table row.
WHEN MATCHED | NOT MATCHED
Use these clauses to instruct Oracle how to respond to the results of the
join_condition. You can specify these two clauses in either order.

Oracle9i: SQL Tuning Workshop 14-14

 Example of Using the MERGE Statement in
Data Warehousing

MERGE INTO customer C USING cust_src s

ON (c.customer_id = s.src_customer_id)
WHEN MATCHED THEN
UPDATE SET c.cust_address = s.cust_address
WHEN NOT MATCHED THEN
INSERT (customer_id, cust_first_name, ...)
VALUES (src_customer_id, src_first_name,
...);

Copyright © Oracle Corporation, 2002. All rights reserved.

Example of MERGE Statement

This an example of using the MERGE statement in data warehousing. customer is a large
fact table and cust_src is a smaller table with rows that must be inserted into customer
conditionally. A table such as the cust_src table is sometimes said to contain the “deltas”
or changes that in this case need to be applied to the other table.
This MERGE statement indicates that the customer table must be merged with the rows
returned from the evaluation of the ON clause. The ON clause in this case is the cust_src
table, but it can be an arbitrary query. Each row from the cust_src table is checked for a
match to any row in the customer table by satisfying the join condition specified by the
ON clause. If so, each row in the customer table is updated using the UPDATE SET
clause of the MERGE statement. If no such row exists in the customer table, then the rows
are inserted into the customer table.

Oracle9i: SQL Tuning Workshop 14-15

 Summary

In this lesson, you should have learned how to:

• Use the new WITH clause
• Simultaneously insert data into multiple tables
• Use the MERGE statement to conditionally update or
insert data

Copyright © Oracle Corporation, 2002. All rights reserved.

Summary
In this lesson, you should have learned about some techniques used in data warehousing that
help to transform, transport, and load data.

Oracle9i: SQL Tuning Workshop 14-16