ERserver
iSeries
DB2 Universal Database for iSeries -
Database Performance and Query Optimization


ERserver
iSeries
DB2 Universal Database for iSeries -
Database Performance and Query Optimization
© Copyright International Business Machines Corporation 2000, 2001. All rights reserved.
US Government Users Restricted Rights – Use, duplication or disclosure restricted by GSA ADP Schedule Contract
with IBM Corp.
 Contents
About DB2 UDB for iSeries Database Performance and Query Optimization . . . . . . . . . vii
Who should read the Database Performance and Query Optimization book . . . . . . . . . . . vii
Assumptions relating to SQL statement examples . . . . . . . . . . . . . . . . . . . viii
How to interpret syntax diagrams . . . . . . . . . . . . . . . . . . . . . . . . . viii
What’s new for V5R1 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . ix

Chapter 1. Database performance and query optimization: Overview . . . . . . . . . . . . 1

| Creating queries . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2

Chapter 2. Data access on DB2 UDB for iSeries: data access paths and methods . . . . . . . 3
Table scan . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3
Index . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3
Encoded vector index . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3
Data access: data access methods . . . . . . . . . . . . . . . . . . . . . . . . . . 3
Data access methods: Summary . . . . . . . . . . . . . . . . . . . . . . . . . . 5
Ordering query results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7
Enabling parallel processing for queries. . . . . . . . . . . . . . . . . . . . . . . . 7
Spreading data automatically. . . . . . . . . . . . . . . . . . . . . . . . . . . . 8
Table scan access method . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8
Parallel table prefetch access method . . . . . . . . . . . . . . . . . . . . . . . . 10
Parallel table scan method . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11
Index scan-key selection access method . . . . . . . . . . . . . . . . . . . . . . . 13
Parallel index scan-key selection access method (available only when the DB2 UDB Symmetric
Multiprocessing feature is installed) . . . . . . . . . . . . . . . . . . . . . . . . 14
Index scan-key positioning access method . . . . . . . . . . . . . . . . . . . . . . 15
Parallel index scan-key positioning access method (available only when the DB2 UDB Symmetric
Multiprocessing feature is installed) . . . . . . . . . . . . . . . . . . . . . . . . 19
Index Only Access Method . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21
Parallel table or index based preload access method . . . . . . . . . . . . . . . . . . 22
Index-from-index access method . . . . . . . . . . . . . . . . . . . . . . . . . . 22
Hashing access method . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 23
Bitmap processing method . . . . . . . . . . . . . . . . . . . . . . . . . . . . 24
| Sort access method . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 28

Chapter 3. The DB2 UDB for iSeries query optimizer: Overview . . . . . . . . . . . . . . 31

How the query optimizer makes your queries more efficient . . . . . . . . . . . . . . . . . 31
Optimizer decision-making rules . . . . . . . . . . . . . . . . . . . . . . . . . . 31
Cost estimation for queries . . . . . . . . . . . . . . . . . . . . . . . . . . . . 32
General query optimization tips . . . . . . . . . . . . . . . . . . . . . . . . . . 34
Access plan validation. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 34
Join optimization . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 35
Grouping optimization . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 50
Ordering optimization . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 54
View implementation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 55

Chapter 4. Optimizing query performance using query optimization tools . . . . . . . . . . 59

Verify the performance of SQL applications . . . . . . . . . . . . . . . . . . . . . . . 60
Examine query optimizer debug messages in the job log . . . . . . . . . . . . . . . . . . 60
Query optimization performance information messages . . . . . . . . . . . . . . . . . 61
Query optimization performance information messages and open data paths . . . . . . . . . 66
Gather information about embedded SQL statements with the PRTSQLINF command . . . . . . . 67
Gather statistics about your queries with the database monitor . . . . . . . . . . . . . . . . 69
Start Database Monitor (STRDBMON) command . . . . . . . . . . . . . . . . . . . . 70

© Copyright IBM Corp. 2000, 2001 iii

 End Database Monitor (ENDDBMON) command . . . . . . . . . . . . . . . . . . . . 71
Database monitor performance rows . . . . . . . . . . . . . . . . . . . . . . . . 71
Query optimizer index advisor . . . . . . . . . . . . . . . . . . . . . . . . . . . 72
Database monitor examples . . . . . . . . . . . . . . . . . . . . . . . . . . . 73
Gather statistics about your queries with memory-resident database monitor APIs. . . . . . . . . 78
Memory-resident database monitor external API description . . . . . . . . . . . . . . . . 79
Memory-resident database monitor external table description . . . . . . . . . . . . . . . 79
Sample SQL queries . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 80
Memory-resident database monitor row identification . . . . . . . . . . . . . . . . . . 80
View the effectiveness of your queries with Visual Explain . . . . . . . . . . . . . . . . . 80
Change the attributes of your queries with the Change Query Attributes (CHGQRYA) command . . . 81
Control queries dynamically with the query options file QAQQINI . . . . . . . . . . . . . . . 82
Specifying the QAQQINI file . . . . . . . . . . . . . . . . . . . . . . . . . . . 82
Creating the QAQQINI query options file . . . . . . . . . . . . . . . . . . . . . . . 83
Control long-running queries with the DB2 UDB for iSeries Predictive Query Governor . . . . . . . 89
Cancelling a query . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 90
Query governor implementation considerations . . . . . . . . . . . . . . . . . . . . 90
Query governor considerations for user applications: Setting the time limit . . . . . . . . . . 90
| Controlling the default reply to the query governor inquiry message . . . . . . . . . . . . . 90
Testing performance with the query governor . . . . . . . . . . . . . . . . . . . . . 91
Examples of setting query time limits . . . . . . . . . . . . . . . . . . . . . . . . 91
Control parallel processing for queries . . . . . . . . . . . . . . . . . . . . . . . . . 92
Controlling system wide parallel processing for queries . . . . . . . . . . . . . . . . . 92
Controlling job level parallel processing for queries . . . . . . . . . . . . . . . . . . . 93
Query optimization tools: Comparison table . . . . . . . . . . . . . . . . . . . . . . . 95

Chapter 5. Using indexes to speed access to large tables . . . . . . . . . . . . . . . . 97

Coding for effective indexes: Avoid numeric conversions . . . . . . . . . . . . . . . . . . 97
Coding for effective indexes: Avoid arithmetic expressions . . . . . . . . . . . . . . . . . 98
Coding for effective indexes: Avoid character string padding . . . . . . . . . . . . . . . . . 98
Coding for effective indexes: Avoid the use of like patterns beginning with % or _ . . . . . . . . . 98
Coding for effective indexes: Be aware of the instances where DB2 UDB for iSeries does not use an
index . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 99
Coding for effective indexes: Using indexes with sort sequence . . . . . . . . . . . . . . . 100
Coding for effective indexes: Using indexes and sort sequence with selection, joins, or grouping 100
Coding for effective indexes: Ordering . . . . . . . . . . . . . . . . . . . . . . . 100
Examples of indexes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 101
Index example: Equals selection with no sort sequence table . . . . . . . . . . . . . . . 101
Index example: Equals selection with a unique-weight sort sequence table . . . . . . . . . . 101
Index example: Equal selection with a shared-weight sort sequence table . . . . . . . . . . 101
Index example: Greater than selection with a unique-weight sort sequence table. . . . . . . . 102
Index example: Join selection with a unique-weight sort sequence table . . . . . . . . . . . 102
Index example: Join selection with a shared-weight sort sequence table . . . . . . . . . . . 102
Index example: Ordering with no sort sequence table . . . . . . . . . . . . . . . . . . 102
Index example: Ordering with a unique-weight sort sequence table . . . . . . . . . . . . . 103
Index example: Ordering with a shared-weight sort sequence table. . . . . . . . . . . . . 103
Index example: Ordering with ALWCPYDTA(*OPTIMIZE) and a unique-weight sort sequence table 103
Index example: Grouping with no sort sequence table . . . . . . . . . . . . . . . . . 104
Index example: Grouping with a unique-weight sort sequence table . . . . . . . . . . . . 104
Index example: Grouping with a shared-weight sort sequence table . . . . . . . . . . . . 104
Index example: Ordering and grouping on the same columns with a unique-weight sort sequence
table . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 104
Index example: Ordering and grouping on the same columns with ALWCPYDTA(*OPTIMIZE) and a
unique-weight sort sequence table . . . . . . . . . . . . . . . . . . . . . . . . 105
Index example: Ordering and grouping on the same columns with a shared-weight sort sequence
table . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 105

iv DB2 UDB for iSeries Database Performance and Query Optimization V5R1
 Index example: Ordering and grouping on the same columns with ALWCPYDTA(*OPTIMIZE) and a
shared-weight sort sequence table . . . . . . . . . . . . . . . . . . . . . . . . 106
Index example: Ordering and grouping on different columns with a unique-weight sort sequence
table . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 106
Index example: Ordering and grouping on different columns with ALWCPYDTA(*OPTIMIZE) and a
unique-weight sort sequence table . . . . . . . . . . . . . . . . . . . . . . . . 106
Index example: Ordering and grouping on different columns with ALWCPYDTA(*OPTIMIZE) and a
shared-weight sort sequence table . . . . . . . . . . . . . . . . . . . . . . . . 107
What are encoded vector indexes? . . . . . . . . . . . . . . . . . . . . . . . . 107

Chapter 6. Application design tips for database performance . . . . . . . . . . . . . . 111

Database application design tips: Use structure parameter passing techniques . . . . . . . . . 111
Database application design tips: Background information on parameter passing . . . . . . . . 111
Database application design tips: Differences in structure parameter passing techniques . . . . . 112
Database application design tips: Use live data . . . . . . . . . . . . . . . . . . . . . 112
Database application design tips: Reduce the number of open operations . . . . . . . . . . . 114
Database application design tips: Retain cursor positions . . . . . . . . . . . . . . . . . 116
Database application design tips: Retaining cursor positions for non-ILE program calls . . . . . 116
Database application design tips: Retaining cursor positions across ILE program calls . . . . . . 116
Database application design tips: General rules for retaining cursor positions for all program calls 117

Chapter 7. Programming techniques for database performance . . . . . . . . . . . . . 119

Programming techniques for database performance: Use the OPTIMIZE clause . . . . . . . . . 120
Programming techniques for database performance: Use FETCH FOR n ROWS. . . . . . . . . 120
Programming techniques for database performance: Improve SQL blocking performance when
using FETCH FOR n ROWS . . . . . . . . . . . . . . . . . . . . . . . . . . 121
Programming techniques for database performance: Use INSERT n ROWS . . . . . . . . . . 121
Programming techniques for database performance: Control database manager blocking . . . . . 121
Programming techniques for database performance: Optimize the number of columns that are
selected with SELECT statements . . . . . . . . . . . . . . . . . . . . . . . . . 122
Programming techniques for database performance: Eliminate redundant validation with SQL
PREPARE statements . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 123
Programming techniques for database performance: Page interactively displayed data with
REFRESH(*FORWARD) . . . . . . . . . . . . . . . . . . . . . . . . . . . . 123

Chapter 8. General DB2 UDB for iSeries performance considerations . . . . . . . . . . . 125

Effects on database performance when using long object names . . . . . . . . . . . . . . 125
Effects of precompile options on database performance . . . . . . . . . . . . . . . . . . 125
Effects of the ALWCPYDTA parameter on database performance . . . . . . . . . . . . . . 126
Tips for using VARCHAR and VARGRAPHIC data types in databases . . . . . . . . . . . . 127

Appendix A. Database monitor: DDS . . . . . . . . . . . . . . . . . . . . . . . . 129

Database monitor physical file DDS . . . . . . . . . . . . . . . . . . . . . . . . . 129
Optional database monitor logical file DDS. . . . . . . . . . . . . . . . . . . . . . . 136
Database monitor logical table 1000 - Summary Row for SQL Information . . . . . . . . . . 136
Database monitor logical table 3000 - Summary Row for Table Scan . . . . . . . . . . . . 147
Database monitor logical table 3001 - Summary Row for Index Used . . . . . . . . . . . . 151
Database monitor logical table 3002 - Summary Row for Index Created . . . . . . . . . . . 157
Database monitor logical table 3003 - Summary Row for Query Sort . . . . . . . . . . . . 164
Database monitor logical table 3004 - Summary Row for Temp Table . . . . . . . . . . . . 168
Database monitor logical table 3005 - Summary Row for Table Locked . . . . . . . . . . . 173
Database monitor logical table 3006 - Summary Row for Access Plan Rebuilt . . . . . . . . . 176
Database monitor logical table 3007 - Summary Row for Optimizer Timed Out . . . . . . . . 180
Database monitor logical table 3008 - Summary Row for Subquery Processing . . . . . . . . 183
Database monitor logical table 3010 - Summary for HostVar & ODP Implementation . . . . . . 185
Database monitor logical table 3014 - Summary Row for Generic QQ Information . . . . . . . 186

Contents v
 Database monitor logical table 3018 - Summary Row for STRDBMON/ENDDBMON . . . . . . 192
Database monitor logical table 3019 - Detail Row for Rows Retrieved . . . . . . . . . . . . 193
Database monitor logical table 3021 - Summary Row for Bitmap Created . . . . . . . . . . 195
Database monitor logical table 3022 - Summary Row for Bitmap Merge . . . . . . . . . . . 198
Database monitor logical table 3023 - Summary for Temp Hash Table Created . . . . . . . . 201
Database monitor logical table 3025 - Summary Row for Distinct Processing . . . . . . . . . 204
Database monitor logical table 3027 - Summary Row for Subquery Merge . . . . . . . . . . 206
Database monitor logical table 3028 - Summary Row for Grouping . . . . . . . . . . . . . 209

Appendix B. Memory Resident Database Monitor: DDS . . . . . . . . . . . . . . . . . 215

External table description (QAQQQRYI) . . . . . . . . . . . . . . . . . . . . . . . . 215
External table description (QAQQTEXT). . . . . . . . . . . . . . . . . . . . . . . . 219
External table description (QAQQ3000) . . . . . . . . . . . . . . . . . . . . . . . . 219
External table description (QAQQ3001) . . . . . . . . . . . . . . . . . . . . . . . . 221
External table description (QAQQ3002) . . . . . . . . . . . . . . . . . . . . . . . . 223
External table description (QAQQ3003) . . . . . . . . . . . . . . . . . . . . . . . . 225
External table description (QAQQ3004) . . . . . . . . . . . . . . . . . . . . . . . . 226
External table description (QAQQ3007) . . . . . . . . . . . . . . . . . . . . . . . . 228
External table description (QAQQ3008) . . . . . . . . . . . . . . . . . . . . . . . . 229
External table description (QAQQ3010) . . . . . . . . . . . . . . . . . . . . . . . . 229

Index . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 231

vi DB2 UDB for iSeries Database Performance and Query Optimization V5R1
 About DB2 UDB for iSeries Database Performance and Query
Optimization
This book explains to programmers and database administrators:
v How to use the tools and functions that are available in DB2 UDB for iSeries for getting the best
performance out of your database applications
v How to run queries that make full use of the capabilities of the DB2 UDB for iSeries integrated
database.

For more information on DB2 UDB for iSeries guidelines and examples for implementation in an
application programming environment, see the following information in the Database and Files Systems
category of the iSeries Information Center:
| v SQL Reference
| v SQL Programming Concepts
| v SQL Programming with Host Languages
| v SQL Call Level Interfaces (ODBC)
| v Database Programming
| v Query/400 Use
| v ODBC
| v SQLJ
Java Database Connectivity (JDBC)information can be found in the IBM Developer Kit for Java under
Programming in the iSeries Information Center.

For additional information on advanced database functions, see the DATABASE 2/400 Advanced Database
Functions book, GG24-4249.

Who should read the Database Performance and Query Optimization

book
| This information is for programmers and database administrators who understand basic database
| applications and want to understand how to tune queries. This information shows how to use the available
| tools for debugging query performance problems.

| You should be familiar with languages and interfaces, including the following:
| v COBOL for iSeries
| v ILE COBOL for iSeries
| v iSeries PL/I
| v ILE C for iSeries
| v ILE C++
| v VisualAge C++ for iSeries
| v REXX
| v RPG III (part of RPG for iSeries)
| v ILE RPG for iSeries
| v Query/400
| v The OPNQRYF command
| v Call level interfaces (CLI)
| v ODBC

© Copyright IBM Corp. 2000, 2001 vii

| v JDBC

Assumptions relating to SQL statement examples

| The examples of queries that are shown in this book are based on the sample tables in Appendix A, ″DB2
| UDB for iSeries Sample Tables,″ of the SQL Programming Concepts book. For the SQL examples,
| assume the following:
v They are shown in the interactive SQL environment or they are written in ILE C or in COBOL. EXEC
SQL and END-EXEC are used to delimit an SQL statement in a COBOL program. A description of how
to use SQL statements in a COBOL program is provided in ″Coding SQL Statements in COBOL
Applications″ of the SQL Programming with Host Languages book. A description of how to use SQL
statements in an ILE C program is provided in ″Coding SQL Statements in C Applications″ of the SQL
Programming with Host Languages book.
v Each SQL example is shown on several lines, with each clause of the statement on a separate line.
v SQL keywords are highlighted.
v Table names provided in Appendix A, ″DB2 UDB for iSeries Sample Tables″ of the SQL Programming
Concepts use the collection CORPDATA. Table names that are not found in Appendix A, ″DB2 UDB for
iSeries Sample Tables,″ should use collections you create.
v Calculated columns are enclosed in parentheses, (), and brackets, [].
v The SQL naming convention is used.
v The APOST and APOSTSQL precompiler options are assumed although they are not the default options
in COBOL. Character string constants within SQL and host language statements are delimited by
apostrophes (’).
v A sort sequence of *HEX is used, unless otherwise noted.
v The complete syntax of the SQL statement is usually not shown in any one example. For the complete
description and syntax of any of the statements described in this guide, see the SQL Reference.

Whenever the examples vary from these assumptions, it is stated.

Because this guide is for the application programmer, most of the examples are shown as if they were
written in an application program. However, many examples can be slightly changed and run interactively
by using interactive SQL. The syntax of an SQL statement, when using interactive SQL, differs slightly
from the format of the same statement when it is embedded in a program.

How to interpret syntax diagrams

Throughout this book, syntax is described using the structure defined as follows:
v Read the syntax diagrams from left to right, from top to bottom, following the path of the line.
The

─── symbol indicates the beginning of a statement.
The ───
symbol indicates that the statement syntax is continued on the next line.
The
─── symbol indicates that a statement is continued from the previous line.
The ───
 symbol indicates the end of a statement.
Diagrams of syntactical units other than complete statements start with the
─── symbol and end with
the ───
symbol.
v Required items appear on the horizontal line (the main path).


required_item


v Optional items appear below the main path.

required_item

optional_item

If an optional item appears above the main path, that item has no effect on the execution of the
statement and is used only for readability.

viii DB2 UDB for iSeries Database Performance and Query Optimization V5R1
 optional_item


required_item


v If you can choose from two or more items, they appear vertically, in a stack.
If you must choose one of the items, one item of the stack appears on the main path.


required_item required_choice1

required_choice2

If choosing one of the items is optional, the entire stack appears below the main path.


required_item

optional_choice1
optional_choice2

If one of the items is the default, it will appear above the main path and the remaining choices will be
shown below.
default_choice


required_item

optional_choice
optional_choice

v An arrow returning to the left, above the main line, indicates an item that can be repeated.

required_item  repeatable_item


If the repeat arrow contains a comma, you must separate repeated items with a comma.
,

required_item  repeatable_item


A repeat arrow above a stack indicates that you can repeat the items in the stack.
v Keywords appear in uppercase (for example, FROM). They must be spelled exactly as shown. Variables
appear in all lowercase letters (for example, column-name). They represent user-supplied names or
values.
v If punctuation marks, parentheses, arithmetic operators, or other such symbols are shown, you must
enter them as part of the syntax.

What’s new for V5R1

The major new features covered in this book include:
v The Database Monitor DDS for physical and logical files have been updated for Visual Explain.
v New QAQQINI options have been added.
v Encoded vector indexes have been updated.
v Partial join re-ordering has been updated.
v Expressions for Like Pattern and Escape Character have been added.

About DB2 UDB for iSeries Database Performance and Query Optimization ix
x DB2 UDB for iSeries Database Performance and Query Optimization V5R1
 Chapter 1. Database performance and query optimization:
Overview
| The goal of database performance tuning is to minimize the response time of your queries and to make
| the best use of your system’s resources by minimizing network traffic, disk I/O, and CPU time. This goal
| can only be achieved by understanding the logical and physical structure of your data, understanding the
| applications used on your system, and understanding how the many conflicting uses of your database may
| impact database performance.

| The best way to avoid performance problems is to ensure that performance issues are part of your
| ongoing development activities. Many of the most significant performance improvements are realized
| through careful design at the beginning of the database development cycle. To most effectively optimize
| performance, you must identify the areas that will yield the largest performance increases over the widest
| variety of situations and focus your analysis on those areas.

| Understanding access paths and the query optimizer

| Since iSeries automatically manages many hardware resources automatically, and uses a cost-based
| optimization formula to determine the most efficient access plan for running an SQL statement, it is
| important to know how the system determines the most efficient access method and what factors
| determine their selection by the system. These topics are covered in Data access methods. In addition, a
| clear understanding of the iSeries query optimizer will also help you design queries that leverage the query
| optimizer’s cost estimation and decision-making rules.

| Improving your queries

| Once you are familiar with these concepts, you can incrementally improve your queries by reviewing the
| material found in the following topics:
|| Topic Description
| Optimizing query performance using Describes how you can use query optimization tools to improve data retrieval
| query optimization tools times by gathering statistics about your queries or controlling the processing of
| your queries. With the results that these tools provide, you can then change the
| data access method chosen by the system or create the correct indexes and
| use them effectively.
| Using indexes to speed access to Describes the index-based retrieval method for accessing tables and how to
| large tables create effective indexes by avoiding such things as numeric conversions,
| arithmetic expressions, character string padding, and the use of like patterns.
| Increasing database performance Describes how the correct design of user applications can improve performance.
| through application design Application design considerations include parameter passing techniques, using
| live data, reducing the number of open operations, and retaining cursor
| positions.
| Improving database performance Describes how the correct programming techniques can improve performance.
| using programming techniques Among the techniques covered are: using the OPTIMIZE clause, using FETCH
| nROWS, using INSERT n ROWS, controlling the database manager blocking,
| optimizing the number of columns selected with SELECT statements, eliminating
| redundant validation, and paging interactively displayed data.
| General iSeries performance Describes some general system considerations and how they affect the
| performance of your queries.
|

© Copyright IBM Corp. 2000, 2001 1

|
| Creating queries
| You can create queries through any of the following interfaces:
| v SQL
| v Open Query File (OPNQRYF) command
| v Open database connectivity (ODBC)
| v Query/400
| v Java Database Connectivity (JDBC)
| v Call level interface (CLI)
| The query optimizer optimizes all of the queries that you create using these interfaces.

2 DB2 UDB for iSeries Database Performance and Query Optimization V5R1
 Chapter 2. Data access on DB2 UDB for iSeries: data access
paths and methods
This section introduces the data access methods that DB2 Universal Database for iSeries and the
Licensed Internal Code use to process queries and access data. The data access methods are grouped
into nonkeyed, keyed, and temporary result file access methods.

The iSeries fundamentally uses two methods to retrieve data specified in a query; through an index (keyed
access methods) or directly through the table (nonkeyed access methods). These access methods can be
combined in many ways to retrieve data. A data access method can be a table scan, an index, a
combination of both, or an encoded vector index.

Table scan
| A table scan, or arrival sequence, uses the order of rows as they are stored in the table to locate data that
| is specified in a query. Processing tables using the table scan is similar to processing sequential or direct
| files on traditional systems.

Index
An index, or keyed sequence access path, provides access to a table that is arranged according to the
contents of key columns. The keyed sequence is the order in which rows are retrieved. The access path is
automatically maintained whenever rows are added to or deleted from the table, or whenever the contents
of the index columns are changed. The best examples of an index is an index that is created with the
CREATE INDEX statement, or a keyed logical file that is created with the CRTLF command.

Columns that are good candidates for creating indexes are:

v Columns that are frequently referenced in row selection predicates.
v Columns that are frequently referenced in grouping or ordering.
v Columns that are used to join tables (see “Join optimization” on page 35).

Encoded vector index

An encoded vector index provides access to a database table by assigning codes to distinct key values
and then representing these values in an array. The elements of the array can be 1, 2, or 4 bytes in
length, depending on the number of distinct values that must be represented. Because of their compact
size and relative simplicity, encoded vector indexes provide for faster scans that can be more easily
processed in parallel.

| You create encoded vector indexes by using the CREATE ENCODED VECTOR INDEX statement. See
| What are encoded vector indexes for information on the use and maintenance of encoded vector indexes.

| For additional information about accelerating your queries with encoded vector indexes , go to the DB2
| Universal Database for iSeries web pages.

Data access: data access methods

The Licensed Internal Code and DB2 Universal Database for iSeries share the work on access methods.
The Licensed Internal Code does the low-level processing which includes selection, join functions,
hashing, and index creation.

© Copyright IBM Corp. 2000, 2001 3

The query optimization process chooses the most efficient access method for each query and keeps this
information in the access plan. The type of access is dependent on the number of rows, the expected
number of page faults 1, and other criteria.

You can use the tools and tips that are described later in this book to influence the way in which the query
optimizer implements your queries.

The optimizer uses any of the following methods to retrieve data. See “Data access methods: Summary”
on page 5 for a summary of these methods:
v Table scan method (a dataspace is an internal object that contains the data in a table)
v Parallel table prefetch method
v Index scan-key selection method
v Index scan-key positioning method
v Parallel table or index preload
v Index-from-index method
v Index only access method
v Hashing method
v Bitmap processing method
v Sort access method

Accessing data with the DB2 UDB Symmetric Multiprocessing methods

The DB2 UDB Symmetric Multiprocessing feature provides the optimizer with additional methods for
retrieving data that include parallel processing.

Symmetrical multiprocessing (SMP) is a form of parallelism achieved on a single system where multiple
processors (CPU and I/O processors) that share memory and disk resource work simultaneously towards
achieving a single end result. This parallel processing means that the database manager can have more
than one (or all) of the system processors working on a single query simultaneously. The performance of a
CPU bound query can be significantly improved with this feature on multiple-processor systems by
distributing the processor load across more than one processor on the system.

The following methods are available to the optimizer once the DB2 UDB Symmetric Multiprocessing
feature has been installed on your system:
v Parallel table scan method
v Parallel index scan-key selection method
v Parallel index scan-key positioning method
v Parallel index only access method
v Parallel hashing method
v Parallel bitmap processing method

Additional considerations:

The following topics provide additional background information on the access methods:
v “Ordering query results” on page 7
v “Enabling parallel processing for queries” on page 7
v “Spreading data automatically” on page 8

1. An interrupt that occurs when a program refers to a 4K-byte page that is not in main storage.

4 DB2 UDB for iSeries Database Performance and Query Optimization V5R1
Data access methods: Summary
The following table provides a summary of data management methods that are discussed in this book.
Table 1. Summary of data access methods
Selection
Access Method Process Good When Not Good When Selected When Advantages
“Table scan Reads all rows. Approx. > 20% Approx. < 20% No ordering, Minimizes page
access method” Selection criteria rows selected. rows selected. grouping, or joining I/O through
on page 8 applied to data in and approx. > 20% pre-fetching.
dataspace. rows selected.
“Parallel table Data retrieved Approx. > 20% Approx. < 20% No ordering, Minimizes wait
prefetch access from auxiliary rows selected. rows selected. grouping, or joining time for page
method” on storage in parallel 1. Adequate Query is CPU and approx. > 20% I/O through
page 10 streams. Reads all active memory bound. rows selected, and parallel table
rows. Selection available. the system job has prefetching.
criteria applied to been configured to
2. Query would
data in dataspace. take advantage of
otherwise be
I/O parallelism.
I/O bound.
3. Data spread
across multiple
disk units.
“Parallel table Data read and Approx. > 10% Approx. < 10% 1. DB2 UDB Significant
scan method” selected in parallel rows selected, rows selected. Symmetric performance
on page 11 tasks. large table. Query is I/O Multiprocessing especially on
1. Adequate bound on a installed. multiprocessors.
active memory uniprocessor 2. CPU bound or
available. system.
running on a
2. Data spread multiprocessor
across multiple system.
disk units.
3. DB2 UDB
Symmetric
Multiprocessing
installed.
4. Multi-processor
system.
“Index scan-key Selection criteria Ordering, Large number of Index is required and Dataspace
selection access applied to index. grouping, and rows selected. cannot use index accessed only
method” on joining. scan-key positioning for rows
page 13 method. matching index
scan-key
selection
criteria.
“Parallel index Selection criteria Size of index is Large number of When ordering of Better I/O
scan-key applied to index in much less than rows selected. results not required. overlap because
selection access parallel tasks. the dataspace. parallel tasks
method DB2 UDB perform the I/O.
(available only Symmetric Can fully utilize
when the DB2 Multiprocessing multiprocessor
UDB Symmetric must be installed. systems.
Multiprocessing
feature is
installed)” on
page 14

Chapter 2. Data access on DB2 UDB for iSeries: data access paths and methods 5
Table 1. Summary of data access methods (continued)
Selection
Access Method Process Good When Not Good When Selected When Advantages
“Index scan-key Selection criteria Approx. < 20% Approx. > 20% Selection columns Index and
positioning applied to range of rows selected. rows selected. match left-most keys dataspace
access method” index entries. and approx. < 20% accessed only
on page 15 Commonly used rows selected. for rows
option. matching
selection
criteria.
“Parallel index Selection criteria Approx. < 20% Large number of 1. When ordering of 1. Index and
scan-key applied to range of rows selected. rows selected. results not dataspace
positioning index entries in DB2 UDB required. accessed
access method parallel tasks. Symmetric only for rows
2. Selection
(available only Multiprocessing matching
columns match
when the DB2 must be installed. selection
left-most keys
UDB Symmetric criteria.
and approx. <
Multiprocessing
20% rows 2. Better I/O
feature is
selected. overlap
installed)” on
because
page 19
parallel
tasks
perform the
I/O.
3. Can fully
utilize a
multiprocessor
systems.
“Index-from- Key row Ordering, grouping Approx. > 20% No existing index to Index and
index access positioning on and joining. rows selected. satisfy ordering but dataspace
method” on permanent index. existing index does accessed only
page 22 Builds temporary satisfy selection and for rows
index over selecting approx. < matching
selected index 20% rows. selection
entries. criteria.
“Sort access Order data read Approx. > 20% Approx. < 20% Ordering specified; See table scan
method” on using table scan rows selected or rows selected or either no index and index
page 28 processing or large result set of small result set of exists to satisfy the scan-key
index scan-key rows. rows. ordering or a large positioning in
positioning. result set is this table.
expected.
“Index Only Done in All columns used Approx. < 20% All columns used in Reduced I/O to
Access Method” combination with in the query exist rows selected or the query exist as the dataspace.
on page 21 any of the other as key columns. small result set of key columns.
index access rows.
methods
“Parallel table or Index or table data Excessive random Active memory is Excessive random Random page
index based loaded in parallel activity would already activity would result I/O is avoided
preload access to avoid random otherwise occur overcommitted. from processing the which can
method” on access. against the object query and active improve I/O
page 22 and active memory is available bound queries.
memory is which can hold the
available to hold entire object.
the entire object.

6 DB2 UDB for iSeries Database Performance and Query Optimization V5R1
Table 1. Summary of data access methods (continued)
Selection
Access Method Process Good When Not Good When Selected When Advantages
“Hashing access Rows with Longer running Short running Join or grouping Reduces
method” on common values grouping and join queries. specified. random I/O
page 23(Parallel are correlate data queries. when compared
or non-parallel) with a common to index
value together. methods. If DB2
UDB Symmetric
Multiprocessing
is installed,
possible
exploitation of
SMP
parallelism.
“Bitmap Key position/index Selection can be Approx. >25% Indexes match Reduces page
processing scan-key selection applied to index rows selected. selection criteria. I/O to the data
method” on used to build and either approx. space. Allows
page 24 bitmap. Bitmap >5% or approx. multiple indexes
used to avoid <25% rows per table.
touching rows in selected or an OR
table. operator is
involved in
selection that
precludes the use
of only one index.

Ordering query results

You must specify an ORDER BY clause (or OPNQRYF KEYFLD parameter) to guarantee a particular
ordering of the results. Before parallel access methods were available, the database manager processed
table rows (and keyed sequences) in a sequential manner. This caused the sequencing of the results to be
somewhat predictable (generally rows were retrieved in the order in which they were stored in the
dataspace) even though ordering was not included in the original query request. Because parallel methods
cause blocks of table rows and key values to be processed concurrently, the ordering of the retrieved
results becomes more random and unpredictable.

An ORDER BY clause is the only way to guarantee the specific sequencing of the results. However, an
ordering request should only be specified when absolutely required, because the sorting of the results can
increase both CPU utilization and response time.

Enabling parallel processing for queries

You must enable parallel processing for queries when you submit them or when you code your
applications. The optimizer does not automatically use parallelism as the chosen access method.

You can use the system-value QQRYDEGREE, the query options file, or the DEGREE parameter on the
Change Query Attributes (CHGQRYA) command to control the degree of parallelism that the query
optimizer uses. See “Control parallel processing for queries” on page 92 for information on how to control
parallel processing.

A set of database system tasks is created at system startup for use by the database manager. The
database manager uses the tasks to process and retrieve data from different disk devices. Since these
tasks can be run on multiple processors simultaneously, the elapsed time of a query can be reduced. Even
though much of the I/O and CPU processing of a parallel query is done by the tasks, the accounting of the

Chapter 2. Data access on DB2 UDB for iSeries: data access paths and methods 7
I/O and CPU resources used are transferred to the application job. The summarized I/O and CPU
resources for this type of application continue to be accurately displayed by the Work with Active Jobs
(WRKACTJOB) command.

Spreading data automatically

DB2 Universal Database for iSeries automatically spreads the data across the disk devices available in the
auxiliary storage pool (ASP) where the data is allocated. This ensures that the data is spread without user
intervention. The spreading allows the database manager to easily process the blocks of rows on different
disk devices in parallel.

Even though DB2 Universal Database for iSeries spreads data across disk devices within an ASP,
sometimes the allocation of the data extents (contiguous sets of data) might not be spread evenly. This
occurs when there is uneven allocation of space on the devices, or when a new device is added to the
ASP. The allocation of the data space may be spread again by saving, deleting, and then restoring the
table.

Table scan access method

All rows in the table are read. The selection criteria are applied to each row, and only the rows that match
the criteria are returned to the calling application. The rows in the table are processed in no guaranteed
order. If you want the result in a particular sequence, you must specify the ORDER BY clause (or
OPNQRYF KEYFLD parameter).

Table scan can be efficient for the following reasons:

v It minimizes the number of page I/O operations because all rows in a given page are processed, and
once the page is in main storage, the page is not retrieved again.
v The database manager can easily predict the sequence of pages from the dataspace for retrieval. For
this reason, the database manager can schedule asynchronous I/O of the pages into main storage from
auxiliary storage. This is commonly referred to as prefetching. This is done so that the page is
available in main storage when the database manager needs to access the data.

Where the table scan access method is most effective

This selection method is good when a large percentage of the rows is to be selected. A large percentage
is generally 20% or more.

Where the table scan access method is least effective

Table scan processing is not efficient when a small percentage of rows in the table will be selected.
Because all rows in the table are examined, this leads to unnecessary use of I/O and processing unit
resources.

Considerations for table scan access

Table scan processing can be adversely affected when rows are selected from a table that contains
deleted rows. This is because the delete operation only marks rows as deleted. For table scan processing,
the database manager reads all of the deleted rows, even though none of the deleted rows are ever
selected. You should use the Reorganize Physical File Member (RGZPFM) CL command to eliminate
deleted rows. By specifying REUSEDLT(*YES) on the physical file, you can also reuse the deleted row
space. All SQL tables are created with REUSEDLT(*YES).

PRTSQLINF command messages

The messages created by the PRTSQLINF CL command to describe a query in an SQL program which is
using the dataspace selection method would appear as follows:

8 DB2 UDB for iSeries Database Performance and Query Optimization V5R1
| SQL4010 Table scan access for table 1.

Selection algorithms for table scan access method

The Licensed Internal Code can use one of two algorithms for selection when a table scan is processed,
derived-column selection and dataspace-only selection. The dataspace-only selection has two forms -
dataspace looping and dataspace-only filtering. Dataspace looping processes large sets of records
efficiently, while dataspace-only filtering is another step to eliminate records prior to derived operations.

All access methods use dataspace filtering, but dataspace looping is only used when a table scan is
processing a high percentage of records.

The following pseudocode illustrates the derived column selection algorithm:

DO UNTIL END OF TABLE

1. Address the next (or first) row

2. Map all column values to an internal buffer, performing all derived

operations.

3. Evaluate the selection criteria to a TRUE or FALSE value using

the column values as they were copied to internal buffer.

4. IF the selection is TRUE

THEN
Copy the values from the internal buffer into the
user’s answer buffer.
ELSE
No operation
END

The table-scan selection algorithm is as follows:

DO UNTIL END OF TABLE

1. Calculate a search limit. This limit is usually the number of

rows which are already in active memory, or have already
had an I/O request done to be loaded into memory.

2. DO UNTIL (search limit reached

or row selection criteria is TRUE)
a. Address the next (or first) row

b. Evaluate any selection criteria which does not

require a derived value directly for the dataspace
row.

END

3. IF the selection is true

THEN
a. Map all column values to an internal buffer, performing all
derived operations.

b. Copy the values from the internal buffer into the

user’s answer buffer.
ELSE
No operation
END

The table-scan selection algorithm provides better performance than derived column selection for two
reasons:
v Data movement and computations are only done on rows that are selected.

Chapter 2. Data access on DB2 UDB for iSeries: data access paths and methods 9
 v The loop in step 2 of the table-scan selection algorithm is generated into an executable code burst.
When a small percentage of rows is actually selected, DB2 Universal Database for iSeries will be
running this small program until a row is found.

Guidelines for coding queries

| No action is necessary for queries that use the table scan selection algorithm of the table scan access
| method. Any query interface can use this improvement. However, the following guidelines determine
| whether a selection predicate can be implemented as a dataspace selection:
| v The optimizer always ensures that the operands for any selection item are compatible, therefore you
| can improve your queries by making sure the operands are compatible before processing the query.
| v Neither operand of the predicate can be any kind of a derived value, function, substring, concatenation,
| or numeric expression.
| v When both operands of a selection predicate are numeric columns, both columns must have the same
| type, scale, and precision; otherwise, one operand is mapped into a derived value. For example, a
| DECIMAL(3,1) must only be compared against another DECIMAL(3,1) column.
| v When one operand of a selection predicate is a numeric column and the other is a constant or host
| variable, then the types must be the same and the precision and scale of the constant or host variable
| must be less than or equal to that of the column.
| v Selection predicates involving packed decimal or numeric types of columns can only be done if the table
| was created by the SQL CREATE TABLE statement.
| v A varying-length character column cannot be referenced in the selection predicate.
| v When one operand of a selection predicate is a character column and the other is a constant or host
| variable, then the length of the host variable cannot be greater than that of the column.
| v Comparison of character-column data must not require CCSID or keyboard shift translation.

It can be important to avoid derived-column selection because the reduction in CPU and response time for
table scan selection can be large, in some cases as high as 70-80%. The queries that will benefit the most
from dataspace only selection are those where less than 60% of the table is actually selected. The lower
the percentage of rows selected, the more noticeable the performance benefit will be.

Parallel table prefetch access method

DB2 Universal Database for iSeries can also use parallel table prefetch processing to shorten the
processing time that is required for long-running, I/O-bound table scan queries.

This method has the same characteristics as the table scan method; however, the I/O processing is done
in parallel. This is accomplished by starting multiple input streams for the table to prefetch the data.

Where the parallel table prefetch access method is most effective

This method is most effective when the following are true:

v The data is spread across multiple disk devices.
v The optimizer has determined that the query will be I/O bound.
v There is an ample amount of main storage available to hold the data that is collected from every input
stream.

DB2 Universal Database for iSeries data spreading

As mentioned previously, DB2 Universal Database for iSeries automatically spreads the data across the
disk devices without user intervention, allowing the database manager to prefetch table data in parallel.
The database manager uses tasks to retrieve data from different disk devices. Usually the request is for
an entire extent (contiguous set of data). This improves performance because the disk device can use

10 DB2 UDB for iSeries Database Performance and Query Optimization V5R1
 smooth sequential access to the data. Because of this optimization, parallel prefetch can preload data to
active memory faster than the SETOBJACC CL command.

Even though DB2 Universal Database for iSeries spreads data across disk devices within an ASP,
sometimes the allocation of the dataspace extents may not be spread evenly. This occurs when there is
uneven allocation of space on the devices or a new device is added to the ASP. The allocation of the
dataspace can be respread by saving, deleting, and restoring the table.

How the query optimizer selects queries that use this method

The query optimizer selects the candidate queries which can take advantage of this type of
implementation. The optimizer selects the candidates by estimating the CPU time required to process the
query and comparing the estimate to the amount of time required for input processing. When the
estimated input processing time exceeds the CPU time, the query optimizer indicates that the query may
be implemented with parallel I/O.

| If DB2 UDB Symmetric Multiprocessing is installed, then the query optimizer usually prefers the DB2 UDB
| Symmetric Multiprocessing parallel methods.

Processing requirements

Parallel table prefetch requires that input and output parallel processing must be enabled by the system
value QQRYDEGREE, by the query option file, or by the DEGREE parameter on the Change Query
Attributes (CHGQRYA) command. See “Control parallel processing for queries” on page 92 for information
on how to control parallel processing. Because queries being processed with parallel table prefetch
aggressively use main storage and disk I/O resources, the number of queries that use parallel table
prefetch should be limited and controlled. Parallel prefetch uses multiple disk arms, but it makes little use
of multiple CPUs for any given query. Parallel prefetch I/O will use I/O resources intensely. Allowing a
parallel prefetch query on a system with an overcommitted I/O subsystem may intensify the
over-commitment problem.

You should run the job in a shared storage pool with the *CALC paging option because this causes more
efficient use of active memory. DB2 Universal Database for iSeries uses the automated system tuner to
determine how much memory this process is allowed to use. At run-time, the Licensed Internal Code will
allow parallel table prefetch to be used only if the memory statistics indicate that it will not overcommit the
memory resources. For more information on the paging option, see the Automatic System Tuning section
of the Work Management topic.

Parallel table prefetch requires that enough memory be available to cache the data that is being retrieved
by the multiple input streams. For large tables, the typical extent size is 1 MB. This means that 2 MBof
memory must be available to use two input streams concurrently. Increasing the amount of available
memory in the pool allows more input streams to be used. If plenty of memory is available, the entire
dataspace for the table may be loaded into active memory when the query is opened.

PRTSQLINF command messages

The messages created by the PRTSQLINF command to describe a query in an SQL program which is
using the parallel table prefetch access method would appear as follows:
SQL4023 Parallel dataspace prefetch used.

Parallel table scan method

DB2 Universal Database for iSeries can use this parallel access method to shorten the processing time
that is required for long-running table scan queries when the DB2 UDB Symmetric Multiprocessing feature
is installed. The parallel table scan method reduces the I/O processing time like the parallel table prefetch
access method. In addition, if running on a system that has more than one processor, this method can

Chapter 2. Data access on DB2 UDB for iSeries: data access paths and methods 11
reduce the elapsed time of a query by splitting the table scan processing into tasks that can be run on the
multiple processors simultaneously. All selection and column processing is performed in the task. The
application’s job schedules the work requests to the tasks and merges the results into the result buffer that
is returned to the application.

Where the parallel table scan access method is most effective

This method is most effective when the following are true:

v The data is spread across multiple disk devices.
v The system has multiple processors that are available.
v There is an ample amount of main storage available to hold the data buffers and result buffers.
v When used for large tables in an OLAP or batch environment.

How the query optimizer selects queries that use this method

As mentioned previously, DB2 Universal Database for iSeries automatically spreads the data across the
disk devices without user intervention, allowing the database manager to prefetch table data in parallel.
This allows each task to concentrate on its share of the striped data stored away. This way there is no
contention on any of the tasks to gain access to the data and perform their portion of the query.

The query optimizer selects the candidate queries that can take advantage of this type of implementation.
The optimizer selects the candidates by estimating the CPU time required to process the query and
comparing the estimate to the amount of time required for input processing. The optimizer reduces its
estimated elapsed time for table scan based on the number of tasks it calculates should be used. It
calculates the number of tasks based on the number of processors in the system, the amount of memory
available in the job’s pool, and the current value of the DEGREE query attribute. If the parallel table scan
is the fastest access method, it is then chosen.

Processing requirements

Parallel table scan requires that SMP parallel processing must be enabled either by the system value
QQRYDEGREE, the query option file, or by the DEGREE parameter on the Change Query Attributes
(CHGQRYA) command. See “Control parallel processing for queries” on page 92 for information on how to
control parallel processing.

Parallel table scan cannot be used for queries that require any of the following:
v Specification of the *ALL commitment control level.
v Nested loop join implementation. See “Nested loop join implementation” on page 35.
v Backward scrolling. For example, parallel table scan cannot normally be used for queries defined by the
Open Query File (OPNQRYF) command, which specify ALWCPYDTA(*YES) or ALWCPYDTA(*NO),
because the application might attempt to position to the last row and retrieve previous rows.
SQL-defined queries that are not defined as scrollable can use this method. Parallel table scan can be
used during the creation of a temporary result, such as a sort or hash operation, no matter what
interface was used to define the query. OPNQRYF can be defined as not scrollable by specifying the
*OPTIMIZE parameter value for the ALWCPYDTA parameter, which enables the usage of most of the
parallel access methods.
v Restoration of the cursor position. For instance, a query requiring that the cursor position be restored as
the result of the SQL ROLLBACK HOLD statement or the ROLLBACK CL command. SQL applications
using a commitment control level other than *NONE should specify *ALLREAD as the value for
precompiler parameter ALWBLK to allow this method to be used.
v Update or delete capability.

12 DB2 UDB for iSeries Database Performance and Query Optimization V5R1
 You should run the job in a shared storage pool with the *CALC paging option, as this will cause more
efficient use of active memory. For more information on the paging option see the Automatic System
Tuning section of the Work Management topic in the iSeries Information Center.

Parallel table scan requires active memory to buffer the data that is being retrieved, and to separate result
buffers for each task. A typical total amount of memory that is needed for each task is about 2 megabytes.
For example, about 8 megabytes of memory must be available in order to use 4 parallel table scan tasks
concurrently. Increasing the amount of available memory in the pool allows more input streams to be used.
Queries that access tables with large varying length character columns, or queries that generate result
values that are larger than the actual row length of the table might require more memory for each task.

The performance of parallel table scan can be severely limited if numerous row locking conflicts or data
mapping errors occur.

Index scan-key selection access method

This access method requires indexes. The entire index is read, and any selection criteria that references
the key columns of the index are applied against the index. The advantage of this method is that the
dataspace is only accessed to retrieve rows that satisfy the selection criteria applied against the index. Any
additional selection not performed through the index scan-key selection method is performed at the
dataspace level.

The index scan-key selection access method can be very expensive if the search condition applies to a
large number of rows because:
v The whole index is processed.
v For every key selected from the index, a random I/O to the dataspace occurs.

How the query optimizer selects queries that use this method

Normally, the optimizer would choose to use table scan processing when the search condition applies to a
large number of rows. The optimizer only chooses the index scan-key selection method if less than 20% of
the keys are selected or if an operation forces the use of an index. Operations that might force the use of
an index include:
v Ordering
v Grouping
v Joining

| In these cases, the optimizer may choose to create a temporary index rather than use an existing index.
| When the optimizer creates a temporary index, it uses a 64K page size for primary dials and an 8K page
| size for secondary dials. An index created using the SQL CREATE INDEX statement uses 64K page size.
| For indexes that are created using the CRTLF command, or for SQL indexes created before V4R5M0, the
| index size is normally 16K.

The optimizer also processes as much of the selection as possible while building the temporary index.
Nearly all temporary indexes built by the optimizer are select/omit or sparse indexes. Finally, the optimizer
can use multiple parallel tasks when creating the index. The page size difference, corresponding
performance improvement from swapping in fewer pages, and the ability to use parallel tasks to create the
index may be enough to overcome the overhead cost of creating an index. Dataspace selection is used for
building of temporary indexes.

If index scan-key selection access method is used because the query specified ordering (an index was
required) the query performance might be improved by using the following parameters to allow the
ordering to be done with the query sort.
v For SQL, the following combinations of precompiler parameters:
– ALWCPYDTA(*OPTIMIZE), ALWBLK(*ALLREAD), and COMMIT(*CHG or *CS)

Chapter 2. Data access on DB2 UDB for iSeries: data access paths and methods 13
 – ALWCPYDTA(*OPTIMIZE) and COMMIT(*NONE)
v For OPNQRYF, the following parameters:
– *ALWCPYDTA(*OPTIMIZE) and COMMIT(*NO)
– ALWCPYDTA(*OPTIMIZE) and COMMIT(*YES) and the commitment control level is started with a
commit level of *NONE, *CHG, or *CS

When a query specifies a select/omit index and the optimizer decides to build a temporary index, all of the
selection from the select/omit index is put into the temporary index after any applicable selection from the
query.

Parallel index scan-key selection access method (available only when

the DB2 UDB Symmetric Multiprocessing feature is installed)
For the parallel index scan-key selection access method, the possible key values are logically partitioned.
Each partition is processed by a separate task just as in the index scan-key selection access method. The
number of partitions processed concurrently is determined by the query optimizer. Because the keys are
not processed in order, this method cannot be used by the optimizer if the index is being used for
ordering. Key partitions that contain a larger portion of the existing keys from the index are further split as
processing of other partitions complete.

Where the parallel index scan-key selection access method is most effective

The following example illustrates a query where the optimizer could choose the index scan-key selection
method:
CREATE INDEX X1 ON EMPLOYEE(LASTNAME,WORKDEPT)

DECLARE BROWSE2 CURSOR FOR

SELECT * FROM EMPLOYEE
WHERE WORKDEPT = ’E01’
OPTIMIZE FOR 99999 ROWS

OPNQRYF example:
OPNQRYF FILE((EMPLOYEE))
QRYSLT(’WORKDEPT *EQ ’’E01’’’)

If the optimizer chooses to run this query in parallel with a degree of four, the following might be the logical
key partitions that get processed concurrently:
LASTNAME values LASTNAME values
leading character leading character
partition start partition end
’A’ ’F’
’G’ ’L’
’M’ ’S’
’T’ ’Z’

If there were fewer keys in the first and second partition, processing of those key values would complete
sooner than the third and fourth partitions. After the first two partitions are finished, the remaining key
values in the last two might be further split. The following shows the four partitions that might be
processed after the first and second partition are finished and the splits have occurred:
LASTNAME values LASTNAME values
leading character leading character
partition start partition end
’O’ ’P’
’Q’ ’S’
’V’ ’W’
’X’ ’Z’

Processing requirements

14 DB2 UDB for iSeries Database Performance and Query Optimization V5R1
 Parallel index scan-key selection cannot be used for queries that require any of the following:
v Specification of the *ALL commitment control level.
v Nested loop join implementation. See “Nested loop join implementation” on page 35.
| v Backward scrolling. For example, parallel index scan-key selection cannot be used for queries defined
| by the Open Query File (OPNQRYF) command which specify ALWCPYDTA(*YES) or
| ALWCPYDTA(*NO), because the application might attempt to position to the last row and retrieve
| previous rows. OPNQRYF can be defined as not scrollable by specifying the *OPTIMIZE parameter
| value for the ALWCPYDTA parameter, which enables the usage of most of the parallel access methods.
| SQL defined queries that are not defined as scrollable can use this method. Parallel index scan-key
| selection can be used during the creation of a temporary result, such as a sort or hash operation, no
| matter what interface was used to define the query.
v Restoration of the cursor position (for instance, a query requiring that the cursor position be restored as
the result of the SQL ROLLBACK HOLD statement or the ROLLBACK CL command). SQL applications
using a commitment control level other than *NONE should specify *ALLREAD as the value for
precompiler parameter ALWBLK to allow this method to be used.
v Update or delete capability.

You should run the job in a shared pool with *CALC paging option as this will cause more efficient use of
active memory. For more information on the paging option see the Automatic System Tuning section of the
Work Management topic in the iSeries Information Center.

Parallel index scan-key selection requires that SMP parallel processing be enabled either by the system
value QQRYDEGREE, the query options file, or by the DEGREE parameter on the Change Query
Attributes (CHGQRYA) command. See “Control parallel processing for queries” on page 92 for information
on how to control parallel processing.

Index scan-key positioning access method

| This access method is very similar to the index scan-key selection access method. They both require a
| keyed sequence index. In the index scan-key selection access method, processing starts at the beginning
| of the index and continues to the end; all keys are paged in. In the index scan-key positioning access
| method, selection is against the index directly on a range of keys that match some or all of the selection
| criteria. Only those keys from this range are paged in and any remaining index selection is performed by
| the index scan-key selection method. Any selection not performed through index scan-key positioning or
| index scan-key selection is performed at the dataspace level. Because index scan-key positioning only
| retrieves a subset of the keys in the index, the performance of the index scan-key positioning method is
| better than the performance of the index scan-key selection method.

Where the index scan-key positioning access method is most efficient

The index scan-key positioning method is most efficient when a small percentage of rows are to be
selected (less than approximately 20%). If more than approximately 20% of the rows are to be selected,
the optimizer generally chooses to:
v Use table scan processing (if index is not required)
v Use index scan-key selection (if an index is required)
v Use query sort routine (if conditions apply)

How the query optimizer selects queries that use this method

For queries that do not require an index (no ordering, grouping, or join operations), the optimizer tries to
find an existing index to use for index scan-key positioning. If no existing index can be found, the optimizer
stops trying to use keyed access to the data because it is faster to use table scan processing than it is to
build an index and then perform index scan-key positioning.

Chapter 2. Data access on DB2 UDB for iSeries: data access paths and methods 15
The following example illustrates a query where the optimizer could choose the index scan-key positioning
method:
CREATE INDEX X1 ON EMPLOYEE(WORKDEPT)

DECLARE BROWSE2 CURSOR FOR

SELECT * FROM EMPLOYEE
WHERE WORKDEPT = ’E01’
OPTIMIZE FOR 99999 ROWS

OPNQRYF example:
OPNQRYF FILE((EMPLOYEE))
QRYSLT(’WORKDEPT *EQ ’’E01’’’)

In this example, the database support uses X1 to position to the first index entry with the WORKDEPT
value equal to ’E01’. For each key equal to ’E01’, it randomly accesses the dataspace 2 and selects the
row. The query ends when the index scan-key selection moves beyond the key value of E01.

Note that for this example all index entries processed and rows retrieved meet the selection criteria. If
additional selection is added that cannot be performed through index scan-key positioning (such as
selection columns which do not match the first key columns of an index over multiple columns) the
optimizer uses index scan-key selection to perform as much additional selection as possible. Any
remaining selection is performed at the dataspace level.

The messages created by the PRTSQLINF CL command to describe this query in an SQL program would
appear as follows:
SQL4008 Index X1 used for table 1.
SQL4011 Key row positioning used on table 1.

The index scan-key positioning access method has additional processing capabilities. One such capability
is to perform range selection across several values. For example:
CREATE INDEX X1 EMPLOYEE(WORKDEPT)

DECLARE BROWSE2 CURSOR FOR

SELECT * FROM EMPLOYEE
WHERE WORKDEPT BETWEEN ’E01’ AND ’E11’
OPTIMIZE FOR 99999 ROWS

OPNQRYF example:
OPNQRYF FILE((EMPLOYEE))
QRYSLT(’WORKDEPT *EQ %RANGE(’’E01’’ ’’E11’’)’)

In the previous example, the database support positions to the first index entry equal to value ’E01’ and
rows are processed until the last index entry for ’E11’ is processed.

PRTSQLINF command messages

The messages created by PRTSQLINF CL command to describe this query in an SQL program would
appear as follows:
SQL4008 Index X1 used for table 1.
SQL4011 Key row positioning used on table 1.

Multi-range index scan-key positioning

2. random accessing occurs because the keys may not be in the same sequence as the rows in the dataspace

16 DB2 UDB for iSeries Database Performance and Query Optimization V5R1
A further extension of this access method, called multi-range index scan-key positioning, is available. It
allows for the selection of rows for multiple ranges of values for the first key columns of an index over
multiple columns.
CREATE INDEX X1 ON EMPLOYEE(WORKDEPT)

DECLARE BROWSE2 CURSOR FOR

SELECT * FROM EMPLOYEE
WHERE WORKDEPT BETWEEN ’E01’ AND ’E11’
OR WORKDEPT BETWEEN ’A00’ AND ’B01’
OPTIMIZE FOR 99999 ROWS

OPNQRYF example:
OPNQRYF FILE((EMPLOYEE))
QRYSLT(’WORKDEPT *EQ %RANGE(’’E01’’ ’’E11’’)
*OR WORKDEPT *EQ %RANGE(’’A00’’ ’’B01’’)’)

In the previous example, the positioning and processing technique is used twice, once for each range of
values.

The messages created by PRTSQLINF CL command to describe this query in an SQL program would
appear as follows:
SQL4008 Index X1 used for table 1.
SQL4011 Key row positioning used on table 1.

All of the index scan-key positioning examples have so far only used one key, the left-most key, of the
index. Index scan-key positioning also handles more than one key (although the keys must be contiguous
to the left-most key).
CREATE INDEX X2
ON EMPLOYEE(WORKDEPT,LASTNAME,FIRSTNME)

DECLARE BROWSE2 CURSOR FOR

SELECT * FROM EMPLOYEE
WHERE WORKDEPT = ’D11’
AND FIRSTNME = ’DAVID’
OPTIMIZE FOR 99999 ROWS

OPNQRYF example:
OPNQRYF FILE((EMPLOYEE))
QRYSLT(’WORKDEPT *EQ ’’D11’’
*AND FIRSTNME *EQ ’’DAVID’’’)

Because the two selection keys (WORKDEPT and FIRSTNME) are not contiguous, there is no multiple
key position support for this example. Therefore, only the WORKDEPT = ’D11’ part of the selection can be
applied against the index (single key index scan-key positioning). While this may be acceptable, it means
that the processing of rows starts with the first key of ’D11’ and then uses index scan-key selection to
process the FIRSTNME = ’DAVID’ against all 9 entries with WORKDEPT key value = ’D11’.

By creating the following index, X3, the above example query would run using multiple keys to do the
index scan-key positioning.
CREATE INDEX X3
ON EMPLOYEE(WORKDEPT, FIRSTNME, LASTNAME)

Multiple key index scan-key positioning support can apply both pieces of selection as index scan-key
positioning. This improves performance considerably. A starting value is built by concatenating the two
selection values into ’D11DAVID’ and selection is positioned to the index entry whose left-most two keys
have that value.

Chapter 2. Data access on DB2 UDB for iSeries: data access paths and methods 17
The messages created by the PRTSQLINF CL command when used to describe this query in an SQL
program would look like this:
SQL4008 Index X3 used for table 1.
SQL4011 Key row positioning used on table 1.

This next example shows a more interesting use of multiple index scan-key positioning.

CREATE INDEX X3 ON EMPLOYEE(WORKDEPT,FIRSTNME)

DECLARE BROWSE2 CURSOR FOR

SELECT * FROM EMPLOYEE
WHERE WORKDEPT = ’D11’
AND FIRSTNME IN (’DAVID’,’BRUCE’,’WILLIAM’)
OPTIMIZE FOR 99999 ROWS

OPNQRYF example:
OPNQRYF FILE((EMPLOYEE))
QRYSLT(’WORKDEPT *EQ ’’D11’’
*AND FIRSTNME *EQ %VALUES(’’DAVID’’ ’’BRUCE’’
’’WILLIAM’’)’)

The query optimizer analyzes the WHERE clause and rewrites the clause into an equivalent form:
DECLARE BROWSE2 CURSOR FOR
SELECT * FROM EMPLOYEE
WHERE (WORKDEPT = ’D11’ AND FIRSTNME = ’DAVID’)
OR (WORKDEPT = ’D11’ AND FIRSTNME = ’BRUCE’)
OR (WORKDEPT = ’D11’ AND FIRSTNME = ’WILLIAM’)
OPTIMIZE FOR 99999 ROWS

OPNQRYF example:
OPNQRYF FILE((EMPLOYEE))
QRYSLT(’(WORKDEPT *EQ ’’D11’’ *AND FIRSTNME *EQ
’’DAVID’’)
*OR (WORKDEPT *EQ ’’D11’’ *AND FIRSTNME *EQ ’’BRUCE’’)
*OR (WORKDEPT *EQ ’’D11’’ *AND FIRSTNME *EQ ’’WILLIAM’’)’)

In the rewritten form of the query there are actually 3 separate ranges of key values for the concatenated
values of WORKDEPT and FIRSTNME:
Index X3 Start value Index X3 Stop value
’D11DAVID’ ’D11DAVID’
’D11BRUCE’ ’D11BRUCE’
’D11WILLIAM’ ’D11WILLIAM’

Index scan-key positioning is performed over each range, significantly reducing the number of keys
selected to just 3. All of the selection can be accomplished through index scan-key positioning.

The complexity of this range analysis can be taken to a further degree in the following example:
DECLARE BROWSE2 CURSOR FOR
SELECT * FROM EMPLOYEE
WHERE (WORKDEPT = ’D11’
AND FIRSTNME IN (’DAVID’,’BRUCE’,’WILLIAM’))
OR (WORKDEPT = ’E11’
AND FIRSTNME IN (’PHILIP’,’MAUDE’))
OR (FIRSTNME BETWEEN ’CHRISTINE’ AND ’DELORES’
AND WORKDEPT IN (’A00’,’C01’))

OPNQRYF example:
OPNQRYF FILE((EMPLOYEE))
QRYSLT(’(WORKDEPT *EQ ’’D11’’
*AND FIRSTNME *EQ %VALUES(’’DAVID’’ ’’BRUCE’’ ’’WILLIAM’’))

18 DB2 UDB for iSeries Database Performance and Query Optimization V5R1
 *OR (WORKDEPT *EQ ’’E11’’
*AND FIRSTNME *EQ %VALUES(’’PHILIP’’ ’’MAUDE’’))
*OR (FIRSTNME *EQ %RANGE(’’CHRISTINE’’ ’’DELORES’’)
*AND WORKDEPT *EQ %VALUES(’’A00’’ ’’C01’’))’)

The query optimizer analyzes the WHERE clause and rewrites the clause into an equivalent form:
DECLARE BROWSE2 CURSOR FOR
SELECT * FROM EMPLOYEE
WHERE (WORKDEPT = ’D11’ AND
FIRSTNME = ’DAVID’)
OR (WORKDEPT = ’D11’ AND
FIRSTNME = ’BRUCE’)
OR (WORKDEPT = ’D11’ AND
FIRSTNME = ’WILLIAM’)
OR (WORKDEPT = ’E11’ AND
FIRSTNME = ’PHILIP’)
OR (WORKDEPT = ’E11’ AND
FIRSTNME = ’MAUDE’)
OR (WORKDEPT = ’A00’ AND
FIRSTNME BETWEEN
’CHRISTINE’ AND ’DELORES’)
OR (WORKDEPT = ’C01’ AND FIRSTNME BETWEEN
’CHRISTINE’ AND ’DELORES’)
OPTIMIZE FOR 99999 ROWS

OPNQRYF example:
OPNQRYF FILE((EMPLOYEE))
QRYSLT(’(WORKDEPT *EQ ’’D11’’ *AND FIRSTNME *EQ
’’DAVID’’)
*OR (WORKDEPT *EQ ’’D11’’ *AND FIRSTNME *EQ ’’BRUCE’’)
*OR (WORKDEPT *EQ ’’D11’’ *AND FIRSTNME *EQ
’’WILLIAM’’)
*OR (WORKDEPT *EQ ’’E11’’ *AND FIRSTNME *EQ ’’PHILIP’’)
*OR (WORKDEPT *EQ ’’E11’’ *AND FIRSTNME *EQ ’’MAUDE’’)
*OR (WORKDEPT *EQ ’’A00’’ *AND
FIRSTNME *EQ %RANGE(’’CHRISTINE’’ ’’DELORES’’))
*OR (WORKDEPT *EQ ’’C01’’ *AND
FIRSTNME *EQ %RANGE(’’CHRISTINE’’ ’’DELORES’’))’)

In the query there are actually 7 separate ranges of key values for the concatenated values of
WORKDEPT and FIRSTNME:
Index X3 Start value Index X3 Stop value
’D11DAVID’ ’D11DAVID’
’D11BRUCE’ ’D11BRUCE’
’D11WILLIAM’ ’D11WILLIAM’
’E11MAUDE’ ’E11MAUDE’
’E11PHILIP’ ’E11PHILIP’
’A00CHRISTINE’ ’A00DELORES’
’C01CHRISTINE’ ’C01DELORES’

Index scan-key positioning is performed over each range. Only those rows whose key values fall within
one of the ranges are returned. All of the selection can be accomplished through index scan-key
positioning. This significantly improves the performance of this query.

Parallel index scan-key positioning access method (available only

when the DB2 UDB Symmetric Multiprocessing feature is installed)
Using the parallel index scan-key positioning access method, the existing key ranges are processed by
separate tasks concurrently in separate database tasks. The number of concurrent tasks is controlled by
the optimizer. The query will start processing the key ranges of the query up to the degree of parallelism
being used. As processing of those ranges completes, the next ones on the list are started. As processing
for a range completes and there are no more ranges in the list to process, ranges that still have keys left
to process are split, just as in the parallel index scan-key selection method. The database manager
attempts to keep all of the tasks that are being used busy, each processing a separate key range.
Whether using the single value, range of values, or multi-range index scan-key positioning, the ranges can
be further partitioned and processed simultaneously. Because the keys are not processed in order, this
method can not be used by the optimizer if the index is being used for ordering.

Chapter 2. Data access on DB2 UDB for iSeries: data access paths and methods 19
How the query optimizer uses this method

Consider the following example if the SQL statement is run using parallel degree of four.
DECLARE BROWSE2 CURSOR FOR
SELECT * FROM EMPLOYEE
WHERE (WORKDEPT = ’D11’ AND
FIRSTNME = ’DAVID’)
OR (WORKDEPT = ’D11’ AND
FIRSTNME = ’BRUCE’)
OR (WORKDEPT = ’D11’ AND
FIRSTNME = ’WILLIAM’)
OR (WORKDEPT = ’E11’ AND
FIRSTNME = ’PHILIP’)
OR (WORKDEPT = ’E11’ AND
FIRSTNME = ’MAUDE’)
OR (WORKDEPT = ’A00’ AND
FIRSTNME BETWEEN
’CHRISTINE’ AND ’DELORES’)
OR (WORKDEPT = ’C01’ AND FIRSTNME BETWEEN
’CHRISTINE’ AND ’DELORES’)
OPTIMIZE FOR 99999 ROWS

OPNQRYF example:
OPNQRYF FILE((EMPLOYEE))
QRYSLT(’(WORKDEPT *EQ ’’D11’’ *AND FIRSTNME *EQ ’’DAVID’’)
*OR (WORKDEPT *EQ ’’D11’’ *AND FIRSTNME *EQ ’’BRUCE’’)
*OR (WORKDEPT *EQ ’’D11’’ *AND FIRSTNME *EQ ’’WILLIAM’’)
*OR (WORKDEPT *EQ ’’E11’’ *AND FIRSTNME *EQ ’’PHILIP’’)
*OR (WORKDEPT *EQ ’’E11’’ *AND FIRSTNME *EQ ’’MAUDE’’)
*OR (WORKDEPT *EQ ’’A00’’ *AND
FIRSTNME*EQ %RANGE(’’CHRISTINE’’ ’’DELORES’’))
*OR (WORKDEPT *EQ ’’C01’’ *AND
FIRSTNME *EQ %RANGE(’’CHRISTINE’’ ’’DELORES’’))’)

The key ranges the database manager starts with are as follows:
Index X3 Start value Index X3 Stop value
Range 1 ’D11DAVID’ ’D11DAVID’
Range 2 ’D11BRUCE’ ’D11BRUCE’
Range 3 ’D11WILLIAM’ ’D11WILLIAM’
Range 4 ’E11MAUDE’ ’E11MAUDE’
Range 5 ’E11PHILIP’ ’E11PHILIP’
Range 6 ’A00CHRISTINE’ ’A00DELORES’
Range 7 ’C01CHRISTINE’ ’C01DELORES’

Ranges 1 to 4 are processed concurrently in separate tasks. As soon as one of those four completes,
range 5 is started. When another range completes, range 6 is started, and so on. When one of the four
ranges in progress completes and there are no more new ones in the list to start, the remaining work left
in one of the other key ranges is split and each half is processed separately.

Processing requirements

Parallel index scan-key positioning cannot be used for queries that require any of the following:
v Specification of the *ALL commitment control level.
v Nested loop join implementation. See “Nested loop join implementation” on page 35.
v Backward scrolling. For example, parallel index scan-key positioning cannot be used for queries defined
by the Open Query File (OPNQRYF) command, which specify ALWCPYDTA(*YES) or
ALWCPYDTA(*NO), because the application might attempt to position to the last row and retrieve
previous rows. SQL-defined queries that are not defined as scrollable can use this method. Parallel
index scan-key positioning can be used during the creation of a temporary result, such as a sort or hash
operation, no matter what interface was used to define the query. OPNQRYF can be defined as not
scrollable by specifying the *OPTIMIZE parameter value for the ALWCPYDTA parameter, which enables
the usage of most of the parallel access methods.
v Restoration of the cursor position. For instance, a query requiring that the cursor position be restored as
the result of the SQL ROLLBACK HOLD statement or the ROLLBACK CL command. SQL applications

20 DB2 UDB for iSeries Database Performance and Query Optimization V5R1
 using a commitment control level other than *NONE should specify *ALLREAD as the value for
precompiler parameter ALWBLK to allow this method to be used.
v Update or delete capability.

You should run the job in a shared pool with the *CALC paging option as this will cause more efficient use
of active memory. For more information on the paging option see the Automatic System Tuning section of
the Work Management topic in the iSeries Information Center.

Parallel index scan-key selection requires that SMP parallel processing be enabled either by the system
value QQRYDEGREE, by the query options file PARALLEL_DEGREE option, or by the DEGREE
parameter on the Change Query Attributes (CHGQRYA) command. See “Control parallel processing for
queries” on page 92 for information on how to control parallel processing.

Index Only Access Method

The index-only access method can be used in conjunction with any of the index scan-key selection or
index scan-key positioning access methods, including the parallel options for these methods. (The parallel
options are available only when the DB2 UDB Symmetric Multiprocessing feature is installed.) The
processing for the selection does not change from what has already been described for these methods.

However, all of the data is extracted from the index rather than performing a random I/O to the data space.
The index entry is then used as the input for any derivation or result mapping that might have been
specified on the query.

Where the index-only method is most effective

The optimizer chooses this method when:

v All of the columns that are referenced within the query can be found within a permanent index or within
the key columns of a temporary index that the optimizer has decided to create.
v The data values must be able to be extracted from the index and returned to the user in a readable
format; in other words, none of the key columns that match the query columns have:
– Absolute value specified
– Alternative collating sequence or sort sequence specified
– Zoned or digit force specified
v The query does not use a left outer join or an exception join.
v For non-SQL users, no variable length or null capable columns can require key feedback.

The following example illustrates a query where the optimizer could choose to perform index only access.
CREATE INDEX X2
ON EMPLOYEE(WORKDEPT,LASTNAME,FIRSTNME)

DECLARE BROWSE2 CURSOR FOR

SELECT FIRSTNME FROM EMPLOYEE
WHERE WORKDEPT = ’D11’
OPTIMIZE FOR 99999 ROWS

OPNQRYF example:
OPNQRYF FILE((EMPLOYEE))
QRYSLT(’WORKDEPT *EQ ’’D11’’’)

In this example, the database manager uses X2 to position to the index entries for WORKDEPT=’D11’ and
then extracts the value for the column FIRSTNME from those entries.

Chapter 2. Data access on DB2 UDB for iSeries: data access paths and methods 21
 Note that the index key columns do not have to be contiguous to the leftmost key of the index for index
only access to be performed. Any key column in the index can be used to provide data for the index only
query. The index is used simply as the source for the data so the database manager can finish processing
the query after the selection has been completed.

Note: Index only access is implemented on a particular table, so it is possible to perform index only
access on some or all of the tables of a join query.

PRTSQLINF command messages

The messages created by the PRTSQLINF command to describe this query in an SQL program are as
follows:
SQL4008 Index X2 used for table 1.
SQL4011 Key row positioning used on table 1.
SQL4022 Index only access used on table 1.

Parallel table or index based preload access method

| Some queries implemented with index scan-key selection can require a lot of random I/O in order to
| access the dataspace. Because of this, a high percentage of the data in the dataspace is referenced. DB2
| Universal Database for iSeries attempts to avoid this random I/O by initiating index- or table-based preload
| when query processing begins. The entire table or index is loaded into active memory in parallel as is
| done for parallel table prefetch. This requires that you have enough memory in the pool to load the entire
| object.

| After the table or index is loaded into memory, random access to the data is achieved without further I/O.
| The DB2 Universal Database for iSeries cost-based query optimizer recognizes the queries and objects
| that benefit from table or index preloads if I/O parallel processing has been enabled. See “Control parallel
| processing for queries” on page 92 for information on how to control parallel processing. If DB2 UDB
| Symmetric Multiprocessing is installed, then the query optimizer usually prefers the DB2 UDB Symmetric
| Multiprocessing parallel methods.

The parallel preload method can be used with any of the other data access methods. The preload is
started when the query is opened and control is returned to the application before the preload is finished.
The application continues fetching rows using the other database access methods without any knowledge
of preload.

Index-from-index access method

| The database manager can build a temporary index from an existing index without having to read all of the
| rows in the dataspace. Generally speaking, this selection method is one of the most efficient. The
| temporary index that is created contains entries for only those rows that meet the selection predicates.
| This is similar to the index created by a select/omit logical file (commonly referred to as a sparse index).

Where the index-from-index access method is most effective

The optimizer chooses this method when:

v The query requires an index because it uses grouping, ordering, or join processing.
v A permanent index exists that has selection columns as the left-most keys and the left-most keys are
very selective (i.e., index scan key positioning can be used).
v The selection columns are not the same as the ordering, grouping, or join-to columns.

How the query optimizer uses this method

To use the index-from-index access method, the database manager:

1. Uses index scan-key positioning on the permanent index with the query selection criteria

22 DB2 UDB for iSeries Database Performance and Query Optimization V5R1
2. Builds index entries in the new temporary index using selected entries.
3. Key columns of the temporary index match the grouping, ordering or join columns.

The result is an index containing entries in the required keyed sequence (grouping, ordering, join) for rows
that match the selection criteria.

A common index-from-index access method example follows:

CREATE INDEX X1 ON EMPLOYEE(WORKDEPT)

DECLARE BROWSE2 CURSOR FOR

SELECT * FROM EMPLOYEE
WHERE WORKDEPT = ’D11’
ORDER BY LASTNAME
OPTIMIZE FOR 99999 ROWS

OPNQRYF example:
OPNQRYF FILE((EMPLOYEE))
QRYSLT(’WORKDEPT *EQ ’’D11’’’)
KEYFLD((LASTNAME))

For this example, a temporary select/omit index is created with the primary key column LASTNAME. It
contains index entries for only those rows where WORKDEPT = ’D11’ assuming less than approximately
20% of the entries are selected.

PRTSQLINF command messages

The messages created by the PRTSQLINF CL command to describe this query in an SQL program are as
follows:
SQL4012 Index created from index X1 for table 1.
SQL4011 Key row positioning used on table 1.

Alternatives to the index-from-index method

Rather than using the index-from-index access method, you can use the query sort routine. See “Sort
access method” on page 28 for more information.

This decision is based on the number of rows to be retrieved.

Hashing access method

The hashing access method provides an alternative method for those queries (groupings and joins) that
must process data in a grouped or correlated manner. Indexes are used to sort and group the data and
are effective in some cases for implementing grouping and join query operations. However, if the optimizer
had to create a temporary index for that query, extra processor time and resources are used when
creating this index before the requested query can be run.

Where the hashing access method is most effective

The hashing access method can complement indexes or serve as an alternative. For each selected row,
the specified grouping or join value in the row is run through a hashing function. The computed hash value
is then used to search a specific partition of the hash table. A hash table is similar to a temporary work
table, but has a different structure that is logically partitioned based on the specified query. If the row’s
source value is not found in the table, then this marks the first time that this source value has been
encountered in the database table. A new hash table entry is initialized with this first-time value and
additional processing is performed based on the query operation. If the row’s source value is found in the
table, the hash table entry for this value is retrieved and additional query processing is performed based

Chapter 2. Data access on DB2 UDB for iSeries: data access paths and methods 23
on the requested operation (such as grouping or joining). The hash method can only correlate (or group)
identical values; the hash table rows are not guaranteed to be sorted in ascending or descending order.

Where this method can be used

The hashing method can be used only when the ALWCPYDTA(*OPTIMIZE) option has been specified
unless a temporary result is required, since the hash table built by the database manager is a temporary
copy of the selected rows.

How this method works

The hashing algorithm allows the database manager to build a hash table that is well-balanced, given that
the source data is random and distributed. The hash table itself is partitioned based on the requested
query operation and the number of source values being processed. The hashing algorithm then ensures
that the new hash table entries are distributed evenly across the hash table partitions. This balanced
distribution is necessary to guarantee that scans in different partitions of the hash tables are processing
the same number of entries. If one hash table partition contains a majority of the hash table entries, then
scans of that partition are going to have to examine the majority of the entries in the hash table. This is
not very efficient.

Since the hash method typically processes the rows in a table sequentially, the database manager can
easily predict the sequence of memory pages from the database table needed for query processing. This
is similar to the advantages of the table scan access method. The predictability allows the database
manager to schedule asynchronous I/O of the table pages into main storage (also known as pre-fetching).
Pre-fetching enables very efficient I/O operations for the hash method leading to improved query
performance.

In contrast, query processing with a keyed sequence access method causes a random I/O to the database
table for every key value examined. The I/O operations are random since the keyed-order of the data in
the index does not match the physical order of the rows in the database table. Random I/O can reduce
query performance because it leads to unnecessary use of I/O and processor unit resources.

An index can also be used by the hash method to process the table rows in keyed order. The index can
significantly reduce the number of table rows that the hash method has to process. This can offset the
random I/O costs associated with indexes.

The hash table creation and population takes place before the query is opened. Once the hash table has
been completely populated with the specified database rows, the hash table is used by the database
manager to start returning the results of the queries. Additional processing might be required on the
resulting hash table rows, depending on the requested query operations.

Since blocks of table rows are automatically spread, the hashing access method can also be performed in
parallel so that several groups of rows are being hashed at the same time. This shortens the amount of
time it takes to hash all the rows in the database table.

If the DB2 UDB Symmetric Multiprocessing feature is installed, the hashing methods can be performed in
parallel.

Bitmap processing method

As the name implies, this method generates bitmaps that are used during access to the data space. The
bitmap processing method is used to:
v Minimize the random I/O that occurs on a data space when using an index in conjunction with the key
position or when using the index scan-key selection method.
v Schedule the I/O more efficiently by skipping large sections of data within the dataspace when the
skip-sequential method is used.

24 DB2 UDB for iSeries Database Performance and Query Optimization V5R1
v Allow multiple indexes to be used to access a particular dataspace.

How the bitmap processing method works

In this method, the optimizer chooses one or more indexes to be used to aid in selecting rows from the
dataspace. Temporary bitmaps are allocated (and initialized), one for each index. Each bitmap contains
one bit for each row in the underlying data space. For each index, index scan-key row positioning and
index scan-key row selection methods are used to apply selection criteria when initializing the bitmap.

For each index entry selected, the bit associated with that row is set to 1 (that is, turned on). The data
space is not accessed. When the processing of the index is complete, the bitmap contains the information
on which rows are to be selected from the underlying data space. This process is repeated for each index.
If two or more indexes are used, the temporary bitmaps are logically ANDed and ORed together to obtain
one resulting bitmap. Once the resulting bitmap is built, it is used to avoid mapping in rows from the
dataspace unless they are selected by the query. It is used to help schedule the selection of rows from the
dataspace or to provide another level of filtering prior to the underlying dataspace being accessed.

The indexes used to generate the bitmaps are not actually used to access the selected rows. For this
reason, they are called tertiary indexes. Conversely, indexes used to access the final rows are called
primary indexes. Primary indexes are used for ordering, grouping, joining, and for selection when no
bitmap is used.

Where the method is used

Bitmaps are always preprocessed before the optimizer starts to process the query through the primary
access method. The bitmap processing method is used in conjunction with primary access methods table
scan, index scan-key row positioning, or index scan-key row selection. Bitmap processing, like parallel
table prefetch and parallel table/index preload, does not actually select the rows from the data space; it
simply assists the primary methods.

If the bitmap is used in conjunction with the table scan method, the bitmap initiates skip-sequential
processing. The table scan (and parallel table scan) uses the bitmap to ″skip over″ pages with no selected
rows (i.e., no bits in the bitmap are set to 1). This has several advantages:
v No CPU processing is used to process nonselected rows.
v I/O is minimized and the memory is not filled with the contents of the entire data space.

Example: Bitmap processing method used in conjunction with table scan method

The following example illustrates a query where the query optimizer chooses the bitmap processing
method in conjunction with the table scan:
CREATE INDEX IX1 ON EMPLOYEE (WORKDEPT)
CREATE INDEX IX2 ON EMPLOYEE (SALARY)

DECLARE C1 CURSOR FOR

SELECT * FROM EMPLOYEE
WHERE WORKDEPT = ’E01’ OR SALARY>50000
OPTIMIZE FOR 99999 ROWS

OPNQRYF example:
OPNQRYF FILE((EMPLOYEE))
QRYSLT(’WORKDEPT *EQ ’’E01’’ *OR SALARY > 50000’)

In this example, both indexes IX1 and IX2 are used. The database manager first generates a bitmap from
the results of applying selection WORKDEPT = ’E01’ against index IX1 (using index scan-key positioning).
The database manager then generates a bitmap from the results of applying selection SALARY>50000
against index IX2 (again using index scan-key positioning).

Chapter 2. Data access on DB2 UDB for iSeries: data access paths and methods 25
Next, the database manager combines these two bitmaps into one resulting bitmap by logically ORing the
individual bitmaps together. Finally, a table scan is initiated. The table scan uses the bitmap to skip
through the data space rows, retrieving only those selected by the bitmap. This improves performance by
skipping over large portions of data.

This example also shows an additional capability provided with bitmap processing (use of an index for
ANDed selection was already possible but bitmap processing now allows more than one index). When
using bitmap processing, multiple index usage is possible with selections where OR is the major Boolean
operator.

The messages created by the PRTSQLINF command when used to describe this query would look like:
SQL4010 Table scan for table 1.
SQL4032 Index IX1 used for bitmap processing of table 1.
SQL4032 Index IX2 used for bitmap processing of table 1.
CPI4329 Arrival sequence access was used for file EMPLOYEE.
CPI4388 2 access path(s) used for bitmap processing of file EMPLOYEE.

Example: Bitmap processing used in conjunction with the index scan-key positioning access
method

If the bitmap is used in conjunction with either the index scan-key row positioning or index scan-key row
selection method, it implies that the bitmap (generated from tertiary indexes) is being used to aid a primary
index access. The following example illustrates a query where bitmap processing is used in conjunction
with the index scan-key positioning for a primary index:
CREATE INDEX PIX ON EMPLOYEE (LASTNAME)
CREATE INDEX TIX1 ON EMPLOYEE (WORKDEPT)
CREATE INDEX TIX2 ON EMPLOYEE (SALARY)

DECLARE C1 CURSOR FOR

SELECT * FROM EMPLOYEE
WHERE WORKDEPT = ’E01’ OR SALARY>50000
ORDER BY LASTNAME

OPNQRYF example:
OPNQRYF FILE((EMPLOYEE))
QRYSLT(’WORKDEPT *EQ ’’E01’’ *OR SALARY > 50000’)
KEYFLD(LASTNAME)

In this example, indexes TIX1 and TIX2 are used in bitmap processing. The database manager first
generates a bitmap from the results of applying selection WORKDEPT = ’E01’ against index TIX1 (using
index scan-key positioning). It then generates a bitmap from the results of applying selection
SALARY>50000 against index TIX2 (again using index scan-key positioning).

The database manager then combines these two bitmaps into one bitmap using OR logic. An index
scan-key selection method is initiated using (primary) index PIX. For each entry in index PIX, the bitmap is
checked. If the entry is selected by the bitmap, then the data space row is retrieved and processed.

The messages created by the PRTSQLINF CL command, when used to describe this query, would look
like:
SQL4008 Index PIX used for table 1.
SQL4032 Index TIX1 used for bitmap processing of table 1.
CPI4328 Access path of file PIX was used by query.
CPI4338 2 access path(s) used for bitmap processing of file EMPLOYEE.

Example: Bitmap processing used in conjunction with join queries

Bitmap processing can be used for join queries, as well. Since bitmap processing is on a per-table basis,
each table of a join can independently use or not use bitmap processing.

26 DB2 UDB for iSeries Database Performance and Query Optimization V5R1
The following example illustrates a query where bitmap processing is used against the second table of a
join query but not on the first table:
CREATE INDEX EPIX ON EMPLOYEE(EMPNO)
CREATE INDEX TIX1 ON EMPLOYEE(WORKDEPT)
CREATE INDEX TIX2 ON EMPLOYEE(SALARY)
DECLARE C1 CURSOR FOR
SELECT * FROM PROJECT, EMPLOYEE
WHERE RESEMP=EMPNO AND
(WORKDEPT=’E01’ OR SALARY>50000)

Using the OPNQRYF command:

OPNQRYF FILE((PROJECT) (EMPLOYEE)) FORMAT(RESULTFILE)
JFLD((1/RESPEMP 2/EMPNO))
QRYSLT(’2/WORKDEPT=’’E01’’ *OR 2/SALARY>50000’)

In this example, the optimizer decides that the join order is table PROJECT to table EMPLOYEE. Table
scan is used on table PROJECT. For table EMPLOYEE, index EPIX is used to process the join (primary
index). Indexes TIX1 and TIX2 are used in bitmap processing.

The database manager positions to the first row in table PROJECT. It then performs the join using index
EPIX. Next, it generates a bitmap from the results of applying selection WORKDEPT=’E01’ against index
TIX1 (using index scan-key positioning). It then generates a bitmap from the results of applying selection
SALARY>50000 against index TIX2 (again using index scan-key positioning).

Next, the database manager combines these two bitmaps into one bitmap using OR logic. Finally, the
entry that EPIX is currently positioned to is checked against the bitmap. The entry is either selected or
rejected by the bitmap. If the entry is selected, the rows are retrieved from the underlying data space.
Next, index EPIX is probed for the next join row. When an entry is found, it is compared against the
bitmap and either selected or rejected. Note that the bitmap was generated only once (the first time it was
needed) and is just reused after that.

The query optimizer debug messages put into the job log would look like:
CPI4327 File PROJECT processed in join position 1.
CPI4326 File EMPLOYEE processed in join position 2.
CPI4338 2 access path(s) used for bitmap processing of file EMPLOYEE.

Bitmap processing and composite key indexes

Bitmap processing alleviates some of the problems associated with having composite key indexes
(multiple key columns in one index).

For example, given an SQL query:

DECLARE C1 CURSOR FOR
SELECT * FROM EMPLOYEE
WHERE WORKDEPT=’D11’ AND
FIRSTNAME IN (’DAVID’, ’BRUCE’, ’WILLIAM’)

Or the same query using the OPNQRYF command:

OPNQRYF FILE((EMPLOYEE))
QRYSLT(’WORKDEPT=’’D11’’ *AND
FIRSTNME = %VALUES(’’DAVID’’ ’’BRUCE’’ ’’WILLIAM’’)’)

An index with keys (WORKDEPT, FIRSTNAME) would be the best index to use to satisfy this query.
However, two indexes, one with a key of WORKDEPT and the other with a key of FIRSTNME could be
used in bitmap processing, with their resulting bitmaps ANDed together and table scan used to retrieve the
result.

Chapter 2. Data access on DB2 UDB for iSeries: data access paths and methods 27
 With the bitmap processing method, you can create several indexes, each with only one key column, and
have the optimizer use them as general purpose indexes for many queries. You can avoid problems
involved with trying to determine the best composite key indexes for all queries being performed against a
table. Bitmap processing, in comparison to using a multiple key-column index, allows more ease of use,
but at some cost to performance.

Note: Keep in mind that you will always achieve the best performance by using composite key indexes.

Considerations for bitmap processing

Some additional points regarding bitmap processing:

v As long as the DB2 UDB Symmetric Multiprocessing feature is installed, you can use parallel
processing whenever you use bitmap processing. In this case, the bitmap is built from the results of
performing either parallel index scan-key positioning or parallel index scan-key selection on the tertiary
index.
v Bitmaps are preprocessed at the first I/O request for the query. Therefore, the first row fetched may take
longer to retrieve than subsequent rows.
| v Bitmaps, by their nature, contain static selection, which is controlled by ALWCPYDTA. Once the bitmap
| is generated, it will not select any new or modified rows. Therefore, bitmap processing is used only
| when ALWCPYDTA(*OPTIMIZE) is specified.
| For example, suppose that an OPNQRYF statement specifying (QRYSLT(’QUANTITY >5’) is opened
| using bitmap processing and the first row is read. While updates to the table will not cause a record to
| be selected, if you reverse this example where the bitmap was generated before the update and the
| update caused the record not to be selected, the optimizer will ensure that you do not get this record
| because of duplicate selection that is applied to the underlying dataspace once it is retrieved. Through a
| separate database operation, all rows where QUANTITY is equal to 4 are updated so QUANTITY is
| equal to 10. Since the bitmap was already built (during the first row fetch from the OPNQRYF open
| identifier), these updated rows will not be retrieved on subsequent fetches through the OPNQRYF open
| identifier.
| The exception to this is when the query contains grouping or one or more aggregate functions (for
| example, SUM, COUNT, MIN, MAX), in which case static data is already being made.
| v The query optimizer does not use bitmap processing for a query that is insert, update, or
| delete-capable. For OPNQRYF, you must set the OPTION parameter to *INP and the SEQONLY
| parameter to *YES. There must not be any overrides to SEQONLY(*NO)).

| Sort access method

| The sort access method provides an alternative method for those queries that must process data in an
| ordered method (ORDER BY). An index can be used to sort the data, and is effective in some cases for
| implementing ordering. However, if the optimizer had to create a temporary index for that query, it would
| use extra processor time and resources when creating the index before the requested query can be run.

| Where this method can be used

| The optimizer chooses this method in the following circumstances:

| v For SQL (see “Effects of the ALWCPYDTA parameter on database performance” on page 126) when
| you specify either of the following precompile options:
| – ALWCPYDTA(*OPTIMIZE), ALWBLK(*ALLREAD), and COMMIT(*CHG or *CS)
| – ALWCPYDTA(*OPTIMIZE) and COMMIT(*NONE)
| v For OPNQRYF, when you specify either of the following options:
| – ALWCPYDTA(*OPTIMIZE) and COMMIT(*NO)
| – ALWCPYDTA(*OPTIMIZE) and COMMIT(*YES) and commitment control is started with a commit
| level of *NONE, or *CHG, or *CS

28 DB2 UDB for iSeries Database Performance and Query Optimization V5R1
| v If a temporary result is required prior to the ordering function.
| v If the number of order by keys exceeds 120 or the combined length of the sort keys exceeds 2000
| bytes.

| How this method works

| The sort algorithm reads the rows into a sort space and sorts the rows based on the specified ordering
| keys. The rows are then returned to the user from the ordered sort space.

Chapter 2. Data access on DB2 UDB for iSeries: data access paths and methods 29
30 DB2 UDB for iSeries Database Performance and Query Optimization V5R1
Chapter 3. The DB2 UDB for iSeries query optimizer:
Overview
This overview of the query optimizer provides guidelines for designing queries that will perform and will
use system resources more efficiently. This overivew covers queries that are optimized by the query
optimizer and includes interfaces such as SQL, OPNQRYF, APIs (QQQQRY), ODBC, and Query/400
queries. Whether or not you apply the guidelines, the query results will still be correct.

Note: The information in this overview is complex. You might find it helpful to experiment with an iSeries
system as you read this information to gain a better understanding of the concepts.

When you understand how DB2 Universal Database for iSeries processes queries, it is easier to
understand the performance impacts of the guidelines discussed in this overview. There are two major
components of DB2 Universal Database for iSeries query processing:
v How the system accesses data. See “Data access: data access methods” on page 3.
These methods are the algorithms that are used to retrieve data from the disk. The methods include
index usage and row selection techniques. In addition, parallel access methods are available with the
DB2 UDB Symmetric Multiprocessing operating system feature.
v Query optimizer. See “How the query optimizer makes your queries more efficient”.
The query optimizer identifies the valid techniques which could be used to implement the query and
selects the most efficient technique.

How the query optimizer makes your queries more efficient

The optimizer is an important part of DB2 Universal Database for iSeries because the optimizer:
v Makes the key decisions which affect database performance.
v Identifies the techniques which could be used to implement the query.
v Selects the most efficient technique.
Data manipulation statements such as SELECT specify only what data the user wants, not how to retrieve
that data. This path to the data is chosen by the optimizer and stored in the access plan. This topic covers
the techniques employed by the query optimizer for performing this task including:
v “Cost estimation for queries” on page 32
v “Access plan validation” on page 34
v “Join optimization” on page 35
v “Grouping optimization” on page 50

Optimizer decision-making rules

The optimizer uses a general set of guidelines to choose the best method for accessing data. The
optimizer:
v Determines the default filter factor for each predicate in the selection clause.
v Extracts attributes of the table from internally stored information.
v Performs a key range estimate to determine the true filter factor of the predicates when the selection
predicates match the left-most keys of an index.
v Determines the cost of table scan processing if an index is not required.
v Determines the cost of creating an index over a table if an index is required. This index is created by
performing either a table scan or creating an index-from-index.
v Determines the cost of using a sort routine or hashing method if selection conditions apply and an index
is required.

© Copyright IBM Corp. 2000, 2001 31

v For each index available, generally in the order of most recently created to oldest, the optimizer does
the following until its time limit is exceeded:
– Extracts attributes of the index from internally stored statistics.
– Determines if the index meets the selection criteria.
– Determines the cost of using the index by using the estimated page faults and the predicate filter
factors to help determine the cost.
– Compares the cost of using this index with the previous cost (current best).
– Picks the cheaper one.
– Continues to search for best index until its time limit is exceeded or no more indexes.

The time limit controls how much time the optimizer spends choosing an implementation. It is based on
how much time was spent so far and the current best implementation cost found. The idea is to prevent
the optimizer from spending more time optimizing the query than it would take to actually execute the
query. Dynamic SQL queries are subject to the optimizer time restrictions. Static SQL queries
optimization time is not limited. For OPNQRYF, if you specify OPTALLAP(*YES), the optimization time is
not limited.

For small tables, the query optimizer spends little time in query optimization. For large tables, the query
optimizer considers more indexes. Generally, the optimizer considers five or six indexes (for each table of
a join) before running out of optimization time. Because of this, it is normal for the optimizer to spend
longer lengths of time analyzing queries against larger tables.

Cost estimation for queries

At run time, the optimizer chooses an optimal access method for the query by calculating an
implementation cost based on the current state of the database. The optimizer models the access cost of
each of the following:
v Reading rows directly from the table (table scan processing)
v Reading rows through an index (using either index scan-key selection or index scan-key positioning)
v Creating an index directly from the dataspace
v Creating an index from an existing index (index-from-index)
v Using the query sort routine or hashing method (if conditions are satisfied)
v Using bitmap processing

The cost of a particular method is the sum of:

v The start-up cost
v The cost associated with the given optimization mode.
– Costs associated with optimization modes when using SQL: For SQL, the precompile option
ALWCPYDTA and the OPTIMIZE FOR n ROWS clause indicate to the query optimizer the
optimization goal to be achieved.
- The optimizer can optimize SQL queries with one of two goals:
1. Minimize the time required to retrieve the first buffer of rows from the table. This goal biases
the optimization towards not creating an index.
Either a data scan or an existing index is preferred. This mode can be specified in two ways:
a. The OPTIMIZE FOR n ROWS allows the users to specify the number of rows they expect
to retrieve from the query.
The optimizer uses this value to determine the percentage of rows that will be returned
and optimizes accordingly. A small value instructs the optimizer to minimize the time
required to retrieve the first n rows.
b. Specifying ALWCPYDTA(*NONE) or (*YES) as a precompiler parameter allows the
optimizer to minimize the time that it requires to retrieve the first 3% of the resulting rows.

32 DB2 UDB for iSeries Database Performance and Query Optimization V5R1
 This option is effective only if the OPTIMIZE FOR n ROWS was not specified.
2. Minimize the time to process the whole query assuming that all selected rows are returned to
the application. This option does not bias the optimizer to any particular access method. This
mode can be specified in two ways:
a. The OPTIMIZE FOR n ROWS allows the users to specify the number of rows they expect
to retrieve from the query.
The optimizer uses this value to determine the percentage of rows that will be returned
and optimizes accordingly. A value greater than or equal to the expected number of
resulting rows instructs the optimizer to minimize the time required to run the entire query.
b. ALWCPYDTA(*OPTIMIZE) specified as a precompiler parameter.
This option is effective only if the OPTIMIZE FOR n ROWS is not specified.
– Costs associated with optimization modes when using OPNQRYF:
- The cost associated with the given optimization parameter (*FIRSTIO, *ALLIO, or *MINWAIT).
v *FIRSTIO — Minimize the time required to retrieve the first buffer of rows from the table. Biases
the optimization toward not creating an index. Either a data scan or an existing index is
preferred. When *FIRSTIO is selected, users may also pass in the number of rows they expect
to retrieve from the query. The optimizer uses this value to determine the percentage of rows
that will be returned and optimizes accordingly. A small value would minimize the time required
to retrieve the first n rows, similar to *FIRSTIO. A large value would minimize the time to
retrieve all n rows, similar to *ALLIO.
v *ALLIO — Minimize the time to process the whole query assuming that all query rows are read
from the table. This option does not bias the optimizer to any particular access method.

Note: If you specify ALWCPYDTA(*OPTIMIZE) and the query optimizer decides to use the sort
routine, your query resolves according to the *ALLIO optimize parameter.
v *MINWAIT–Minimize delays when reading rows from the table. Minimize I/O time at the
expense of open time. This option biases optimization toward either creating a temporary index
or performing a sort. Either an index is created or an existing index is used.
v The cost of any index creations
v The cost of the expected number of page faults to read the rows and the cost of processing the
expected number of rows.
Page faults and number of rows processed may be predicted by statistics the optimizer can obtain from
the database objects, including:
– Table size
– Row size
– Index size
– Key size

Page faults can also be greatly affected if index only access can be performed, thus eliminating any
random input and output to the dataspace.

A weighted measure of the expected number of rows to process is based on what the relational
operators in the row selection predicates, default filter factors, are likely to retrieve:
– 10% for equal
– 33% for less-than, greater-than, less-than-equal-to, or greater-than-equal-to
– 90% for not equal
– 25% for BETWEEN range (OPNQRYF %RANGE)
– 10% for each IN list value (OPNQRYF %VALUES)

Chapter 3. The DB2 UDB for iSeries query optimizer: Overview 33

 Key range estimate is a method the optimizer uses to gain more accurate estimates of the number of
expected rows to be selected from one or more selection predicates. The optimizer estimates by
applying the selection predicates against the left-most keys of an existing index. The default filter factors
can then be further refined by the estimate based on the key range. If an index exists whose left-most
keys match columns used in row selection predicates, that index can be used to estimate the number of
entries that match the selection criteria. The estimate of the number of entries is based on the number
of pages and key density of the machine index and is done without actually accessing the entries.
Creating indexes over columns that are used in selection predicates can significantly help optimization.

Page faults and the number of rows processed are dependent on the type of access the optimizer
chooses. Refer to Chapter 2, “Data access on DB2 UDB for iSeries: data access paths and methods”
on page 3 for more information on access methods.

General query optimization tips

Here are some tips to help your queries run as fast as possible:
v Create indexes whose leftmost key columns match your selection predicates to help supply the
optimizer with selectivity values (key range estimates).
v For join queries, create indexes that match your join columns to help the optimizer determine the
average number of matching rows.
v Minimize extraneous mapping by specifying only columns of interest on the query. For example, specify
only the columns you need to query on the SQL SELECT statement instead of specifying SELECT *.
Also, you should specify FOR FETCH ONLY if the columns do not need to be updated.
v If your queries often use table scan access method, use the RGZPFM (Reorganize Physical File
Member) command to remove deleted rows from tables or the CHGPF (Change Physical File)
REUSEDLT (*YES) command to reuse deleted rows.

For embedded SQL, consider using the following precompile options:

v Specify ALWCPYDTA(*OPTIMIZE) to allow the query optimizer to create temporary copies of data so
better performance can be obtained.
v Use CLOSWLCSR(*ENDJOB) or CLOSQLCSR(*ENDACTGRP) to allow open data paths to remain
open for future invocations.
v Specify DLYPRP(*YES) to delay SQL statement validation until an OPEN, EXECUTE, or DESCRIBE
statement is run. This option improves performance by eliminating redundant validation.
v Use ALWBLK(*ALLREAD) to allow row blocking for read-only cursors.

For OPNQRYF (Open Query File) queries, consider using the following parameters:
v Use ALWCPYDTA(*OPTIMIZE) to let the query optimizer create temporary copies of data if it can obtain
better performance by doing so.
v Use OPTIMIZE(*FIRSTIO) to bias the optimizer to use an existing index instead of creating a temporary
index.

Access plan validation

An access plan is a control structure that describes the actions necessary to satisfy each query request.
An access plan contains information about the data and how to extract it. For any query, whenever
optimization occurs, the query optimizer develops an optimized plan of how to access the requested data.
The information is kept in what is called a mini plan. The mini plan, along with the query definition
template (QDT), is used to interface with the optimizer and make an access plan.
v For dynamic SQL, an access plan is created, but the plan is not saved. A new access plan is created
each time the PREPARE statement runs.
v For an iSeries program that contains static embedded SQL, the access plan is saved in the associated
space of the program or package that contains embedded SQL statements.

34 DB2 UDB for iSeries Database Performance and Query Optimization V5R1
v For OPNQRYF, an access plan is created but is not saved. A new access plan is created each time the
OPNQRYF command is processed.
v For Query/400, an access plan is saved as part of the query definition object.

Join optimization
A join operation is a complex function that requires special attention in order to achieve good performance.
This section describes how DB2 Universal Database for iSeries implements join queries and how
optimization choices are made by the query optimizer. It also describes design tips and techniques which
help avoid or solve performance problems. Among the topics discussed are:
v Nested loop join implementation
v Hash joins
v Cost estimation and index selection for join secondary dials
v Tips for improving the performance of join queries

Nested loop join implementation

DB2 Universal Database for iSeries provides a nested loop join method. For this method, the processing
of the tables in the join are ordered. This order is called the join order. The first table in the final join
order is called the primary table. The other tables are called secondary tables. Each join table position
is called a dial. During the join, DB2 Universal Database for iSeries:
1. Accesses the first primary table row selected by the predicates local to the primary table.
2. Builds a key value from the join columns in the primary table.
3. Uses index scan-key positioning to locate the first row that satisfies the join condition for the first
secondary table using an index with keys matching the join condition or local row selection columns of
the secondary table.
4. Applies bitmap selection, if applicable.
5. Determines if the row is selected by applying any remaining selection local to the first secondary dial.
If the secondary dial row is not selected then the next row that satisfies the join condition is located.
Steps 1 through 5 are repeated until a row that satisfies both the join condition and any remaining
selection is selected from all secondary tables
6. Returns the result join row.
7. Processes the last secondary table again to find the next row that satisfies the join condition in that
dial.
During this processing, when no more rows that satisfy the join condition can be selected, the
processing backs up to the logical previous dial and attempts to read the next row that satisfies its join
condition.
8. Ends processing when all selected rows from the primary table are processed.

Nested loop join characteristics:

Note the following characteristics of a nested loop join:

v If ordering or grouping is specified and all the columns are over a single table and that table is eligible
to be the primary, then that table becomes the primary table and is processed with an index over the
table.
v If ordering and grouping is specified on two or more tables, DB2 Universal Database for iSeries breaks
the processing of the query into two parts:
1. Perform the join selection omitting the ordering or grouping processing and write the result rows to a
temporary work table. This allows the optimizer to consider any table of the join query as a
candidate for the primary table.
2. The ordering or grouping processing is then performed on the data in the temporary work table.

Chapter 3. The DB2 UDB for iSeries query optimizer: Overview 35

 The query optimizer might also decide to break the query into these two parts to improve performance
when the SQL ALWCPYDTA(*OPTIMIZE) precompiler parameter or the OPNQRYF KEYFLD, and
ALWCPYDTA(*OPTIMIZE) parameters are specified.
v All rows that satisfy the join condition from each secondary dial are located using an index. Rows are
retrieved from secondary tables in random sequence. This random disk I/O time often accounts for a
large percentage of the processing time of the query. Since a given secondary dial is searched once for
each row selected from the primary and the preceding secondary dials that satisfy the join condition for
each of the preceding secondary dials, a large number of searches may be performed against the later
dials. Any inefficiencies in the processing of the later dials can significantly inflate the query processing
time. This is the reason why attention to performance considerations for join queries can reduce the
run-time of a join query from hours to minutes.
v Again, all selected rows from secondary dials are accessed through an index. If an efficient index
cannot be found, a temporary index is created. Some join queries build temporary indexes over
secondary dials even when an index exists for all of the join keys. Because efficiency is very important
for secondary dials of longer running queries, the query optimizer may choose to build a temporary
index which contains only entries which pass the local row selection for that dial. This preprocessing of
row selection allows the database manager to process row selection in one pass instead of each time
rows are matched for a dial.

Hash join
The hash join method is similar to nested loop join. Instead of using indexes to locate the matching rows
in a secondary table, however, a hash temporary result table is created that contains all of the rows
selected by local selection against the table. The structure of the hash table is such that rows with the
same join value are loaded into the same hash table partition (clustered). The location of the rows for any
given join value can be found by applying a hashing function to the join value.

Advantages of hash joins over nested loop joins

Hash join has several advantages over nested loop join:

v The structure of a hash temporary result table is simpler than that of an index, so less CPU processing
is required to build and probe a hash table.
v The rows in the hash result table contain all of the data required by the query so there is no need to
access the dataspace of the table with random I/O when probing the hash table.
v Like join values are clustered, so all matching rows for a given join value can usually be accessed with
a single I/O request.
v The hash temporary result table can be built using SMP parallelism.
v Unlike indexes, entries in hash tables are not updated to reflect changes of column values in the
underlying table. The existence of a hash table does not affect the processing cost of other updating
jobs in the system.

Queries that cannot use hash join

Hash join cannot be used for queries that:

v Perform subqueries unless all subqueries in the query can be transformed to inner joins.
v Perform a UNION or UNION ALL.
v Perform left outer or exception join.
v Use a DDS created join logical file.
v Require live access to the data as specified by the *NO or *YES parameter values for the ALWCPYDTA
precompiler parameter. Hash join is used only for queries running with ALWCPYDTA(*OPTIMIZE). This
parameter can be specified either on precompiler commands, the STRSQL CL command, or the
OPNQRYF CL command. The Client Access/400 ODBC driver and Query Management driver always
uses this mode.
v Hash join can be used with OPTIMIZE(*YES) if a temporary result is required to run the query.

36 DB2 UDB for iSeries Database Performance and Query Optimization V5R1
v Hash join cannot be used for queries involving physical files or tables that have read triggers.
v Require that the cursor position be restored as the result of the SQL ROLLBACK HOLD statement or
the ROLLBACK CL command. For SQL applications using commitment control level other than *NONE,
this requires that *ALLREAD be specified as the value for the ALWBLK precompiler parameter.

Hash join and parallel processing

The query attribute DEGREE, which can be changed by using the Change Query attribute CL command
(CHGQRYA), does not enable or disable the optimizer from choosing to use hash join. However, hash join
queries can use SMP parallelism if the query attribute DEGREE is set to either *OPTIMIZE, *MAX, or
*NBRTASKS.

Types of queries that can effectively use hash join

Hash join is used in many of the same cases where a temporary index would have been built. Join queries
which are most likely to be implemented using hash join are those where either:
v All rows in the various tables of the join are involved in producing result rows.
v Significant non-join selection is specified for the tables of the join which reduces the number of rows in
the tables that are involved with the join result.

Example of hash join that processes all rows

The following is an example of a join query that would process all of the rows from the queried tables:
SELECT *
FROM EMPLOYEE, EMP_ACT
WHERE EMPLOYEE.EMPNO = EMP_ACT.EMPNO
OPTIMIZE FOR 99999999 ROWS

OPNQRYF example :
OPNQRYF FILE((EMPLOYEE EMP_ACT)) FORMAT(FORMAT1)
JFLD((1/EMPNO 2/EMPNO *EQ))
ALWCPYDTA(*OPTIMIZE)

This query is implemented using the following steps:

1. A temporary hash table is built over table EMP_ACT with a key of EMPNO. This occurs when the
query is opened.
2. For each row retrieved from the EMPLOYEE table, the temporary hash table will be probed for any
matching join values.
3. For each matching row found, a result row is returned.

The messages created by the PRTSQLINF CL command to describe this hash join query in an SQL
program would appear as follows:
SQL402A Hashing algorithm used to process join.
SQL402B Table EMPLOYEE used in hash join step 1.
SQL402B Table EMP_ACT used in hash join step 2.

Example of hash join on query that is limited by local selection

The following is an example of a join query that would have the queried tables of the join queried
significantly reduced by local selection:
SELECT EMPNO, LASTNAME, DEPTNAME
FROM EMPLOYEE, DEPARTMENT
WHERE EMPLOYEE.WORKDEPT = DEPARTMENT.DEPTNO
AND EMPLOYEE.HIREDATE BETWEEN 1996-01-30 AND 1995-01-30
AND DEPARTMENT.DEPTNO IN (’A00’, ’D01’, ’D11’, ’D21’, ’E11’)
OPTIMIZE FOR 99999999 ROWS

Chapter 3. The DB2 UDB for iSeries query optimizer: Overview 37

OPNQRYF example:
OPNQRYF FILE((EMPLOYEE DEPARTMENT))
FORMAT(FORMAT2)
QRYSLT(’1/HIREDATE *EQ %RANGE(’’1996-01-30’’ ’’1995-01-30’’)
*AND 2/DEPTNO *EQ %VALUES(’’A00’’ ’’D01’’ ’’D11’’ ’’D21’’
’’E11’’’)
JFLD((1/WORKDEPT 2/DEPTNO *EQ))
ALWCPYDTA(*OPTIMIZE)

This query is implemented using the following steps:

1. A temporary hash table is built over table DEPARTMENT with key values of DEPTNO containing rows
matching the selection predicate, DEPTNO IN (’A00’, ’D01’, ’D11’, ’D21’, ’E11’). This occurs when the
query is opened.
2. For each row retrieved from the EMPLOYEE table matching the selection predicate, HIREDATE
BETWEEN 1996-01-30 and 1995-01-30, the temporary hash table will be probed for the matching join
values.
3. For each matching row found, a result row is returned.

The messages created by the PRTSQLINF CL command to describe this hash join query in an SQL
program would appear as follows:
SQL402A Hashing algorithm used to process join.
SQL402B Table EMPLOYEE used in hash join step 1.
SQL402B Table DEPARTMENT used in hash join step 2.

Example of hash join on query where ordering, grouping, non-equal selection, or result columns
are selected

When ordering, grouping, non-equal selection specified with operands derived from columns of different
tables, or result columns are derived from columns of different tables, the hash join processing will be
done and the result rows of the join will be written to a temporary table. Then, as a second step, the query
will be completed using the temporary table.

The following is an example of a join query with selection specified with operands derived from columns of
different tables:
SELECT EMPNO, LASTNAME, DEPTNAME
FROM EMPLOYEE, DEPARTMENT
WHERE EMPLOYEE.WORKDEPT = DEPARTMENT.DEPTNO
AND EMPLOYEE.EMPNO > DEPARTMENT.MGRNO
OPTIMIZE FOR 99999999 ROWS

OPNQRYF example:
OPNQRYF FILE((EMPLOYEE DEPARTMENT)
FORMAT(FORMAT2)
JFLD((1/WORKDEPT 2/DEPTNO *EQ) (1/EMPNO 2/MGRNO
*GT))

This query is implemented using the following steps:

1. A temporary hash table is built over table DEPARTMENT with a key of DEPTNO. This occurs when the
query is opened.
2. For each row retrieved from the EMPLOYEE table, the temporary hash table will be probed for the
matching join values.
3. For each matching row found, a result row is written to a temporary table.
4. After all of the join result rows are written to the temporary table, rows that are selected by EMPNO >
MGRNO are read from the temporary table and returned to the application.

38 DB2 UDB for iSeries Database Performance and Query Optimization V5R1
The messages created by the PRTSQLINF CL command to describe this hash join query in an SQL
program would appear as follows:
SQL402A Hashing algorithm used to process join.
SQL402B Table EMPLOYEE used in hash join step 1.
SQL402B Table DEPARTMENT used in hash join step 2.
SQL402C Temporary result table created for hash join query.

Join optimization algorithm

The query optimizer must determine the join columns, join operators, local row selection, index usage, and
dial ordering for a join query.

The join columns and join operators depend on the:

v Join column specifications of the query
v Join order
v Interaction of join columns with other row selection
v Index used.

Join specifications which are not implemented for the dial are either deferred until they can be processed
in a later dial or, if an inner join was being performed for this dial, processed as row selection.

For a given dial, the only join specifications which are usable as join columns for that dial are those being
joined to a previous dial. For example, for the second dial the only join specifications that can be used to
satisfy the join condition are join specifications which reference columns in the primary dial. Likewise, the
third dial can only use join specifications which reference columns in the primary and the second dials and
so on. Join specifications which reference later dials are deferred until the referenced dial is processed.

For any given dial, only one type of join operator is normally implemented. For example, if one inner join
specification has a join operator of ’=’ and the other has a join operator of ’>’, the optimizer attempts to
implement the join with the ’=’ operator. The ’>’ join specification is processed as row selection after a
matching row for the ’=’ specification is found. In addition, multiple join specifications that use the same
operator are implemented together.

Note: For OPNQRYF, only one type of join operator is allowed for either a left outer or an exception join.
That is, the join operator for all join conditions must be the same.

When looking for an existing index to access a secondary dial, the query optimizer looks at the left-most
key columns of the index. For a given dial and index, the join specifications which use the left-most key
columns can be used. For example:
DECLARE BROWSE2 CURSOR FOR
SELECT * FROM EMPLOYEE, EMP_ACT
WHERE EMPLOYEE.EMPNO = EMP_ACT.EMPNO
AND EMPLOYEE.HIREDATE = EMP_ACT.EMSTDATE
OPTIMIZE FOR 99999 ROWS

OPNQRYF example:
OPNQRYF FILE((EMPLOYEE, EMP_ACT)) FORMAT(FORMAT1)
JFLD((1/EMPNO 2/EMP_ACT *EQ)(1/HIREDATE 2/EMSTDATE
*EQ))

For the index over EMP_ACT with key columns EMPNO, PROJNO, and EMSTDATE, the join operation is
performed only on column EMPNO. After the join is performed, index scan-key selection is done using
column EMSTDATE.

The query optimizer also uses local row selection when choosing the best use of the index for the
secondary dial. If the previous example had been expressed with a local predicate as:

Chapter 3. The DB2 UDB for iSeries query optimizer: Overview 39

 DECLARE BROWSE2 CURSOR FOR
SELECT * FROM EMPLOYEE, EMP_ACT
WHERE EMPLOYEE.EMPNO = EMP_ACT.EMPNO
AND EMPLOYEE.HIREDATE = EMP_ACT.EMSTDATE
AND EMP_ACT.PROJNO = ’123456’
OPTIMIZE FOR 99999 ROWS

OPNQRYF example:
OPNQRYF FILE((EMPLOYEE, EMP_ACT)) FORMAT(FORMAT2)
QRYSLT(’2/PROJNO *EQ ’’123456’’’)
JFLD((1/EMPNO 2/EMP_ACT *EQ)(1/HIREDATE 2/EMSTDATE
*EQ))

the index with key columns EMPNO, PROJNO, and EMSTDATE are fully utilized by combining join and
selection into one operation against all three key columns.

When creating a temporary index, the left-most key columns are the usable join columns in that dial
position. All local row selection for that dial is processed when selecting entries for inclusion into the
temporary index. A temporary index is similar to the index created for a select/omit keyed logical file. The
temporary index for the previous example would have key columns of EMPNO and EMSTDATE.

Since the OS/400 query optimizer attempts a combination of join and local row selection when determining
access path usage, it is possible to achieve almost all of the same advantages of a temporary index by
use of an existing index. In the above example, using either implementation, an existing index may be
used or a temporary index may be created. A temporary index would have been built with the local row
selection on PROJNO applied during the index’s creation; the temporary index would have key columns of
EMP_ACT and EMSTDATE (to match the join selection). If, instead, an existing index was used with key
columns of EMP_ACT, PROJNO, EMSTDATE (or PROJNO, EMP_ACT, EMSTDATE or EMSTDATE,
PROJNO, EMP_ACT or ...) the local row selection could be applied at the same time as the join selection
(rather than prior to the join selection, as happens when the temporary index is created, or after the join
selection, as happens when only the first key column of the index matches the join column).

The implementation using the existing index is more likely to provide faster performance because join and
selection processing are combined without the overhead of building a temporary index. However, the use
of the existing index may have just slightly slower I/O processing than the temporary index because the
local selection is run many times rather than once. In general, it is a good idea to have existing indexes
available with key columns for the combination of join columns and columns using equal selection as the
left-most keys.

Join order optimization

The join order is fixed if any join logical files are referenced. The join order is also fixed if the OPNQRYF
JORDER(*FILE) parameter is specified or the query options file (QAQQINI) FORCE_JOIN_ORDER
parameter is *YES. Otherwise, the following join ordering algorithm is used to determine the order of the
tables:
1. Determine an access method for each individual table as candidates for the primary dial.
2. Estimate the number of rows returned for each table based on local row selection.
If the join query with row ordering or group by processing is being processed in one step, then the
table with the ordering or grouping columns is the primary table.
3. Determine an access method, cost, and expected number of rows returned for each join combination
of candidate tables as primary and first secondary tables.
The join order combinations estimated for a four table inner join would be:

1-2 2-1 1-3 3-1 1-4 4-1 2-3 3-2 2-4 4-2 3-4 4-3
4. Choose the combination with the lowest join cost.
If the cost is nearly the same, then choose the combination which selects the fewest rows.

40 DB2 UDB for iSeries Database Performance and Query Optimization V5R1
5. Determine the cost, access method, and expected number of rows for each remaining table joined to
the previous secondary table.
6. Select an access method for each table that has the lowest cost for that table.
7. Choose the secondary table with the lowest join cost.
If the cost is nearly the same, choose the combination which selects the fewest rows.
8. Repeat steps 4 on page 40 through 7 until the lowest cost join order is determined.

Note: After dial 32, the optimizer uses a different method to determine file join order, which may not be
the lowest cost.

When a query contains a left or right outer join or a right exception join, the join order is not fixed.
However, all from-columns of the ON clause must occur from dials previous to the left or right outer or
exception join. For example:
FROM A INNER JOIN B ON A.C1=B.C1
LEFT OUTER JOIN C ON B. C2=C.C2

The allowable join order combinations for this query would be:

1–2–3, 2–1–3, or 2–3–1

Right outer or right exception joins are implemented as left outer and left exception, respectively with files
flipped. For eaxmple:
FROM A RIGHT OUTER JOIN B ON A.C1=B.C1

is implemented as B LEFT OUTER JOIN A ON B.C1=A.C1. The only allowed join order is 2–1.

When a join logical file is referenced or the join order is forced to the specified table order, the query
optimizer loops through all of the dials in the order specified, and determines the lowest cost access
methods.

Cost estimation and index selection for join secondary dials

In step 3 on page 40 and in step 5, the query optimizer has to estimate a cost and choose an access
method for a given dial combination. The choices made are similar to those for row selection except that
an index must be used.

As the query optimizer compares the various possible access choices, it must assign a numeric cost value
to each candidate and use that value to determine the implementation which consumes the least amount
of processing time. This costing value is a combination of CPU and I/O time and is based on the following
assumptions:
v Table pages and index pages must be retrieved from auxiliary storage. For example, the query optimizer
is not aware that an entire table may be loaded into active memory as the result of a SETOBJACC CL
command. Usage of this command may significantly improve the performance of a query, but the query
optimizer does not change the query implementation to take advantage of the memory resident state of
the table.
v The query is the only process running on the system. No allowance is given for system CPU utilization
or I/O waits which occur because of other processes using the same resources. CPU related costs are
scaled to the relative processing speed of the system running the query.
v The values in a column are uniformly distributed across the table. For example, if 10% of the rows in a
table have the same value, then it is assumed that every tenth row in the table contains that value.
v The values in a column are independent from the values in any other columns in a row. For example, if
a column named A has a value of 1 in 50% of the rows in a table and a column named B has a value of
2 in 50% of the rows, then it is expected that a query which selects rows where A = 1, and B = 2
selects 25% of the rows in the table.

Chapter 3. The DB2 UDB for iSeries query optimizer: Overview 41

The main factors of the join cost calculations for secondary dials are the number of rows selected in all
previous dials and the number of rows which match, on average, each of the rows selected from previous
dials. Both of these factors can be derived by estimating the number of matching rows for a given dial.

When the join operator is something other than equal, the expected number of matching rows is based on
the following default filter factors:
v 33% for less-than, greater-than, less-than-equal-to, or greater-than-equal-to
v 90% for not equal
v 25% for BETWEEN range (OPNQRYF %RANGE)
v 10% for each IN list value (OPNQRYF %VALUES)

For example, when the join operator is less-than, the expected number of matching rows is .33 * (number
of rows in the dial). If no join specifications are active for the current dial, the cartesian product is assumed
to be the operator. For cartesian products, the number of matching rows is every row in the dial, unless
local row selection can be applied to the index.

When the join operator is equal, the expected number of rows is the average number of duplicate rows for
a given value.

The iSeries performs index maintenance (insertion and deletion of key values in an index) and maintains a
running count of the number of unique values for the given key columns in the index. These statistics are
bound with the index object and are always maintained. The query optimizer uses these statistics when it
is optimizing a query. Maintaining these statistics adds no measurable amount of overhead to index
maintenance. This statistical information is only available for indexes which:
v Contain no varying length character keys.

Note: If you have varying length character columns used as join columns, you can create an index
which maps the varying length character column to a fixed character key using the CRTLF CL
command. An index that contains fixed length character keys defined over varying length data
supplies average number of duplicate values statistics.
v Were created or rebuilt on an iSeries system on which Version 2 Release 3 or a later version is
installed.

Note: The query optimizer can use indexes created on earlier versions of OS/400 to estimate if the join
key values have a high or low average number of duplicate values. If the index is defined with
only the join keys, the estimate is done based on the size of the index. In many cases, additional
keys in the index cause matching row estimates through that index to not be valid. The
performance of some join queries may be improved by rebuilding these indexes.

Average number of duplicate values statistics are maintained only for the first 4 left-most keys of the index.
For queries which specify more than 4 join columns, it might be beneficial to create multiple additional
indexes so that an index can be found with average number of duplicate values statistics available within
the 4 left-most key columns. This is particularly important if some of the join columns are somewhat
unique (low average number of duplicate values).

42 DB2 UDB for iSeries Database Performance and Query Optimization V5R1
Figure 1. Average number of duplicate values of a 3 key index

These statistics are maintained as part of index rebuild and creation.

Using the average number of duplicate values for equal joins or the default filter value for the other join
operators, we now have the number of matching rows. The following formula is used to compute the
number of join rows from previous dials.
NPREV = Rp * M2 * FF2 * ..... *Mn * FFn .....
NPREV
The number of join rows from all previous dials.
Rp The number of rows selected from the primary dial.
M2 The number of matching rows for dial 2.
FF2 Filtering reduction factor for predicates local to dial 2 that are not already applied using M2 above.
Mn The number of matching rows for dial n.
FFn Filtering reduction factor for predicates local to dial n that are not already applied using Mn above.

Note: Multiply the pair of matching rows (Mn) and filter reduction filter factors (FFn) for each
secondary dial preceding the current dial.

Now that it has calculated the number of join rows from previous dials, the optimizer is ready to generate a
cost for the access method.

Chapter 3. The DB2 UDB for iSeries query optimizer: Overview 43

Temporary index or hash temporary result table from table: The first access method choice analyzed
by the query optimizer is building a temporary index or hash temporary result table from the table. The
basic formula for costing access of a join secondary dial through a temporary index built from the table or
hash table follows:
JSCOST = CRTDSI +
NPREV *((MATCH * FF * KeyAccess)
+ (MATCH * FF * FCost)) *
FirstIO
JSCOST
Join Secondary cost
CRTDSI
Cost to build the temporary index or a hash temporary result table
NPREV
The number of join rows from all previous dials
MATCH
The number of matching rows (usually average duplicates)
KeyAccess
The cost to access a key in an index or a hash table
FF The filtering factor for local predicates of this dial (excluding selection performed on earlier dials
because of transitive closure)
FCost The cost to access a row from the table
FirstIO
A reduction ratio to reduce the non-startup cost because of an optimization goal to optimize for the
first buffer retrieval. For more information, see “Cost estimation for queries” on page 32.

This secondary dial access method is used if no usable index is found or if the temporary index or hash
table performs better than any existing index. This method can be better than using any existing index
because the row selection is completed when the index or hash table is created if any of the following are
true:
v The number of matches (MATCH) is high.
v The number of join rows from all previous dials (NPREV) is high.
v There is some filtering reduction (FF < 100%).

Temporary index or hash table from index: The basic cost formula for this access method choice is
the same as that of using a temporary index or hash table built from a table, with one exception. The cost
to build the temporary index, CRTDSI, is calculated to include the selection of the rows through an existing
index. This access method is used for join secondary dial access for the same reason. However, the
creation from an index might be less costly.

Use an existing index: The final access method is to use an existing index. The basic formula for
costing access of a join secondary dial through an existing index is:
JSCOST = NPREV *((MATCH * KeyAccess)
+ (MATCH * FCost)) *
FirstIO
JSCOST
Join Secondary cost
NPREV
The number of join rows from all previous dials
MATCH
The number of matching keys which will be found in this index (usually average duplicates)

44 DB2 UDB for iSeries Database Performance and Query Optimization V5R1
KeyAccess
The cost to access a key in an index
FCost The cost to access a row from the table
FirstIO
A reduction ratio to reduce the non-startup cost because of an optimization goal to optimize for the
first buffer retrieval. For more information, see “Cost estimation for queries” on page 32.

If I/O optimization is used (that is, OPNQRYF OPTIMIZE(*FIRSTIO)), this is a likely access method because
the entire cost is reduced. Also, if the number of join rows from all previous dials (NPREV), and the
number of matching keys (MATCH) is low, this may be the most efficient method.

The query optimizer considers using an index which only has a subset of the join columns as the left-most
leading keys when:
v It is able to determine from the average number of duplicate values statistics that the average number
of rows with duplicate values is quite low.
v The number of rows being selected from the previous dials is small.

Predicates generated through transitive closure

For join queries, the query optimizer may do some special processing to generate additional selection.
When the set of predicates that belong to a query logically infer extra predicates, the query optimizer
generates additional predicates. The purpose is to provide more information during join optimization.

Example of predicates being added because of transitive closure:

SELECT * FROM EMPLOYEE, EMP_ACT
WHERE EMPLOYEE.EMPNO = EMP_ACT.EMPNO
AND EMPLOYEE.EMPNO = ’000010’

The optimizer will modify the query to be:

SELECT * FROM EMPLOYEE, EMP_ACT
WHERE EMPLOYEE.EMPNO = EMP_ACT.EMPNO
AND EMPLOYEE.EMPNO = ’000010’
AND EMP_ACT.EMPNO = ’000010’

OPNQRYF example:
OPNQRYF FILE((EMPLOYEE EMP_ACT)) FORMAT(FORMAT1)
QRYSLT(’1/EMPNO *EQ ’’000010’’’)
JFLD((1/EMPNO 2/EMPNO *EQ))

The optimizer will modify the query to be:

OPNQRYF FILE((EMPLOYEE EMP_ACT)) FORMAT(FORMAT1)
QRYSLT(’1/EMPNO *EQ ’’000010’’ *AND
2/EMPNO *EQ ’’000010’’’)
JFLD((1/EMPNO 2/EMPNO *EQ))

The following rules determine which predicates are added to other join dials:
v The dials affected must have join operators of equal.
v The predicate is isolatable, which means that a false condition from this predicate would omit the row.
v One operand of the predicate is an equal join column and the other is a constant or host variable.
v The predicate operator is not LIKE or IN (OPNQRYF %WLDCRD, %VALUES, or *CT).
v The predicate is not connected to other predicates by OR.
v The join type for the dial is an inner join.

The query optimizer generates a new predicate, whether or not a predicate already exists in the WHERE
clause (OPNQRYF QRYSLT parameter).

Chapter 3. The DB2 UDB for iSeries query optimizer: Overview 45

Some predicates are redundant. This occurs when a previous evaluation of other predicates in the query
already determines the result that predicate provides. Redundant predicates can be specified by you or
generated by the query optimizer during predicate manipulation. Redundant predicates with predicate
operators of =, >, >=, <, <=, or BETWEEN (OPNQRYF *EQ, *GT, *GE, *LT, *LE, or %RANGE) are merged
into a single predicate to reflect the most selective range.

Multiple join types for a query

Even though multiple join types (inner, left outer, right outer, left exception, and right exception) can be
specified in the query using the JOIN syntax, the iSeries Licensed Internal Code can only support one join
type of inner, left outer, or left exception join type for the entire query. This requires the optimizer to
determine what the overall join type for the query should be and to reorder files to achieve the correct
semantics.

Note: This section does not apply to OPNQRYF.

The optimizer will evaluate the join criteria along with any row selection that may be specified in order to
determine the join type for each dial and for the entire query. Once this information is known the optimizer
will generate additional selection using the relative row number of the tables to simulate the different types
of joins that may occur within the query.

Since null values are returned for any unmatched rows for either a left outer or an exception join, any
isolatable selection specified for that dial, including any additional join criteria that may be specified in the
WHERE clause, will cause all of the unmatched rows to be eliminated (unless the selection is for an IS
NULL predicate). This will cause the join type for that dial to be changed to an inner join (or an exception
join) if the IS NULL predicate was specified.

In the following example a left outer join is specified between the tables EMPLOYEE and DEPARTMENT.
In the WHERE clause there are two selection predicates that also apply to the DEPARTMENT table.
SELECT EMPNO, LASTNAME, DEPTNAME, PROJNO
FROM CORPDATA.EMPLOYEE XXX LEFT OUTER JOIN CORPDATA.DEPARTMENT YYY
ON XXX.WORKDEPT = YYY.DEPTNO
LEFT OUTER JOIN CORPDATA.PROJECT ZZZ
ON XXX.EMPNO = ZZZ.RESPEMP
WHERE XXX.EMPNO = YYY.MGRNO AND
YYY.DEPTNO IN (’A00’, ’D01’, ’D11’, ’D21’, ’E11’)

The first selection predicate, XXX.EMPNO = YYY.MGRNO, is an additional join condition that will be
added to the join criteria and evaluated as an ″inner join″ join condition. The second is an isolatable
selection predicate that will eliminate any unmatched rows. Either one of these selection predicates will
cause the join type for the DEPARTMENT table to be changed from a left outer join to an inner join.

Even though the join between the EMPLOYEE and the DEPARTMENT table was changed to an inner join
the entire query will still need to remain a left outer join to satisfy the join condition for the PROJECT table.

Note: Care must be taken when specifying multiple join types since they are supported by appending
selection to the query for any unmatched rows. This means that the number of resulting rows that
satisfy the join criteria can become quite large before any selection is applied that will either select
or omit the unmatched rows based on that individual dial’s join type.

For more information on how to use the JOIN syntax see either Joining Data from More Than One Table in
the SQL Programming Concepts book or the SQL Reference book.

Sources of join query performance problems

The optimization algorithms described above benefit most join queries, but the performance of a few
queries may be degraded. This occurs when:
v An index is not available which provides average number of duplicate values statistics for the potential
join columns.

46 DB2 UDB for iSeries Database Performance and Query Optimization V5R1
 Note: “Cost estimation and index selection for join secondary dials” on page 41 provides suggestions
on how to avoid the restrictions about indexes statistics or create additional indexes over the
potential join columns if they do not exist.
v The query optimizer uses default filter factors to estimate the number of rows being selected when
applying local selection to the table because indexes do not exist over the selection columns.
Creating indexes over the selection columns allows the query optimizer to make a more accurate
filtering estimate by using key range estimates.
v The particular values selected for the join columns yield a significantly greater number of matching rows
than the average number of duplicate values for all values of the join columns in the table (i.e. the data
is not uniformly distributed).
Use DDS to build a logical file with an index with select/omit specifications matching the local row
selection. This provides the query optimizer with a more accurate estimate of the number of matching
rows for the keys which are selected.

Note: The optimizer can better determine from the select/omit index that the data is not uniformly
distributed.
v The query optimizer makes the wrong assumption about the number of rows which will be retrieved
from the answer set.
For SQL programs, specifying the precompile option ALWCPYDTA(*YES) makes it more likely that the
queries in that program will use an existing index. Likewise, specifying ALWCPYDTA(*OPTIMIZE)
makes it more likely that the queries in that program will create a temporary index. The SQL clause
OPTIMIZE FOR n ROWS can also be used to influence the query optimizer.
For the OPNQRYF command, the wrong performance option for the OPTIMIZE keyword may have
been specified. Specify *FIRSTIO to make the use of an existing index more likely. Specify *ALLIO to
make the creation of a temporary index more likely.

Tips for improving the performance of join queries

If you are looking at a join query which is performing poorly or you are about to create a new application
which uses join queries, the following checklist may be useful.
Table 2. Checklist for Creating an Application that Uses Join Queries
What to Do How It Helps
Check the database design. Make This gives the query optimizer a better opportunity to select an efficient access
sure that there are indexes method because it can determine the average number of duplicate values. Many
available over all of the join queries may be able to use the existing index to implement the query and avoid
columns and/or row selection the cost of creating a temporary index.
columns. If using CRTLF, make
sure that the index is not shared.
Check the query to see whether Since the query optimizer does not add predicates for predicates connected by
some complex predicates should OR or non-isolatable predicates, or predicate operators of LIKE or IN, modifying
be added to other dials to allow the query by adding these predicates may help.
the optimizer to get a better idea
of the selectivity of each dial.
Create an index which includes This step helps if the statistical characteristics are not uniform for the entire table.
Select/Omit specifications which For example, if there is one value which has a high duplication factor and the rest
match that of the query using of the column values are unique, then a select/omit index allows the optimizer to
CRTLF CL command. skew the distribution of values for that key and make the right optimization for the
selected values.

Chapter 3. The DB2 UDB for iSeries query optimizer: Overview 47

Table 2. Checklist for Creating an Application that Uses Join Queries (continued)
What to Do How It Helps
Specify ALWCPYDTA(*OPTIMIZE) If the query is creating a temporary index, and you feel that the processing time
or ALWCPYDTA(*YES) would be better if the optimizer only used the existing index, specify
ALWCPYDTA(*YES).

If the query is not creating a temporary index, and you feel that the processing
time would be better if a temporary index was created, specify
ALWCPYDTA(*OPTIMIZE).

Alternatively, specify the OPTIMIZE FOR n ROWS to inform the optimizer of the
application has intention to read every resulting row. To do this set n to a large
number. You could also set n to a small number before ending the query.
For OPNQRYF, specify If the query is creating a temporary index and you feel that the processing time
OPTIMIZE(*FIRSTIO) or would be better if it would only use the existing index, then specify
OPTIMIZE(*ALLIO) OPTIMIZE(*FIRSTIO). If the query is not creating a temporary index and you feel
that the processing time would be better if a temporary index was created then
specify OPTIMIZE(*ALLIO).

48 DB2 UDB for iSeries Database Performance and Query Optimization V5R1
Table 2. Checklist for Creating an Application that Uses Join Queries (continued)
What to Do How It Helps
Use a join logical file or use the A join in which one table is joined with all secondary tables consecutively is
query options file (QAQQINI) sometimes called a star join. In the case of a star join where all secondary join
FORCE_JOIN_ORDER parameter predicates contain a column reference to a particular table, there may be
of *YES. OPNQRYF users can performance advantages if that table is placed in join position one. In Example A,
specify JORDER(*FILE). all tables are joined to table EMPLOYEE. The query optimizer can freely
determine the join order. The query should be changed to force EMPLOYEE into
join position one by using the query options file (QAQQINI)
FORCE_JOIN_ORDER parameter of *YES or OPNQRYF JORDER(*FILE) as
shown in example B. Note that in these examples the join type is a join with no
default values returned (this is an inner join.). The reason for forcing the table into
the first position is to avoid random I/O processing. If EMPLOYEE is not in join
position one, every row in EMPLOYEE could be examined repeatedly during the
join process. If EMPLOYEE is fairly large, considerable random I/O processing
occurs resulting in poor performance. By forcing EMPLOYEE to the first position,
random I/O processing is minimized.

Example A: Star join query

DECLARE C1 CURSOR FOR
SELECT * FROM DEPARTMENT, EMP_ACT, EMPLOYEE,
PROJECT
WHERE DEPARTMENT.DEPTNO=EMPLOYEE.WORKDEPT
AND EMP_ACT.EMPNO=EMPLOYEE.EMPNO
AND EMPLOYEE.WORKDEPT=PROJECT.DEPTNO

Example B: Star join query with order forced via FORCE_JOIN_ORDER

DECLARE C1 CURSOR FOR
SELECT * FROM EMPLOYEE, DEPARTMENT, EMP_ACT,
PROJECT
WHERE DEPARTMENT.DEPTNO=EMPLOYEE.WORKDEPT
AND EMP_ACT.EMPNO=EMPLOYEE.EMPNO
AND EMPLOYEE.WORKDEPT=PROJECT.DEPTNO

Example A: Star join query (OPNQRYF)

OPNQRYF FILE((DEPARTMENT EMP_ACT EMPLOYEE
PROJECT)) FORMAT(FORMAT1)
JFLD((1/DEPTNO 3/WORKDEPT *EQ)
(2/EMPNO 3/EMPNO *EQ)
(3/WORKDEPT 4/DEPTNO *EQ))

Example B: Start join query (OPNQRYF) with JORDER(*FILE) parameter

OPNQRYF FILE((EMPLOYEE DEPARTMENT EMP_ACT
PROJECT)) FORMAT(FORMAT1)
JFLD((2/DEPTNO 1/WORKDEPT *EQ)
(3/EMPNO 1/EMPNO *EQ)
(1/WORKDEPT 4/DEPTNO *EQ))
JORDER(*FILE)

Note: Specifying columns from EMPLOYEE in the ORDER BY clause

(OPNQRYF KEYFLD parameter) may also have the effect of placing EMPLOYEE
in join position 1. This allows the query optimizer to choose the best order for the
remaining tables.
Specify ALWCPYDTA(*OPTIMIZE) In the cases where ordering is specified and all key columns are from a single
to allow the query optimizer to use dial, this allows the query optimizer to consider all possible join orders.
a sort routine.
Specify join predicates to prevent This improves performance by reducing the join fan-out. Every secondary table
all of the rows from one table from should have at least one join predicate that references on of its columns as a
being joined to every row in the ’join-to’ column.
other table.

Chapter 3. The DB2 UDB for iSeries query optimizer: Overview 49

Tips for improving performance when selecting data from more than two tables
If the select-statement you are considering accesses two or more tables, all the recommendations
suggested in Chapter 5, “Using indexes to speed access to large tables” on page 97 apply. The following
suggestion is directed specifically to select-statements that access several tables. For joins that involve
more than two tables, you might want to provide redundant information about the join columns. If you give
the optimizer extra information to work with when requesting a join. It can determine the best way to do
the join. The additional information might seem redundant, but is helpful to the optimizer. For example,
instead of coding:
EXEC SQL
DECLARE EMPACTDATA CURSOR FOR
SELECT LASTNAME, DEPTNAME, PROJNO, ACTNO
FROM CORPDATA.DEPARTMENT, CORPDATA.EMPLOYEE,
CORPDATA.EMP_ACT
WHERE DEPARTMENT.MGRNO = EMPLOYEE.EMPNO
AND EMPLOYEE.EMPNO = EMP_ACT.EMPNO
END-EXEC.

When using the OPNQRYF command, specify:

OPNQRYF FILE(CORPDATA/DEPARTMENT CORPDATA/EMPLOYEE CORPDATA/EMP_ACT)
FORMAT(FORMAT1)
JFLD((1/MGRNO 2/EMPNO *EQ) (2/EMPNO 3/EMP_ACT *EQ))

Provide the optimizer with a little more data and code:

EXEC SQL
DECLARE EMPACTDATA CURSOR FOR
SELECT LASTNAME, DEPTNAME, PROJNO, ACTNO
FROM CORPDATA.DEPARTMENT, CORPDATA.EMPLOYEE,
CORPDATA.EMP_ACT
WHERE DEPARTMENT.MGRNO = EMPLOYEE.EMPNO
AND EMPLOYEE.EMPNO = EMP_ACT.EMPNO
AND DEPARTMENT.MGRNO = EMP_ACT.EMPNO
END-EXEC.

When using the OPNQRYF command, specify:

OPNQRYF FILE(CORPDATA/DEPARTMENT CORPDATA/EMPLOYEE CORPDATA/EMP_ACT)
FORMAT(FORMAT1)
JFLD((1/MGRNO 2/EMPNO *EQ) (2/EMPNO 3/EMP_ACT *EQ)
(1/MGRNO 3/EMPNO *EQ))

Grouping optimization
This section describes how DB2 Universal Database for iSeries implements grouping techniques and how
optimization choices are made by the query optimizer. The query optimizer has two choices for
implementing grouping: the hash implementation or the index implementation.

Grouping hash implementation

This technique uses the base hash access method to perform grouping or summarization of the selected
table rows. For each selected row, the specified grouping value is run through the hash function. The
computed hash value and grouping value are used to quickly find the entry in the hash table
corresponding to the grouping value. If the current grouping value already has a row in the hash table, the
hash table entry is retrieved and summarized (updated) with the current table row values based on the
requested grouping column operations (such as SUM or COUNT). If a hash table entry is not found for the
current grouping value, a new entry is inserted into the hash table and initialized with the current grouping
value.

The time required to receive the first group result for this implementation will most likely be longer than
other grouping implementations because the hash table must be built and populated first. Once the hash

50 DB2 UDB for iSeries Database Performance and Query Optimization V5R1
table is completely populated, the database manager uses the table to start returning the grouping results.
Before returning any results, the database manager must apply any specified grouping selection criteria or
ordering to the summary entries in the hash table.

Where the grouping hash method is most effective

The grouping hash method is most effective when the consolidation ratio is high. The consolidation ratio
is the ratio of the selected table rows to the computed grouping results. If every database table row has its
own unique grouping value, then the hash table will become too large. This in turn will slow down the
hashing access method.

The optimizer estimates the consolidation ratio by first determining the number of unique values in the
specified grouping columns (that is, the expected number of groups in the database table). The optimizer
then examines the total number of rows in the table and the specified selection criteria and uses the result
of this examination to estimate the consolidation ratio.

Indexes over the grouping columns can help make the optimizer’s ratio estimate more accurate. Indexes
improve the accuracy because they contain statistics that include the average number of duplicate values
for the key columns.

The optimizer also uses the expected number of groups estimate to compute the number of partitions in
the hash table. As mentioned earlier, the hashing access method is more effective when the hash table is
well-balanced. The number of hash table partitions directly affects how entries are distributed across the
hash table and the uniformity of this distribution.

The hash function performs better when the grouping values consist of columns that have non-numeric
data types, with the exception of the integer (binary) data type. In addition, specifying grouping value
columns that are not associated with the variable length and null column attributes allows the hash
function to perform more effectively.

Index grouping implementation

This implementation utilizes the index scan-key selection or index scan-key positioning access methods to
perform the grouping. An index is required that contains all of the grouping columns as contiguous leftmost
key columns. The database manager accesses the individual groups through the index and performs the
requested summary functions.

Since the index, by definition, already has all of the key values grouped together, the first group result can
be returned in less time than the hashing method. This is because of the temporary result that is required
for the hashing method. This implementation can be beneficial if an application does not need to retrieve
all of the group results or if an index already exists that matches the grouping columns.

When the grouping is implemented with an index and a permanent index does not already exist that
satisfies grouping columns, a temporary index is created. The grouping columns specified within the query
are used as the key columns for this index.

Optimizing grouping by eliminating grouping columns

All of the grouping columns are evaluated to determine if they can be removed from the list of grouping
columns. Only those grouping columns that have isolatable selection predicates with an equal operator
specified can be considered. This guarantees that the column can only match a single value and will not
help determine a unique group. This processing is done to allow the optimizer to consider more indexes to
implement the query and to reduce the number of columns that will be added as key columns to a
temporary index or hash table.

The following example illustrates a query where the optimizer could eliminate a grouping column.

Chapter 3. The DB2 UDB for iSeries query optimizer: Overview 51

 DECLARE DEPTEMP CURSOR FOR
SELECT EMPNO, LASTNAME, WORKDEPT
FROM CORPDATA.EMPLOYEE
WHERE EMPNO = ’000190’
GROUP BY EMPNO, LASTNAME, WORKDEPT

OPNQRYF example:
OPNQRYF FILE(EMPLOYEE) FORMAT(FORMAT1)
QRYSLT(’EMPNO *EQ ’’000190’’’)
GRPFLD(EMPNO LASTNAME WORKDEPT)

In this example, the optimizer can remove EMPNO from the list of grouping columns because of the EMPNO
= ’000190’ selection predicate. An index that only has LASTNAME and WORKDEPT specified as key
columns can be considered to implement the query and if a temporary index or hash is required then
EMPNO will not be used.

Note: Even though EMPNO can be removed from the list of grouping columns, the optimizer might still
choose to use that index if a permanent index exists with all three grouping columns.

| Optimizing grouping by removing read triggers

| For queries involving physical files or tables with read triggers, group by triggers will always involve a
| temporary file prior to the group by processing, and will therefore slow down these queries.

| Note: Read triggers are added when the ADDPFTRG command has been used on the table with
| TRGTIME (*AFTER) and TRGEVENT (*READ).

| The query will run faster is the read trigger is removed (RMVPFTRG TRGTIME (*AFTER) TRGEVENT
| (*READ)).

Optimizing grouping by adding additional grouping columns

The same logic that is applied to removing grouping columns can also be used to add additional grouping
columns to the query. This is only done when you are trying to determine if an index can be used to
implement the grouping.

The following example illustrates a query where the optimizer could add an additional grouping column.
CREATE INDEX X1 ON EMPLOYEE
(LASTNAME, EMPNO, WORKDEPT)

DECLARE DEPTEMP CURSOR FOR

SELECT LASTNAME, WORKDEPT
FROM CORPDATA.EMPLOYEE
WHERE EMPNO = ’000190’
GROUP BY LASTNAME, WORKDEPT

OPNQRYF example:
OPNQRYF FILE ((EMPLOYEE)) FORMAT(FORMAT1)
QRYSLT(’EMPNO *EQ ’’000190’’’)
GRPFLD(LASTNAME WORKDEPT)

For this query request, the optimizer can add EMPNO as an additional grouping column when considering
X1 for the query.

Optimizing grouping by using index skip key processing

Index Skip Key processing can be used when grouping with the keyed sequence implementation algorithm
which uses an existing index. The index skip key processing algorithm:
1. Uses the index to position to a group and

52 DB2 UDB for iSeries Database Performance and Query Optimization V5R1
2. finds the first row matching the selection criteria for the group, and if specified the first non-null MIN or
MAX value in the group
3. Returns the group to the user
4. ″Skip″ to the next group and repeat processing
This will improve performance by potentially not processing all index key values for a group.

Index skip key processing can be used:

v For single table queries using the keyed sequence grouping implementation when:
– There are no column functions in the query, or
– There is only a single MIN or MAX column function in the query and the operand of the MIN or MAX
is the next key column in the index after the grouping columns. There can be no other grouping
functions in the query. For the MIN function, the key column must be an ascending key; for the MAX
function, the key column must be a descending key. If the query is whole table grouping, the
operand of the MIN or MAX must be the first key column.
Example 1, using SQL:
CREATE INDEX IX1 ON EMPLOYEE (SALARY DESC)

DECLARE C1 CURSOR FOR

SELECT MAX(SALARY) FROM EMPLOYEE;

Example 1, using the OPNQRYF command:

OPNQRYF FILE(EMPLOYEE) FORMAT(FORMAT1)
MAPFLD((MAXSAL ’%MAX(SALARY)’))

The query optimizer will chose to use the index IX1. The SLIC runtime code will scan the index until
it finds the first non-null value for SALARY. Assuming that SALARY is not null, the runtime code will
position to the first index key and return that key value as the MAX of salary. No more index keys
will be processed.

Example 2, using SQL:

CREATE INDEX IX2 ON EMPLOYEE (DEPT, JOB,SALARY)

DECLARE C1 CURSOR FOR

SELECT DEPT, MIN(SALARY)
FROM EMPLOYEE
WHERE JOB=’CLERK’
GROUP BY DEPT

Example 2, using the OPNQRYF command:

OPNQRYF FILE(EMPLOYEE) FORMAT(FORMAT2)
QRYSLT(’JOB *EQ ’’CLERK’’’)
GRPFLD((DEPT))
MAPFLD((MINSAL ’%MIN(SALARY)’))

The query optimizer will chose to use Index IX2. The SLIC runtime code will position to the first
group for DEPT where JOB equals ’CLERK’ and will return the SALARY. The code will then skip to
the next DEPT group where JOB equals ’CLERK’.
v For join queries:
– All grouping columns must be from a single table.
– For each dial there can be at most one MIN or MAX column function operand that references the
dial and no other column functions can exist in the query.
– If the MIN or MAX function operand is from the same dial as the grouping columns, then it uses the
same rules as single table queries.

Chapter 3. The DB2 UDB for iSeries query optimizer: Overview 53

 – If the MIN or MAX function operand is from a different dial then the join column for that dial must join
to one of the grouping columns and the index for that dial must contain the join columns followed by
the MIN or MAX operand.
Example 1, using SQL:
CREATE INDEX IX1 ON DEPARTMENT(DEPTNAME)

CREATE INDEX IX2 ON EMPLOYEE(WORKDEPT, SALARY)

DECLARE C1 CURSOR FOR

SELECT DEPTNAME, MIN(SALARY)
FROM DEPARTMENT, EMPLOYEE
WHERE DEPARTMENT.DEPTNO=EMPLOYEE.WORKDEPT
GROUP BY DEPARTMENT.DEPTNO;

Example 1, using the OPNQRYF command:

OPNQRYF FILE(DEPARTMENT EMPLOYEE) FORMAT(FORMAT1)
JFLD((1/DEPTNO 2/WORKDEPT *EQ))
GRPFLD((1/DEPTNO))
MAPFLD((MINSAL ’%MIN(SALARY)’))

Ordering optimization
This section describes how DB2 Universal Database for iSeries implements ordering techniques, and how
optimization choices are made by the query optimizer. The query optimizer can use either index ordering
or a sort to implement ordering.

Sort Ordering implementation

The sort algorithm reads the rows into a sort space and sorts the rows based on the specified ordering
keys. The rows are then returned to the user from the ordered sort space.

Index Ordering implementation

The index ordering implementation requires an index that contains all of the ordering columns as
contiguous leftmost key columns. The database manager accesses the individual rows through the index
in index order, which results in the rows being returned in order to the requester.

This implementation can be beneficial if an application does not need to retrieve all of the ordered results,
or if an index already exists that matches the ordering columns. When the ordering is implemented with an
index, and a permanent index does not already exist that satisfies ordering columns, a temporary index is
created. The ordering columns specified within the query are used as the key columns for this index.

Optimizing ordering by eliminating ordering columns

All of the ordering columns are evaluated to determine if they can be removed from the list of ordering
columns. Only those ordering columns that have isolatable selection predicates with an equal operator
specified can be considered. This guarantees that the column can match only a single value, and will not
help determine in the order.

This processing is done to allow the optimizer to consider more indexes as it implements the query, and to
reduce the number of columns that will be added as key columns to a temporary index. The following SQL
example illustrates a query where the optimizer could eliminate an ordering column.
DECLARE DEPTEMP CURSOR FOR
SELECT EMPNO, LASTNAME, WORKDEPT
FROM CORPDATA.EMPLOYEE
WHERE EMPNO = ’000190’
ORDER BY EMPNO, LASTNAME, WORKDEPT

OPNQRYF example:

54 DB2 UDB for iSeries Database Performance and Query Optimization V5R1
OPNQRYF FILE(EMPLOYEE) FORMAT(FORMAT1)
QRYSLT(’EMPNO *EQ ’’000190’’’)
KEYFLD(EMPNO LASTNAME WORKDEPT)

In this example, the optimizer can remove EMPNO from the list of ordering columns because of the
EMPNO = ’000190’ selection predicate. An index that has only LASTNAME and WORKDEPT specified as
key columns can be considered to implement the query; if a temporary index is required, then EMPNO will
not be used.

Note: Even though EMPNO can be removed from the list of ordering columns, the optimizer might still
choose to use that index if a permanent index exists with all three ordering columns.

Optimizing ordering by adding additional ordering columns

The same logic that is applied to removing ordering columns can also be used to add additional grouping
columns to the query. This is done only when you are trying to determine if an index can be used to
implement the ordering.

The following example illustrates a query where the optimizer could add an additional ordering column.
CREATE INDEX X1 ON EMPLOYEE (LASTNAME, EMPNO, WORKDEPT)

DECLARE DEPTEMP CURSOR FOR

SELECT LASTNAME, WORKDEPT
FROM CORPDATA.EMPLOYEE
WHERE EMPNO = ’000190’
ORDER BY LASTNAME, WORKDEPT

OPNQRYF example:
OPNQRYF FILE ((EMPLOYEE)) FORMAT(FORMAT1)
QRYSLT(’EMPNO *EQ ’’000190’’’)
KEYFLD(LASTNAME WORKDEPT)

For this query request, the optimizer can add EMPNO as an additional ordering column when considering
X1 for the query.

View implementation
Views are implemented by the query optimizer using one of two methods:
v The optimizer combines the query select statement with the select statement of the view (view
composite)
v The optimizer places the results of the view in a temporary table and then replaces the view reference
in the query with the temporary table (view materialization)

This also applies to nested table expressions and common table expressions except where noted.

View composite implementation

The view composite implementation takes the query select statement and combines it with the select
statement of the view to generate a new query. The new, combined select statement query is then run
directly against the underlying base tables.

This single, composite statement is the preferred implementation for queries containing views, since it
requires only a single pass of the data.

Examples:
CREATE VIEW D21EMPL AS
SELECT * FROM CORPDATA.EMPLOYEE
WHERE WORKDEPT=’D21’

Using SQL:

Chapter 3. The DB2 UDB for iSeries query optimizer: Overview 55

 SELECT LASTNAME, FIRSTNME, SALARY
FROM D21EMPL
WHERE JOB=’CLERK’

Using OPNQRYF:
OPNQRYF FILE(D21EMPL)
FORMAT(FORMAT1)
QRYSLT(’JOB *EQ ’’CLERK’’’)

The query optimizer will generate a new query that looks like the following example:
SELECT LASTNAME, FIRSTNME, SALARY
FROM CORPDATA.EMPLOYEE
WHERE WORKDEPT=’D21’ AND JOB=’CLERK’

The query contains the columns selected by the user’s query, the base tables referenced in the query, and
the selection from both the view and the user’s query.

Note: The new composite query that the query optimizer generates is not visible to users. Only the
original query against the view will be seen by users and database performance tools.

View materialization implementation

The view materialization implementation runs the query of the view and places the results in a temporary
result table. The view reference in the user’s query is then replaced with the temporary table, and the
query is run against the temporary result table.

View materialization is done whenever it is not possible to create a view composite. The following types of
queries require view materialization:
v The outermost select of the view contains grouping, the query contains grouping, and refers to a column
derived from a column function in the view in the HAVING or select-list.
v The query is a join and the outermost select of the view contains grouping or DISTINCT.
v The outermost select of the view contains DISTINCT, and the query has UNION, grouping, or DISTINCT
and one of the following:
– Only the query has a shared weight NLSS table
– Only the view has a shared weight NLSS table
– Both the query and the view have a shared weight NLSS table, but the tables are different.
v The query contains a column function and the outermost select of the view contains a DISTINCT
v The view does not contain an access plan. This can occur when a view references a view and a view
composite cannot be created because of one of the reasons listed above. This does not apply to nested
table expressions and common table expressions.
Since a temporary result table is created, access methods that are allowed with ALWCPYDTA(*OPTIMIZE)
may be used to implement the query. These methods include hash grouping, hash join, and bitmaps.

Examples:
CREATE VIEW AVGSALVW AS
SELECT WORKDEPT, AVG(SALARY) AS AVGSAL
FROM CORPDATA.EMPLOYEE
GROUP BY WORKDEPT

SQL example:
SELECT D.DEPTNAME, A.AVGSAL
FROM CORPDATA.DEPARTMENT D, AVGSALVW A
WHERE D.DEPTNO=A.WORKDEPT

OPNQRYF example:

56 DB2 UDB for iSeries Database Performance and Query Optimization V5R1
OPNQRYF FILE(CORPDATA/DEPARTMENT AVGSALVW)
FORMAT(FORMAT1)
JFLD((1/DEPTNO 2/WORKDEPT *EQ))

In this case, a view composite cannot be created since a join query references a grouping view. The
results of AVGSALVW are placed in a temporary result table (*QUERY0001). The view reference
AVGSALVW is replaced with the temporary result table. The new query is then run. The generated query
looks like the following:
SELECT D.DEPTNAME, A.AVGSAL
FROM CORPDATA.DEPARTMENT D, *QUERY0001 A
WHERE D.DEPTNO=A.WORKDEPT

Note: The new query that the query optimizer generates is not visible to users. Only the original query
against the view will be seen by users and database performance tools.

Whenever possible, isolatable selection from the query, except subquery predicates, is added to the view
materialization process. This results in smaller temporary result tables and allows existing indexes to be
used when materializing the view. This will not be done if there is more than one reference to the same
view or common table expression in the query. The following is an example where isolatable selection is
added to the view materialization:
SELECT D.DEPTNAME,A.AVGSAL
FROM CORPDATA.DEPARTMENT D, AVGSALVW A
WHERE D.DEPTNO=A.WORKDEPT
A.WORKDEPT LIKE ’D%’ AND AVGSAL>10000

OPNQRYF example:
OPNQRYF FILE(CORPDATA/DEPARTMENT AVGSALVW)
FORMAT(FORMAT1)
JFLD((1/DEPTNO 2/WORKDEPT *EQ))
QRYSLT(’1/WORKDEPT *EQ %WLDCRD(’’D*’’) *AND 2/AVGSAL *GT 10000’)

The isolatable selection from the query is added to view resulting in a new query to generate the
temporary result table:
SELECT WORKDEPT, AVG(SALARY) AS AVGSAL
FROM CORPDATA.EMPLOYEE
WHERE WORKDEPT LIKE ’D%’
GROUP BY WORKDEPT
HAVING AVG(SALARY)>10000

Chapter 3. The DB2 UDB for iSeries query optimizer: Overview 57

58 DB2 UDB for iSeries Database Performance and Query Optimization V5R1
Chapter 4. Optimizing query performance using query
optimization tools
You can use query optimization tools to improve data retrieval time. Use the results of the tools to:
v Change the data access method chosen by the system. See “Data access methods: Summary” on
page 5.
v Create the correct indexes and use them effectively. See Chapter 5, “Using indexes to speed access to
large tables” on page 97.

Query optimization is an iterative process. Do the following as needed to optimize your queries.

Gather statistics about your queries

There are various ways to gather statistics about your queries. The following is a sampling of the ways
that statistics can be gathered:
v “Verify the performance of SQL applications” on page 60
v “Examine query optimizer debug messages in the job log” on page 60
v “Gather information about embedded SQL statements with the PRTSQLINF command” on page 67
v “Gather statistics about your queries with the database monitor” on page 69
v “Gather statistics about your queries with memory-resident database monitor APIs” on page 78
v “View the effectiveness of your queries with Visual Explain” on page 80
v Use the Operations Navigator SQL Performance monitor.

Control the processing of your queries:

v “Change the attributes of your queries with the Change Query Attributes (CHGQRYA) command” on
page 81
v “Control queries dynamically with the query options file QAQQINI” on page 82
v “Control long-running queries with the DB2 UDB for iSeries Predictive Query Governor” on page 89
v “Control parallel processing for queries” on page 92

Comparing the different tools:

You may want to check out the “Query optimization tools: Comparison table” on page 95 to learn:
v What information each tool can yield about your query
v When in the process a specific tool can analyze your query
v The tasks each tool can perform to improve your query

For additional tips and techniques:

If you are experienced with query optimization, you may want to refer to a list of “General query
optimization tips” on page 34.

Also, the following topics provide programming tips and techniques for optimizing your applications for
query performance:
v Chapter 6, “Application design tips for database performance” on page 111
v Chapter 7, “Programming techniques for database performance” on page 119
v Chapter 8, “General DB2 UDB for iSeries performance considerations” on page 125

© Copyright IBM Corp. 2000, 2001 59

Verify the performance of SQL applications
You can verify the performance of an SQL application by using the following commands:
DSPJOB
You can use the Display Job (DSPJOB) command with the OPTION(*OPNF) parameter to show
the indexes and tables being used by an application that is running in a job.
You can also use DSPJOB with the OPTION(*JOBLCK) parameter to analyze object and row lock
contention. It displays the objects and rows that are locked and the name of the job holding the
lock.
Specify the OPTION(*CMTCTL) parameter on the DSPJOB command to show the isolation level
that the program is running, the number of rows being locked during a transaction, and the
pending DDL functions. The isolation level displayed is the default isolation level. The actual
isolation level, used for any SQL program, is specified on the COMMIT parameter of the
CRTSQLxxx command.
PRTSQLINF
The Print SQL Information (PRTSQLINF) command lets you print information about the embedded
SQL statements in a program, SQL package, or service program. The information includes the
SQL statements, the access plans used during the running of the statement, and a list of the
command parameters used to precompile the source member for the object. For more information
on printing information about SQL Statements, see the PRTSQLINF section in “Gather information
about embedded SQL statements with the PRTSQLINF command” on page 67.
STRDBMON
You can use the Start Database Monitor (STRDBMON) command to capture to a file information
about every SQL statement that runs. See “Gather statistics about your queries with the database
monitor” on page 69 for more information.
CHGQRYA
You can use the Change Query Attribute (CHGQRYA) command to change the query attributes for
the query optimizer. Among the attributes that can be changed by this command are the predictive
query governor, parallelism, and the query options.
STRDBG
You can use the Start Debug (STRDBG) command to put a job into debug mode and, optionally,
add as many as 20 programs and 20 class files and 20 service programs to debug mode. It also
specifies certain attributes of the debugging session. For example, it can specify whether database
files in production libraries can be updated while in debug mode.

Examine query optimizer debug messages in the job log

Query optimizer debug messages issue informational messages to the job log about the implementation of
a query. These messages explain what happened during the query optimization process. For example, you
can learn:
v Why an index was or was not used
v Why a temporary result was required
v Whether joins and blocking are used
v What type of index was advised by the optimizer
v Status of the job’s queries
v Indexes used
v Status of the cursor

The optimizer automatically logs messages for all queries it optimizes, including SQL, call level interface,
ODBC, OPNQRYF, and SQL Query Manager.

60 DB2 UDB for iSeries Database Performance and Query Optimization V5R1
Viewing debug messages:

To view the messages, put your job into debug mode using one of the following methods:
v Use the following command:
STRDBG PGM(Library/program) UPDPROD(*YES)

STRDBG places in the job log information about all SQL statements that run.
v Set the QRYOPTLIB parameter on the Change Query Attributes (CHGQRYA) command to a user library
where the QAQQINI file exists. Set the parameter on the QAQQINI file to MESSAGES_DEBUG, and set
the value to *YES. This option places query optimization information in the job log.
Pressing F10 from the command Entry panel displays the message text. To see the second-level text,
press F1 (Help). The second-level text sometimes offers hints for improving query performance.

See “Query optimization performance information messages” and “Query optimization performance
information messages and open data paths” on page 66 for the specific meanings of the debug messages.

Query optimization performance information messages

You can evaluate the structure and performance of the given SQL statements in a program using
informational messages put in the job log by the database manager. The messages are issued for an SQL
program or interactive SQL when running in the debug mode. The database manager may send any of the
following messages when appropriate. The ampersand variables (&1, &X) are replacement variables that
contain either an object name or some other substitution value when the message appears in the job log.
The messages are:
v “CPI4321 - Access path built for file &1.” on page 62
v “CPI4322 - Access path built from keyed file &1” on page 63
v “CPI4323 - The OS/400 query access plan has been rebuilt” on page 63
v “CPI4324 - Temporary file built for file &1” on page 63
v “CPI4325 - Temporary result file built for query” on page 64
v “CPI4326 - File &1 processed in join position &11” on page 64
v “CPI4327 - File &1 processed in join position 1” on page 64
v “CPI4328 - Access path of file &4 was used by query” on page 64
v “CPI4329 - Arrival sequence access was used for file &1” on page 64
v “CPI432A - Query optimizer timed out for file &1” on page 64
v “CPI432B - Subselects processed as join query” on page 65
v “CPI432C - All access paths were considered for file &1” on page 65
v “CPI432D - Additional access path reason codes were used” on page 65
v “CPI432E - Selection columns mapped to different attributes” on page 65
v CPI432F Access path suggestion for file &1.
v CPI4330 &6 tasks used for parallel &10 scan of file &1.
v CPI4331 &6 tasks used for parallel index created over file &1.
v CPI4332 &1 host variables used in query.
v CPI4333 Hashing algorithm used to process join.
v CPI4334 Query implemented as reusable ODP.
v CPI4335 Optimizer debug messages for hash join step &1 follow:
v CPI4336 Group processing generated.
v CPI4337 Temporary hash table built for hash join step &1.
v “CPI4338 - &1 Access path(s) used for bitmap processing of file &2” on page 65

Chapter 4. Optimizing query performance using query optimization tools 61

v CPI4339 Query options retrieved from Library &1.
v CPI433A Unable to retrieve query options file.
v CPI433C Library &1 not found.
v CPI4341 Performing distributed query.
v CPI4342 Performing distributed join for query.
v CPI4345 Temporary distributed result file &4 built for query.
v CPI4346 Optimizer debug messages for query join step &1 of &2 follow:
v CPI4347 Query being processed in multiple steps.
v CPI4348 The ODP associated with the cursor was hard closed.
v CPI4349 Fast past refresh of the host variable values is not possible.
v CPI434A &1 Starting optimizer debug message for query &2.
v CPI434B &1 Ending debug message for query &2.
v CPI434C The OS/400 Query access plan was not rebuilt.
v “SQL7910 - All SQL cursors closed” on page 66
v “SQL7911 - ODP reused” on page 66
v “SQL7912 - ODP created” on page 66
v “SQL7913 - ODP deleted” on page 66
v “SQL7914 - ODP not deleted” on page 67
v “SQL7915 - Access plan for SQL statement has been built” on page 67
v “SQL7916 - Blocking used for query” on page 67
v “SQL7917 - Access plan not updated” on page 67
v “SQL7918 - Reusable ODP deleted” on page 67
v “SQL7919 - Data conversion required on FETCH or embedded SELECT” on page 67
v “SQL7939 - Data conversion required on INSERT or UPDATE” on page 67

These messages provide feedback on how a query was run and, in some cases, indicate the
improvements that can be made to help the query run faster.

The messages contain message help that provides information about the cause for the message, object
name references, and possible user responses.

The time at which the message is sent does not necessarily indicate when the associated function was
performed. Some messages are sent altogether at the start of a query run.

The causes and user responses for the following messages are paraphrased. The actual message help is
more complete and should be used when trying to determine the meaning and responses for each
message.

The possible user action for each message are described in the following sections:

CPI4321 - Access path built for file &1.

This message indicates that a temporary index was created to process the query. The new index is
created by reading all of the rows in the specified table.

The time required to create an index on each run of a query can be significant. Consider creating a logical
file (CRTLF) or an SQL index (CREATE INDEX SQL statement):
v Over the table named in the message help.
v With key columns named in the message help.
v With the ascending or descending sequencing specified in the message help.

62 DB2 UDB for iSeries Database Performance and Query Optimization V5R1
 v With the sort sequence table specified in the message help.

Consider creating the logical file with select or omit criteria that either match or partially match the query’s
predicates involving constants. The database manager will consider using select or omit logical files even
though they are not explicitly specified on the query.

For certain queries, the optimizer may decide to create an index even when an existing one can be used.
This might occur when a query has an ordering column as a key column for an index, and the only row
selection specified uses a different column. If the row selection results in roughly 20% of the rows or more
to be returned, then the optimizer may create a new index to get faster performance when accessing the
data. The new index minimizes the amount of data that needs to be read.

CPI4322 - Access path built from keyed file &1

This message indicates that a temporary index was created from the access path of an existing keyed
table or index.

| Generally, this action should not take a significant amount of time or resource because only a subset of
| the data in the table needs to be read. This is normally done to allow the optimizer to use an existing
| index for selection while creating one for ordering, grouping, or join criteria. Sometimes even faster
| performance can be achieved by creating a logical file or SQL index that satisfies the index requirement
| stated in the message help.

For more detail, see the previous message, CPI4321.

CPI4323 - The OS/400 query access plan has been rebuilt

This message can be sent for a variety of reasons. The specific reason is provided in the message help.

Most of the time, this message is sent when the queried table environment has changed, making the
current access plan obsolete. An example of the table environment changing is when an index required by
the query no longer exists on the system.

An access plan contains the instructions for how a query is to be run and lists the indexes for running the
query. If a needed index is no longer available, the query is again optimized, and a new access plan is
created, replacing the old one.

The process of again optimizing the query and building a new access plan at runtime is a function of DB2
UDB for iSeries. It allows a query to be run as efficiently as possible, using the most current state of the
database without user intervention.

The infrequent appearance of this message is not a cause for action. For example, this message will be
sent when an SQL package is run the first time after a restore, or anytime the optimizer detects that a
change has occurred (such as a new index was created), that warrants an implicit rebuild. However,
excessive rebuilds should be avoided because extra query processing will occur. Excessive rebuilds may
indicate a possible application design problem or inefficient database management practices. See
CPI434C.

CPI4324 - Temporary file built for file &1

| Before the query processing could begin, the data in the specified table had to be copied into a temporary
| physical table to simplify running the query. The message help contains the reason why this message was
| sent.

If the specified table selects few rows, usually less than 1000 rows, then the row selection part of the
query’s implementation should not take a significant amount of resource and time. However if the query is
taking more time and resources than can be allowed, consider changing the query so that a temporary
table is not required.

Chapter 4. Optimizing query performance using query optimization tools 63

 One way to do this is by breaking the query into multiple steps. Consider using an INSERT statement with
a subselect to select only the rows that are required into a table, and then use that table’s rows for the
rest of the query.

CPI4325 - Temporary result file built for query

| A temporary result table was created to contain the intermediate results of the query. The results are
| stored in an internal temporary table (structure). This allows for more flexibility by the optimizer in how to
| process and store the results. The message help contains the reason why a temporary result table is
| required.

In some cases, creating a temporary result table provides the fastest way to run a query. Other queries
that have many rows to be copied into the temporary result table can take a significant amount of time.
However, if the query is taking more time and resources than can be allowed, consider changing the query
so that a temporary result table is not required.

CPI4326 - File &1 processed in join position &11

This message provides the join position of the specified table when an index is used to access the table’s
data. Join position pertains to the order in which the tables are joined. See the Join optimizationsection
for details.

CPI4327 - File &1 processed in join position 1

This message provides the name of the first or primary table of the join when table access scan method is
used to select rows from the table.

See the previous message, CPI4326, for information on join position and join performance tips.

CPI4328 - Access path of file &4 was used by query

This message names an existing index that was used by the query.

The reason the index was used is given in the message help.

CPI4329 - Arrival sequence access was used for file &1

No index was used to access the data in the specified table. The rows were scanned sequentially in arrival
sequence.

| If an index does not exist, you may want to create one whose key column matches one of the columns in
| the row selection. You should only create an index if the row selection (WHERE clause) selects 20% or
| fewer rows in the table. To force the use of an existing index, change the ORDER BY clause of the query
| to specify the first key column of the index, or ensure that the query is running under a first I/O
| environment.

CPI432A - Query optimizer timed out for file &1

The optimizer stops considering indexes when the time spent optimizing the query exceeds an internal
value that corresponds to the estimated time to run the query and the number of rows in the queried
tables. Generally, the more rows in the tables, the greater the number of indexes that will be considered.

When the estimated time to run the query is exceeded, the optimizer does not consider any more indexes
and uses the current best method to implement the query. Either an index has been found to get the best
performance, or an index will have to be created. If the actual time to execute the query exceeds the
estimated run time this may indicate the optimizer did not consider the best index.

The message help contains a list of indexes that were considered before the optimizer timed out. By
viewing this list of indexes, you may be able to determine if the optimizer timed out before the best index
was considered.

64 DB2 UDB for iSeries Database Performance and Query Optimization V5R1
 To ensure that an index is considered for optimization, specify the logical file associated with the index as
the table to be queried. The optimizer will consider the index of the table specified on the query or SQL
statement first. Remember that SQL indexes cannot be queried.

You may want to delete any indexes that are no longer needed.

CPI432B - Subselects processed as join query

Two or more SQL subselects were combined by the query optimizer and processed as a join query.
Generally, this method of processing is a good performing option.

CPI432C - All access paths were considered for file &1

The optimizer considered all indexes built over the specified table. Since the optimizer examined all
indexes for the table, it determined the current best access to the table.

The message help contains a list of the indexes. With each index a reason code is added. The reason
code explains why the index was or was not used.

CPI432D - Additional access path reason codes were used

Message CPI432A or CPI432C was issued immediately before this message. Because of message length
restrictions, some of the reason codes used by messages CPI432A and CPI432C are explained in the
message help of CPI432D. Use the message help from this message to interpret the information returned
from message CPI432A or CPI432C.

CPI432E - Selection columns mapped to different attributes

This message indicates that the query optimizer was not able to consider the usage of an index to resolve
one or more of the selection specifications of the query. If there was an index available which otherwise
could have been used to limit the processing of the query to just a few rows, then the performance of this
query will be affected.

The attributes of a comparison value and a comparison column must match otherwise a conversion will
occur so that they do match. Generally, this conversion occurs such that the value with the smallest
attributes is mapped to the attributes of the other value. When the attributes of the comparison column
have to be mapped to be compatible with that of the comparison value, the optimizer can no longer use an
index to implement this selection.

CPI4338 - &1 Access path(s) used for bitmap processing of file &2
The optimizer chooses to use one or more indexes, in conjunction with the query selection (WHERE
clause), to build a bitmap. This resulting bitmap indicates which rows will actually be selected.

Conceptually, the bitmap contains one bit per row in the underlying table. Corresponding bits for selected
rows are set to ’1’. All other bits are set to ’0’.

Once the bitmap is built, it is used, as appropriate, to avoid mapping in rows from the table not selected by
the query. The use of the bitmap depends on whether the bitmap is used in combination with the arrival
sequence or with a primary index.

When bitmap processing is used with arrival sequence, either message CPI4327 or CPI4329 will precede
this message. In this case, the bitmap will help to selectively map only those rows from the table that the
query selected.

| When bitmap processing is used with a primary index, either message CPI4326 or CPI4328 will precede
| this message. Rows selected by the primary index will be checked against the bitmap before mapping the
| row from the table. See the Bitmap processing access method for details.

Chapter 4. Optimizing query performance using query optimization tools 65

Query optimization performance information messages and open data
paths
Several of the following SQL run-time messages refer to open data paths.

An open data path (ODP) definition is an internal object that is created when a cursor is opened or when
other SQL statements are run. It provides a direct link to the data so that I/O operations can occur. ODPs
are used on OPEN, INSERT, UPDATE, DELETE, and SELECT INTO statements to perform their
respective operations on the data.

Even though SQL cursors are closed and SQL statements have already been run, the database manager
in many cases will save the associated ODPs of the SQL operations to reuse them the next time the
statement is run. So an SQL CLOSE statement may close the SQL cursor but leave the ODP available to
be used again the next time the cursor is opened. This can significantly reduce the processing and
response time in running SQL statements.

The ability to reuse ODPs when SQL statements are run repeatedly is an important consideration in
achieving faster performance.

The following informational messages are issued at SQL run time:

SQL7910 - All SQL cursors closed

This message is sent when the job’s call stack no longer contains a program that has run an SQL
statement.

Unless CLOSQLCSR(*ENDJOB) or CLOSQLCSR(*ENDACTGRP) was specified, the SQL environment for

reusing ODPs across program calls exists only until the active programs that ran the SQL statements
complete.

Except for ODPs associated with *ENDJOB or *ENDACTGRP cursors, all ODPs are deleted when all the
SQL programs on the call stack complete and the SQL environment is exited.

This completion process includes closing of cursors, the deletion of ODPs, the removal of prepared
statements, and the release of locks.

Putting an SQL statement that can be run in the first program of an application keeps the SQL
environment active for the duration of that application. This allows ODPs in other SQL programs to be
reused when the programs are repeatedly called. CLOSQLCSR(*ENDJOB) or
CLOSQLCSR(*ENDACTGRP) can also be specified.

SQL7911 - ODP reused

This message indicates that the last time the statement was run or when a CLOSE statement was run for
this cursor, the ODP was not deleted. It will now be used again. This should be an indication of very
efficient use of resources by eliminating unnecessary OPEN and CLOSE operations.

SQL7912 - ODP created

No ODP was found that could be used again. The first time that the statement is run or the cursor is
opened for a process, an ODP will always have to be created. However, if this message appears on every
run of the statement or open of the cursor, the tips recommended in “Database application design tips:
Retaining cursor positions for non-ILE program calls” on page 116 should be applied to this application.

SQL7913 - ODP deleted

For a program that is run only once per job, this message could be normal. However, if this message
appears on every run of the statement or open of the cursor, then the tips recommended in “Database
application design tips: Retaining cursor positions for non-ILE program calls” on page 116 should be
applied to this application.

66 DB2 UDB for iSeries Database Performance and Query Optimization V5R1
SQL7914 - ODP not deleted
If the statement is rerun or the cursor is opened again, the ODP should be available again for use.

SQL7915 - Access plan for SQL statement has been built

The DB2 UDB for iSeries precompilers allow the creation of the program objects even when required
tables are missing. In this case the binding of the access plan is done when the program is first run. This
message indicates that an access plan was created and successfully stored in the program object.

SQL7916 - Blocking used for query

SQL will request multiple rows from the database manager when running this statement instead of
requesting one row at a time.

SQL7917 - Access plan not updated

The database manager rebuilt the access plan for this statement, but the program could not be updated
with the new access plan. Another job is currently running the program that has a shared lock on the
access plan of the program.

The program cannot be updated with the new access plan until the job can obtain an exclusive lock on the
access plan of the program. The exclusive lock cannot be obtained until the shared lock is released.

The statement will still run and the new access plan will be used; however, the access plan will continue to
be rebuilt when the statement is run until the program is updated.

SQL7918 - Reusable ODP deleted

A reusable ODP exists for this statement, but either the job’s library list or override specifications have
changed the query.

The statement now refers to different tables or uses different override specifications than are in the
existing ODP. The existing ODP cannot be reused, and a new ODP must be created. To make it possible
to reuse the ODP, avoid changing the library list or the override specifications.

SQL7919 - Data conversion required on FETCH or embedded SELECT

When mapping data to host variables, data conversions were required. When these statements are run in
the future, they will be slower than if no data conversions were required. The statement ran successfully,
but performance could be improved by eliminating the data conversion. For example, a data conversion
that would cause this message to occur would be the mapping of a character string of a certain length to a
host variable character string with a different length. You could also cause this error by mapping a numeric
value to a host variable that is a different type (decimal to integer). To prevent most conversions, use host
variables that are of identical type and length as the columns that are being fetched.

SQL7939 - Data conversion required on INSERT or UPDATE

The attributes of the INSERT or UPDATE values are different than the attributes of the columns receiving
the values. Since the values must be converted, they cannot be directly moved into the columns.
Performance could be improved if the attributes of the INSERT or UPDATE values matched the attributes
of the columns receiving the values.

Gather information about embedded SQL statements with the

PRTSQLINF command
The PRTSQLINF command gathers information about the embedded SQL statements in a program, SQL
package (the object normally used to store the access plan for a remote query), or service program. It
then puts the information in a spooled file. PRTSQLINF provides information about:
v The SQL statements being executed
v The type of access plan used during execution. This includes information about how the query will be
implemented, the indexes used, the join order, whether a sort is done, whether a database scan is
sued, and whether an index is created.

Chapter 4. Optimizing query performance using query optimization tools 67

 v A list of the command parameters used to precompile the source member for the object.
To gather this information, run PRTSQLINF against a saved access plan. This means you must execute or
at least prepare the query (using SQL’s PREPARE statement) before you use the command. It is best to
execute the query because the index created as a result of PREPARE is relatively sparse and may well
change after the first run. PRTSQLINF’s requirement of a saved access plan means the command cannot
be used with OPNQRYF.

PRTSQLINF gives output that is similar to the information you can get from debug messages, but
PRTSQLINF must be run against a saved access plan. The query optimizer automatically logs information
messages about the current query processing when your job is in debug mode. So, query debug
messages work at runtime while PRTSQLINF works retroactively. You can also see this information in the
second level text of the query governor inquiry message CPA4259. The messages are:
| v SQL400A Temporary distributed result file &1 was created to contain join result. Result file was directed
| v SQL400B Temporary distributed result file &1 was created to contain join result. Result file was
| broadcast.
| v SQL400C Optimizer debug messages for distributed query step &1 and &2 follow.
| v SQL400D GROUP BY processing generated.
| v SQL400E Temporary distributed result file &1 was created while processing distributed subquery.
| v SQL4001 Temporary result created.
| v SQL4002 Reusable ODP sort used.
| v SQL4003 UNION.
| v SQL4004 SUBQUERY.
| v SQL4005 Query optimizer timed out for table &1.
| v SQL4006 All indexes considered for table &1.
| v SQL4007 Query implementation for join position &1 table &2.
| v SQL4008 Index &1 used for table &2.
| v SQL4009 Index created for table &1.
| v SQL401A Processing grouping criteria for query containing a distributed table.
| v SQL401B Temporary distributed result table &1 was created while processing grouping criteria.
| v SQL401C Performing distributed join for query.
| v SQL401D Temporary distributed result table &1 was created because table &2 was directed.
| v SQL401E Temporary distributed result table &1 was created because table &2 was broadcast.
| v SQL401F Table &1 used in distributed join.
| v SQL4010 Table scan access for table &1.
| v SQL4011 Index scan-key row positioning used on table &1.
| v SQL4012 Index created from index &1 for table &2.
| v SQL4013 Access plan has not been built.
| v SQL4014 &1 join column pair(s) are used for this join position.
| v SQL4015 From-column &1.&2, to-column &3.&4, join operator &%, join predicate &6.
| v SQL4016 Subselects processed as join query.
| v SQL4017 Host variables implemented as reusable ODP.
| v SQL4018 Host variables implemented as non-reusable ODP.
| v SQL4019 Host variables implemented as file management row positioning reusable ODP.
| v SQL402A Hashing algorithm used to process join.
| v SQL402B Table &1 used in hash join step &2.
| v SQL402C Temporary table created for hash join results.
| v SQL402D Query attributes overridden from query options file &2 in library &1.

68 DB2 UDB for iSeries Database Performance and Query Optimization V5R1
| v SQL4020 Estimated query run time is &1 seconds.
| v SQL4021 Access plan last saved on &1 at &2.
| v SQL4022 Access plan was saved with SRVQRY attributes active.
| v SQL4023 Parallel table prefetch used.
| v SQL4024 Parallel index preload access method used.
| v SQL4025 Parallel table preload access method used.
| v SQL4026 Index only access used on table number &1.
| v SQL4027 Access plan was saved with DB2 UDB Symmetric Multiprocessing installed on the system.
| v SQL4028 The query contains a distributed table.
| v SQL4029 Hashing algorithm used to process the grouping.
| v SQL4030 &1 tasks specified for parallel scan on table &2.
| v SQL4031 &1 tasks specified for parallel index create over table &2.
| v SQL4032 Index &1 used for bitmap processing of table &2.
| v SQL4033 &1 tasks specified for parallel bitmap create using &2.
| v SQL4034 Multiple join classes used to process join.
| v SQL4035 Table &1 used in join class &2.

Gather statistics about your queries with the database monitor

Database monitor statistics provide the most complete information about a query. You can gather
performance statistics for a specific query or for every query on the system. There are several different
ways to gather the statistics:
v Use the “Start Database Monitor (STRDBMON) command” on page 70 and the “End Database Monitor
(ENDDBMON) command” on page 71.
v Use the Start Performance Monitor (STRPFRMON) command with the STRDBMON parameter.
v Use memory resident database monitor APIs. See “Gather statistics about your queries with
memory-resident database monitor APIs” on page 78.

| Note: Database monitors can generate significant CPU and disk storage overhead when in use.

You can monitor a specific job or all jobs on the system. The statistics gathered are placed in the output
database table specified on the command. Each job in the system can be monitored concurrently by two
monitors:
v One started specifically on that job
v One started for all jobs in the system
When a job is monitored by two monitors, each monitor is logging rows to a different output table. You can
identify rows in the output database table by each row’s unique identification number.

What kinds of statistics you can gather

The database monitor provides the same information that is provided with the query optimizer debug
messages (STRDBG) and the Print SQL information (PRTSQLINF) command. The following is a sampling
of the additional information that will be gathered by the database monitors:
v System and job name
v SQL statement and sub-select number
v Start and end timestamp
v Estimated processing time
v Total rows in table queried
v Number of rows selected

Chapter 4. Optimizing query performance using query optimization tools 69

v Estimated number of rows selected
v Estimated number of joined rows
v Key columns for advised index
v Total optimization time
v Join type and method
v ODP implementation

How you can use performance statistics

You can use these performance statistics to generate various reports. For instance, you can include
reports that show queries that:
v Use an abundance of the system resources.
v Take an extremely long time to execute.
v Did not run because of the query governor time limit.
v Create a temporary index during execution
v Use the query sort during execution
v Could perform faster with the creation of a keyed logical file containing keys suggested by the query
optimizer.

Note: A query that is cancelled by an end request generally does not generate a full set of performance
statistics. However, it does contain all the information about how a query was optimized, with the
exception of runtime or multi-step query information.

Start Database Monitor (STRDBMON) command

The STRDBMON command starts the collection of database performance statistics for a specific job or all
jobs on the system. The statistics are placed in an output database table and member specified on the
command. If the output table and/or member does not exist, one is created based upon the table and
format definition of model table QSYS/QAQQDBMN. If the output table and/or member exist, the row
format of the output table must be named QQQDBMN.

You can specify a replace/append option that allows you to clear the member of information before writing
rows or to just append new information to the end of the existing table.

You can also specify a force row write option that allows you to control how many rows are kept in the row
buffer of each job being monitored before forcing the rows to be written to the output table. By specifying a
force row write value of 1, FRCRCD(1), monitor rows will appear in the log as soon as they are created.
FRCRCD(1) also ensures that the physical sequence of the rows are most likely, but not guaranteed, to be
in time sequence. However, FRCRCD(1) will cause the most negative performance impact on the jobs
being monitored. By specifying a larger number for the FRCRCD parameter, the performance impact of
monitoring can be lessened.

Specifying *DETAIL on the TYPE parameter of the STRDBMON command indicates that detail rows, as
well as summary rows, are to be collected. This is only useful for non-SQL queries, those queries which
do not generate a QQQ1000 row. For non-SQL queries the only way to determine the number of rows
returned and the total time to return those rows is to collect detail rows. Currently the only detail row is
QQQ3019, in Appendix A, “Database monitor: DDS” on page 129. While the detail row contains valuable
information, it creates a slight performance degradation for each block of rows returned. Therefore its use
should be closely monitored.

If the monitor is started on all jobs, any jobs waiting on job queues or any jobs started during the
monitoring period will have statistics gathered from them once they begin. If the monitor is started on a
specific job, that job must be active in the system when the command is issued. Each job in the system
can be monitored concurrently by only two monitors:

70 DB2 UDB for iSeries Database Performance and Query Optimization V5R1
 v One started specifically on that job.
v One started on all jobs in the system.
When a job is monitored by two monitors and each monitor is logging to a different output table, monitor
rows will be written to both logs for this job. If both monitors have selected the same output table then the
monitor rows are not duplicated in the output table.

End Database Monitor (ENDDBMON) command

The ENDDBMON command ends the Database Monitor for a specific job or all jobs on the system. If an
attempt to end the monitor on all jobs is issued, there must have been a previous STRDBMON issued for
all jobs. If a particular job is specified on this command, the job must have the monitor started explicitly
and specifically on that job.

For example, consider the following sequence of events:

1. Monitoring was started for all jobs in the system.
2. Monitoring was started for a specific job.
3. Monitoring was ended for all jobs.
In this sequence, the specific job monitor continues to run because an explicit start of the monitor was
done on it. It continues to run until an ENDDBMON on the specific job is issued.

Consider the following sequence:

1. Monitoring was started for all jobs in the system.
2. Monitoring was started for a specific job.
3. Monitoring was ended for the specific job.
In this sequence, monitoring continues to run for all jobs, even over the specific job, until an ENDDBMON
for all jobs is issued.

In the following sequence:

1. Monitoring was started for a specific job.
2. Monitoring was started for all jobs in the system.
3. Monitoring was ended for all jobs.
In this sequence, monitoring continues to run for the specific job until you issue an ENDDBMON for that
job.

In the following sequence:

1. Monitoring was started for a specific job.
2. Monitoring was started for all jobs in the system.
3. Monitoring was ended for the specific job.
In this sequence, monitoring continues to run for all jobs, including the specific job.

When monitoring is ended for all jobs, all of the jobs on the system will be triggered to close the output
table, however, the ENDDBMON command can complete before all of the monitored jobs have written
their final performance rows to the log. Use the Work with Object Locks (WRKOBJLCK) command to see
that all of the monitored jobs no longer hold locks on the output table before assuming the monitoring is
complete.

Database monitor performance rows

| The rows in the database table are uniquely identified by their row identification number. The information
| within the file-based monitor (STRDBMON) is written out based upon a set of logical formats which are
| defined in Appendix A. These logical formats correlate closely to the debug messages and the PRSQLINF

Chapter 4. Optimizing query performance using query optimization tools 71

| messages. The appendix also identifies which physical columns are used for each logical format and what
| information it contains. You can use the logical formats to identify the information that can be extracted
| from the monitor. These rows are defined in several different logical files which are not shipped with the
| system and must be created by the user, if desired. The logical files can be created with the DDS shown
| in “Optional database monitor logical file DDS” on page 136. The column descriptions are explained in the
| tables following each figure.

| Note: The database monitor logical files are keyed logical files that contain some select/omit criteria.
| Therefore, there will be some maintenance overhead associated with these tables while the
| database monitor is active. The user may want to minimize this overhead while the database
| monitor is active, especially if monitoring all jobs. When monitoring all jobs, the number of rows
| generated could be quite large. The logicals are not required to process the results. They simply
| make the extraction of information for the table easier and more direct.

| Minimizing maintenance overhead

| Possible ways to minimize maintenance overhead associated with database monitor logical files:
| v Do not create the database monitor logical files until the database monitor has completed.
| v Create the database monitor logical files using dynamic select/omit criteria (DYNSLT keyword on logical
| file’s DDS).
| v Create the database monitor logical files with rebuild index maintenance specified on the CRTLF
| command (*REBLD option on MAINT parameter).
| By minimizing the maintenance overhead at run time, you are merely delaying the maintenance cost until
| the database monitor logical file is either created or opened. The choice is to either spend the time while
| the database monitor is active or spend the time after the database monitor has completed.

| Query optimizer index advisor

| The query optimizer analyzes the row selection in the query and determines, based on default values, if
| creation of a permanent index would improve performance. If the optimizer determines that a permanent
| index would be beneficial, it returns the key columns necessary to create the suggested index.

| The index advisor information can be found in the Database Monitor logical files QQQ3000, QQQ3001 and
| QQQ3002. The advisor information is stored in columns QQIDXA, QQIDXK and QQIDXD. When the
| QQIDXA column contains a value of ’Y’ the optimizer is advising you to create an index using the key
| columns shown in column QQIDXD. The intention of creating this index is to improve the performance of
| the query.

| In the list of key columns contained in column QQIDXD the optimizer has listed what it considers the
| suggested primary and secondary key columns. Primary key columns are columns that should significantly
| reduce the number of keys selected based on the corresponding query selection. Secondary key columns
| are columns that may or may not significantly reduce the number of keys selected.

| The optimizer is able to perform index scan-key positioning over any combination of the primary key
| columns, plus one additional secondary key column. Therefore it is important that the first secondary key
| column be the most selective secondary key column. The optimizer will use index scan-key selection with
| any of the remaining secondary key columns. While index scan-key selection is not as fast as index
| scan-key positioning it can still reduce the number of keys selected. Hence, secondary key columns that
| are fairly selective should be included.

| Column QQIDXK contains the number of suggested primary key columns that are listed in column
| QQIDXD. These are the left-most suggested key columns. The remaining key columns are considered
| secondary key columns and are listed in order of expected selectivity based on the query. For example,

72 DB2 UDB for iSeries Database Performance and Query Optimization V5R1
| assuming QQIDXK contains the value of 4 and QQIDXD specifies 7 key columns, then the first 4 key
| columns specified in QQIDXK would be the primary key columns. The remaining 3 key columns would be
| the suggested secondary key columns.

| It is up to the user to determine the true selectivity of any secondary key columns and to determine
| whether those key columns should be included when creating the index. When building the index the
| primary key columns should be the left-most key columns followed by any of the secondary key columns
| the user chooses and they should be prioritized by selectivity. The query optimizer index advisor should
| only be used to help analyze complex selection within a query that cannot be easily debugged manually.

| Note: After creating the suggested index and executing the query again, it is possible that the query
| optimizer will choose not to use the suggested index. While the selection criteria is taken into
| consideration by the query optimizer, join, ordering, and grouping criteria are not.

| Database monitor examples

| Suppose you have an application program with SQL statements and you want to analyze and performance
| tune these queries. The first step in analyzing the performance is collection of data. The following
| examples show how you might collect and analyze data using STRDBMON and ENDDBMON.

| Performance data is collected in LIB/PERFDATA for an application running in your current job. The
| following sequence collects performance data and prepares to analyze it.
| 1. STRDBMON FILE(LIB/PERFDATA). If this table does not already exist, the command will create one
| from the skeleton table in QSYS/QAQQDBMN.
| 2. Run your application
| 3. ENDDBMON
| 4. Create logical files over LIB/PERFDATA using the DDS shown in “Optional database monitor logical file
| DDS” on page 136. Creating the logical files is not mandatory. All of the information resides in the base
| table that was specified on the STRDBMON command. The logical files simply provide an easier way
| to view the data.
| You are now ready to analyze the data. The following examples give you a few ideas on how to use this
| data. You should closely study the physical and logical file DDS to understand all the data being collected
| so you can create queries that give the best information for your applications.

| Database monitor performance analysis example 1

| Determine which queries in your SQL application are implemented with table scans. The complete
| information can be obtained by joining two logical files: QQQ1000, which contains information about the
| SQL statements, and QQQ3000, which contains data about queries performing table scans. The following
| SQL query could be used:
| SELECT A.QQTLN, A.QQTFN, A.QQTOTR, A.QQIDXA, C.QQrcdr,
| (B.QQETIM - B.QQSTIM) AS TOT_TIME, B.QQSTTX
| FROM LIB/QQQ3000 A, LIB/QQQ1000 B, LIB/QQQ3019 C
| WHERE A.QQJFLD = B.QQJFLD
| AND A.QQUCNT = B.QQUCNT
| AND A.QQJFLD = C.QQJFLD AND A.QQUCNT = C.QQUCNT

| Sample output of this query is shown in Table 3 on page 74. Key to this example are the join criteria:
WHERE A.QQJFLD = B.QQJFLD
AND A.QQUCNT = B.QQUCNT

A lot of data about many queries is contained in multiple rows in table LIB/PERFDATA. It is not uncommon
for data about a single query to be contained in 10 or more rows within the table. The combination of
defining the logical files and then joining the tables together allows you to piece together all the data for a
query or set of queries. Column QQJFLD uniquely identifies all data common to a job; column QQUCNT is
unique at the query level. The combination of the two, when referenced in the context of the logical files,
connects the query implementation to the query statement information.

Chapter 4. Optimizing query performance using query optimization tools 73

Table 3. Output for SQL Queries that Performed Table Scans
Table Total Index Rows
Lib Name Name Rows Advised Returned TOT_TIME Statement Text
LIB1 TBL1 20000 Y 10 6.2 SELECT * FROM LIB1/TBL1
WHERE FLD1 = ’A’
LIB1 TBL2 100 N 100 0.9 SELECT * FROM LIB1/TBL2
LIB1 TBL1 20000 Y 32 7.1 SELECT * FROM LIB1/TBL1
WHERE FLD1 = ’B’ AND
FLD2 > 9000

If the query does not use SQL, the SQL information row (QQQ1000) is not created. This makes it more
difficult to determine which rows in LIB/PERFDATA pertain to which query. When using SQL, row
QQQ1000 contains the actual SQL statement text that matches the performance rows to the
corresponding query. Only through SQL is the statement text captured. For queries executed using the
OPNQRYF command, the OPNID parameter is captured and can be used to tie the rows to the query. The
OPNID is contained in column QQOPID of row QQQ3014.

Database monitor performance analysis example 2

Similar to the preceding example that showed which SQL applications were implemented with table scans,
the following example shows all queries that are implemented with table scans.
SELECT A.QQTLN, A.QQTFN, A.QQTOTR, A.QQIDXA,
B.QQOPID, B.QQTTIM, C.QQCLKT, C.QQRCDR, D.QQROWR,
(D.QQETIM - D.QQSTIM) AS TOT_TIME, D.QQSTTX
FROM LIB/QQQ3000 A INNER JOIN LIB/QQQ3014 B
ON (A.QQJFLD = B.QQJFLD AND
A.QQUCNT = B.QQUCNT)
LEFT OUTER JOIN LIB/QQQ3019 C
ON (A.QQJFLD = C.QQJFLD AND
A.QQUCNT = C.QQUCNT)
LEFT OUTER JOIN LIB/QQQ1000 D
ON (A.QQJFLD = D.QQJFLD AND
A.QQUCNT = D.QQUCNT)

In this example, the output for all queries that performed table scans are shown in Table 4.

Note: The columns selected from table QQQ1000 do return NULL default values if the query was not
executed using SQL. For this example assume the default value for character data is blanks and
the default value for numeric data is an asterisk (*).
Table 4. Output for All Queries that Performed Table Scans
ODP
Lib Table Total Index Query Open Clock Recs Rows TOT_
Name Name Rows Advised OPNID Time Time Rtned Rtned TIME Statement Text
LIB1 TBL1 20000 Y 1.1 4.7 10 10 6.2 SELECT *
FROM LIB1/TBL1
WHERE FLD1 = ’A’
LIB1 TBL2 100 N 0.1 0.7 100 100 0.9 SELECT *
FROM LIB1/TBL2
LIB1 TBL1 20000 Y 2.6 4.4 32 32 7.1 SELECT *
FROM LIB1/TBL1
WHERE FLD1 = ’A’
AND FLD2 > 9000
LIB1 TBL4 4000 N QRY04 1.2 4.2 724 * * *

74 DB2 UDB for iSeries Database Performance and Query Optimization V5R1
If the SQL statement text is not needed, joining to table QQQ1000 is not necessary. You can determine
the total time and rows selected from data in the QQQ3014 and QQQ3019 rows.

Database monitor performance analysis example 3

Your next step may include further analysis of the table scan data. The previous examples contained a
column titled Index Advised. A Y (yes) in this column is a hint from the query optimizer that the query may
perform better with an index to access the data. For the queries where an index is advised, notice that the
rows selected by the query are low in comparison to the total number of rows in the table. This is another
indication that a table scan may not be optimal. Finally, a long execution time may highlight queries that
may be improved by performance tuning.

The next logical step is to look into the index advised optimizer hint. The following query could be used for
this:
SELECT A.QQTLN, A.QQTFN, A.QQIDXA, A.QQIDXD,
A.QQIDXK, B.QQOPID, C.QQSTTX
FROM LIB/QQQ3000 A INNER JOIN LIB/QQQ3014 B
ON (A.QQJFLD = B.QQJFLD AND
A.QQUCNT = B.QQUCNT)
LEFT OUTER JOIN LIB/QQQ1000 C
ON (A.QQJFLD = C.QQJFLD AND
A.QQUCNT = C.QQUCNT)
WHERE A.QQIDXA = ’Y’

There are two slight modifications from the first example. First, the selected columns have been changed.
Most important is the selection of column QQIDXD that contains a list of possible key columns to use
when creating the index suggested by the query optimizer. Second, the query selection limits the output to
those table scan queries where the optimizer advises that an index be created (A.QQIDXA = ’Y’). Table 5
shows what the results might look like.
Table 5. Output with Recommended Key Columns
Advised Advised
Table Index Key Primary Query
Lib Name Name Advised columns Key OPNID Statement Text
LIB1 TBL1 Y FLD1 1 SELECT * FROM LIB1/TBL1
WHERE FLD1 = ’A’
LIB1 TBL1 Y FLD1, 1 SELECT * FROM LIB1/TBL1
FLD2 WHERE FLD1 = ’B’ AND
FLD2 > 9000
LIB1 TBL4 Y FLD1, 1 QRY04
FLD4

At this point you should determine whether it makes sense to create a permanent index as advised by the
optimizer. In this example, creating one index over LIB1/TBL1 would satisfy all three queries since each
use a primary or left-most key column of FLD1. By creating one index over LIB1/TBL1 with key columns
FLD1, FLD2, there is potential to improve the performance of the second query even more. The frequency
these queries are run and the overhead of maintaining an additional index over the table should be
considered when deciding whether or not to create the suggested index.

If you create a permanent index over FLD1, FLD2 the next sequence of steps would be to:
1. Start the performance monitor again
2. Re-run the application
3. End the performance monitor
4. Re-evaluate the data.

It is likely that the three index-advised queries are no longer performing table scans.

Chapter 4. Optimizing query performance using query optimization tools 75

Additional database monitor examples
The following are additional ideas or examples on how to extract information from the performance monitor
statistics. All of the examples assume data has been collected in LIB/PERFDATA and the documented
logical files have been created.
1. How many queries are performing dynamic replans?
SELECT COUNT(*)
FROM LIB/QQQ1000
WHERE QQDYNR <> ’NA’
2. What is the statement text and the reason for the dynamic replans?
SELECT QQDYNR, QQSTTX
FROM LIB/QQQ1000
WHERE QQDYNR <> ’NA’

Note: You have to refer to the description of column QQDYNR for definitions of the dynamic replan
reason codes.
3. How many indexes have been created over LIB1/TBL1?
SELECT COUNT(*)
FROM LIB/QQQ3002
WHERE QQTLN = ’LIB1’
AND QQTFN = ’TBL1’
4. What key columns are used for all indexes created over LIB1/TBL1 and what is the associated SQL
statement text?
SELECT A.QQTLN, A.QQTFN, A.QQIDXD, B.QQSTTX
FROM LIB/QQQ3002 A, LIB/QQQ1000 B
WHERE A.QQJFLD = B.QQJFLD
AND A.QQUCNT = B.QQUCNT
AND A.QQTLN = ’LIB1’
AND A.QQTFN = ’TBL1’

Note: This query shows key columns only from queries executed using SQL.
5. What key columns are used for all indexes created over LIB1/TBL1 and what was the associated
SQL statement text or query open ID?
SELECT A.QQTLN, A.QQTFN, A.QQIDXD,
B.QQOPID,C.QQSTTX
FROM LIB/QQQ3002 A INNER JOIN LIB/QQQ3014 B
ON (A.QQJFLD = B.QQJFLD AND
A.QQUCNT = B.QQUCNT)
LEFT OUTER JOIN LIB/QQQ1000 C
ON (A.QQJFLD = C.QQJFLD AND
A.QQUCNT = C.QQUCNT)
WHERE A.QQTLN = ’LIB1’
AND A.QQTFN = ’TBL1’

Note: This query shows key columns from all queries on the system.
6. What types of SQL statements are being performed? Which are performed most frequently?
SELECT QQSTOP, COUNT(*)
FROM LIB/QQQ1000
GROUP BY QQSTOP
ORDER BY 2 DESC
7. Which SQL queries are the most time consuming? Which user is running these queries?
SELECT (QQETIM - QQSTIM), QQUSER, QQSTTX
FROM LIB/QQQ1000
ORDER BY 1 DESC
8. Which queries are the most time consuming?
SELECT (A.QQTTIM + B.QQCLKT), A.QQOPID, C.QQSTTX
FROM LIB/QQQ3014 A LEFT OUTER JOIN LIB/QQQ3019 B
ON (A.QQJFLD = B.QQJFLD AND

76 DB2 UDB for iSeries Database Performance and Query Optimization V5R1
 A.QQUCNT = B.QQUCNT)
LEFT OUTER JOIN LIB/QQQ1000 C
ON (A.QQJFLD = C.QQJFLD AND
A.QQUCNT = C.QQUCNT)
ORDER BY 1 DESC

Note: This example assumes detail data has been collected into row QQQ3019.
9. Show the data for all SQL queries with the data for each SQL query logically grouped together.
SELECT A.*
FROM LIB/PERFDATA A, LIB/QQQ1000 B
WHERE A.QQJFLD = B.QQJFLD
AND A.QQUCNT = B.QQUCNT

Note: This might be used within a report that will format the interesting data into a more readable
format. For example, all reason code columns could be expanded by the report to print the
definition of the reason code (that is, physical column QQRCOD = ’T1’ means a table scan
was performed because no indexes exist over the queried table).
10. How many queries are being implemented with temporary tables because a key length of greater
than 2000 bytes or more than 120 key columns was specified for ordering?
SELECT COUNT(*)
FROM LIB/QQQ3004
WHERE QQRCOD = ’F6’
11. Which SQL queries were implemented with nonreusable ODPs?
SELECT B.QQSTTX
FROM LIB/QQQ3010 A, LIB/QQQ1000 B
WHERE A.QQJFLD = B.QQJFLD
AND A.QQUCNT = B.QQUCNT
AND A.QQODPI = ’N’
12. What is the estimated time for all queries stopped by the query governor?
SELECT QQEPT, QQOPID
FROM LIB/QQQ3014
WHERE QQGVNS = ’Y’

Note: This example assumes detail data has been collected into row QQQ3019.
13. Which queries estimated time exceeds actual time?
SELECT A.QQEPT, (A.QQTTIM + B.QQCLKT), A.QQOPID,
C.QQTTIM, C.QQSTTX
FROM LIB/QQQ3014 A LEFT OUTER JOIN LIB/QQQ3019 B
ON (A.QQJFLD = B.QQJFLD AND
A.QQUCNT = B.QQUCNT)
LEFT OUTER JOIN LIB/QQQ1000 C
ON (A.QQJFLD = C.QQJFLD AND
A.QQUCNT = C.QQUCNT)
WHERE A.QQEPT/1000 > (A.QQTTIM + B.QQCLKT)

Note: This example assumes detail data has been collected into row QQQ3019.
14. Should a PTF for queries that perform UNION exists be applied. It should be applied if any queries
are performing UNION. Do any of the queries perform this function?
SELECT COUNT(*)
FROM QQQ3014
WHERE QQUNIN = ’Y’

Note: If result is greater than 0, the PTF should be applied.

15. You are a system administrator and an upgrade to the next release is planned. A comparison
between the two releases would be interesting.
v Collect data from your application on the current release and save this data in LIB/CUR_DATA
v Move to the next release

Chapter 4. Optimizing query performance using query optimization tools 77

 v Collect data from your application on the new release and save this data in a different table:
LIB/NEW_DATA
v Write a program to compare the results. You will need to compare the statement text between the
rows in the two tables to correlate the data.

Gather statistics about your queries with memory-resident database

monitor APIs
The Memory-Resident Database Monitor (DBMon) is a tool that provides another method for monitoring
database performance. This tool is only intended for SQL performance monitoring and is useful for
programmers and performance analysts. The DBMon monitor, with the help of a new set of APIs, takes
database monitoring statistics and manages them for the user in memory. This memory-based monitor
reduces CPU overhead as well as resulting table sizes.

The Start Database Monitor (STRDBMON) can constrain system resources when collecting performance
information. This overhead is mainly attributed to the fact that performance information is written directly to
a database table as the information is collected. The memory-based collection mode reduces the system
resources consumed by collecting and managing performance results in memory. This allows the monitor
to gather database performance statistics with a minimal impact to the performance of the system as
whole (or to the performance of individual SQL statements).

The DBMon monitor collects much of the same information as the STRDBMON monitor, but the
performance statistics are kept in memory. At the expense of some detail, information is summarized for
identical SQL statements to reduce the amount of information collected. The objective is to get the
statistics to memory as fast as possible while deferring any manipulation or conversion of the data until the
performance data is dumped to a result table for analysis.

The DBMon monitor is not meant to replace the STRDBMON monitor. There are circumstances where the
loss of detail in the DBMon monitor will not be sufficient to fully analyze an SQL statement. In these cases,
the STRDBMON monitor should still be used.

The DBMon monitor manages the data in memory, combining and accumulating the information into a
series of row formats. This means that for each unique SQL statement, information is accumulated from
each run of the statement and the detail information is only collected for the most expensive statement
execution.

Each SQL statement is identified by the monitor according to the:

v statement name
v package (or program)
v library that contains the prepared statement
v cursor name that is used
For pure dynamic statements, the statement text is kept in a separate space and the statement
identification will be handled internally via a pointer.

While this system avoids the significant overhead of writing each SQL operation to a table, keeping
statistics in memory comes at the expense of some detail. Your objective should be to get the statistics to
memory as fast as possible, then reserve time for data manipulation or data conversion later when you
dump data to a table.

The DBMon manages the data that is in memory by combining and accumulating the information into the
new row formats. Therefore, for each unique SQL statement, information accumulates from each running
of the statement, and the system only collects detail information for the most expensive statement
execution.

78 DB2 UDB for iSeries Database Performance and Query Optimization V5R1
 Each SQL statement is identified by the monitor by the statement name, the package (or program) and
library that contains the prepared statement and the cursor name that is used. For pure dynamic
statements:
v Statement text is kept in a separate space and
v Statement identification is handled internally via a pointer.

API support for the DBMon monitor

A new set of APIs enable support for the DBMon monitor. An API supports each of the following activities:
v Start the new monitor
v Dump statistics to tables
v Clear the monitor data from memory
v Query the monitor status
v End the new monitor
When you start the new monitor, information is stored in the local address space of each job that the
system monitors. As each statement completes, the system moves information from the local job space to
a common system space. If more statements are executed than can fit in this amount of common system
space, the system drops the statements that have not been executed recently.

Memory-resident database monitor external API description

The memory-resident database monitor is controlled by a set of APIs. For additional information, see the
OS/400 APIs information in the Programming category of the iSeries Information Center.
Table 6. External API Description
QQQSSDBM API to start the SQL monitor
QQQCSDBM API to clear SQL monitor memory
QQQDSDBM API to dump the contents of the SQL monitor to table
QQQESDBM API to end the SQL monitor
QQQQSDBM API to query status of the database monitor

Memory-resident database monitor external table description

The memory resident database monitor uses its own set of tables instead of using the single table with
logical files that the STRDBMON monitor uses. The memory resident database monitor tables closely
match the suggested logical files of the STRDBMON monitor.

| Note: Starting with Version 4 Release 5, newly captured information will not appear through the memory
| resident monitor, and although the file format for these files did not change, the file formats for the
| file based monitor did change.
Table 7. External table Description
QAQQQRYI Query (SQL) information
QAQQTEXT SQL statement text
QAQQ3000 Table scan
QAQQ3001 Index used
QAQQ3002 Index created
QAQQ3003 Sort
QAQQ3004 Temporary table
QAQQ3007 Optimizer time out/ all indexes considered

Chapter 4. Optimizing query performance using query optimization tools 79

Table 7. External table Description (continued)
QAQQ3008 Subquery
QAQQ3010 Host variable values

Sample SQL queries

As with the STRDBMON monitor, it is up to the user to extract the information from the tables in which all
of the monitored data is stored. This can be done through any query interface that the user chooses.

If you are using Operation Navigator with the support for the SQL Monitor, you have the ability to analyze
the results direct through the GUI interface. There are a number of shipped queries that can be used or
modified to extract the information from any of the tables. For a list of these queries, go to Common

queries on analysis of DB Performance Monitor data the DB2 UDB for iSeries website .

Memory-resident database monitor row identification

The join key column QQKEY simplifies the joining of multiple tables together. This column replaces the join
field (QQJFLD) and unique query counters (QQCNT) that the database monitor used. The join key column
contains a unique identifier that allows all of the information for this query to be received from each of the
tables.

This join key column does not replace all of the detail columns that are still required to identify the specific
information about the individual steps of a query. The Query Definition Template (QDT) Number or the
Subselect Number identifies information about each detailed step. Use these columns to identify which
rows belong to each step of the query process:
v QQQDTN - Query Definition Template Number
v QQQDTL - Query Definition Template Subselect Number (Subquery)
v QQMATN - Materialized Query Definition Template Number (View)
v QQMATL - Materialized Query Definition Template Subselect Number (View w/ Subquery)
Use these columns when the monitored query contains a subquery, union, or a view operation. All query
types can generate multiple QDT’s to satisfy the original query request. The system uses these columns to
separate the information for each QDT while still allowing each QDT to be identified as belonging to this
original query (QQKEY).

View the effectiveness of your queries with Visual Explain

You can use the Visual Explain tool with Operations Navigator to create a query graph that graphically
displays the implementation of an SQL statement. You can use this tool to see information about both
static and dynamic SQL statements. Visual Explain supports the following types of SQL statements:
SELECT, INSERT, UPDATE, and DELETE.

Visual Explain can be used to find the most expensive or most time consuming operations in your query.
You can improve query performance by:
v Rewriting your SQL statement
v Changing query attributes and environment settings
v Creating indexes recommended by the query optimizer

You can use Visual Explain to:

v View the statistics that were used during optimization
v Determine whether or not an index was used to access the data. If an index was not used, Visual
Explain can help you determine which columns might benefit from being indexed.

80 DB2 UDB for iSeries Database Performance and Query Optimization V5R1
v View the effects of performing tuning techniques by comparing the before and after pictures of the
implementation.
v Obtain information about each operation (icon) in the query graph, including the total estimated cost and
estimated number of rows returned.

Visual Explain works against the data stored in the database monitor table. That is, a table that contains
the results from executing the STRDBMON command. It does not work with tables resulting from the
memory-resident monitor. There are two ways to invoke the Visual Explain tool. The first, and most
common, is through Operations Navigator. The second is through the Visual Explain API. See the OS/400
APIs information in the Programming category of the iSeries Information Center. You can launch Visual
Explain either of the following windows in Operations Navigator:
v Expand the list of available SQL Performance Monitors. Right-click on an SQL Performance Monitor and
choose the List explainable statements option. This opens the Explainable statements for SQL
performance monitor window.
v Highlight (left click) on the SQL statement that you wish to explain and click the Run Visual Explain
button.
v Enter an SQL statement in the Run SQL Scripts window. Select the statement and choose Explain...
from the context menu, or select Run and Explain... from the Visual Explain pull-down menu.

The database monitor table (results of running the STRDBMON command on the server) can be explained
through Operations Navigator. First you must import the database monitor table into Operations Navigator.
To do this, right-click on the SQL Performance Monitors and choose the Import option. Specify a name for
the performance monitor (name it will be known by within Operations Navigator) and the qualified name of
the database monitor table. Be sure to select Detailed as the type of monitor. Detailed implies the
file-based (STRDBMON) monitor while Summary implies the memory-resident monitor (which is not
supported by Visual Explain). Once the monitor has been imported, follow the steps to launch Visual
Explain from within Operations Navigator.

All the information in the query optimizer debug messages (e.g., table name, estimated number of rows,
index advised information) is shown through Visual Explain in the Attributes section. In fact, Visual Explain
and the file-based monitor (STRDBMON) contain more information than what is contained in the optimizer
debug messages:
v You can see the long and short name of tables being queried.
v You can see which columns were used in the selection to create the temporary index (remember, most
temporary indexes are sparse or select/omit indexes).
v You can see attributes about the environment when the query is executed. For example, it will show
what the ALWCPYDTA setting was.
v It will show the name and the size of the memory pool
v It will show which query INI file was used.

Another benefit of Visual Explain is its ability to differentiate the subselects within the query. If you execute
a query which contains a subquery it is sometimes difficult to determine which optimizer debug messages
belong to which subselect, the outer query or the subquery. Visual Explain handles all this.

Change the attributes of your queries with the Change Query

Attributes (CHGQRYA) command
You can modify different types of attributes of the queries that you will execute during a certain job with the
CHGQRYA command. The types of attributes that you can modify include:
v Predictive Query Governor
v Query Parallelism
v Asynchronous Job

Chapter 4. Optimizing query performance using query optimization tools 81

v Apply CHGQRYA to remote
v Query options file parameter

Before the system starts a query, the system checks the query time limit against the estimated elapsed
query time. The system also uses a time limit of zero to optimize performance on queries without having to
run through several iterations.

You can check the inquiry message CPA4259 for the predicted runtime and for what operations the query
will perform. If the query is cancelled, debug messages will still be written to the job log.

The DB2 Universal Database for iSeries Predictive Query Governor can stop the initiation of a query if the
query’s estimated or predicted runtime (elapsed execution time) is excessive. The governor acts before a
query is run instead of while a query is running. You can use it in any interactive or batch job on iSeries.
You can also use it with all DB2 Universal Database for iSeries query interfaces; it is not limited to use
with SQL queries. See “Control long-running queries with the DB2 UDB for iSeries Predictive Query
Governor” on page 89 for details.

Control queries dynamically with the query options file QAQQINI

The query options file QAQQINI support provides the ability to dynamically modify or override the
environment in which queries are executed through the CHGQRYA command and the QAQQINI file.

The query options file QAQQINI is used to set some attributes used by the Query Optimizer. For each
query that is run the query option values are retrieved from the QAQQINI file in the library specified on the
QRYOPTLIB parameter of the CHGQRYA CL command and used to optimize or implement the query.

Environmental attributes that you can modify through the QAQQINI file include:
v APPLY_REMOTE
v ASYNC_JOB_USAGE
v COMMITMENT_CONTROL_LOCK_LIMIT
v FORCE_JOIN_ORDER
v MESSAGES_DEBUG
v OPEN_CURSOR_CLOSE_COUNT
v OPEN_CURSOR_THRESHOLD
v OPTIMIZE_STATISTIC_LIMITATION
v OPTIMIZATION_GOAL
v PARALLEL_DEGREE
v PARAMETER_MARKER_CONVERSION
v QUERY_TIME_LIMIT
v REOPTIMIZE_ACCESS_PLAN
v UDF_TIME_OUT

Specifying the QAQQINI file

Use the CHGQRYA command with the QRYOPTLIB (query options library) parameter to specify which
library currently contains or will contain the query options file QAQQINI. The query options file will be
retrieved from the library specified on the QRYOPTLIB parameter for each query and remains in effect for
the duration of the job or user session, or until the QRYOPTLIB parameter is changed by the CHGQRYA
command.

If the CHGQRYA command is not issued or is issued but the QRYOPTLIB parameter is not specified, the
library QUSRSYS is searched for the existence of the QAQQINI file. If a query options file is not found for
a query, no attributes will be modified. Since the system is shipped with no INI file in QUSRSYS, you may

82 DB2 UDB for iSeries Database Performance and Query Optimization V5R1
 receive a message indicating that there is no INI file. This message is not an error but simply an indication
that a QAQQINI file that contains all default values is being used. The initial value of the QRYOPTLIB
parameter for a job is QUSRSYS.

Creating the QAQQINI query options file

Each system is shipped with a QAQQINI template file in library QSYS. The QAQQINI file in QSYS is to be
used as a template when creating all user specified QAQQINI files. To create your own QAQQINI file, use
the CRTDUPOBJ command to create a copy of the QAQQINI file in the library that will be specified on the
CHGQRYA QRYOPTLIB parameter. The file name must remain QAQQINI, for example:
| CRTDUPOBJ OBJ(QAQQINI)
| FROMLIB(QSYS)
| OBJTYPE(*FILE)
| TOLIB(MYLIB)
| DATA(*YES)

System-supplied triggers are attached to the QAQQINI file in QSYS therefore it is imperative that the only
means of copying the QAQQINI file is through the CRTDUPOBJ CL command. If another means is used,
such as CPYF, then the triggers may be corrupted and an error will be signaled that the options file cannot
be retrieved or that the options file cannot be updated.

Because of the trigger programs attached to the QAQQINI file, the following CPI321A informational
message will be displayed six times in the job log when the CRTDUPOBJ CL is used to create the file.
This is not an error. It is only an informational message.

CPI321A Information Message: Trigger QSYS_TRIG_&1___QAQQINI___00000&N in library &1 was added

to file QAQQINI in library &1. The ampersand variables (&1, &N) are replacement variables that contain
either the library name or a numeric value.

Note: It is recommended that the file QAQQINI, in QSYS, not be modified. This is the original template
that is to be duplicated into QUSRSYS or a user specified library for use.

QAQQINI query options file format

Query Options File:
UNIQUE
R QAQQINI TEXT(’Query options + file’)

QQPARM 256A VARLEN(10) +

TEXT(’Query +
option parameter’) +
COLHDG(’Parameter’)
QQVAL 256A VARLEN(10) +
TEXT(’Query option +
parameter value’) +
COLHDG(’Parameter Value’)
QQTEXT 1000G VARLEN(100) +
TEXT(’Query +
option text’) +
ALWNULL +
COLHDG(’Query Option’ +
’Text’) +
CCSID(13488) +
DFT(*NULL)
K QQPARM

The QAQQINI file shipped in the library QSYS has been pre-populated with the following rows:
Table 8. QAQQINI File Records. Description
QQPARM QQVAL
APPLY_REMOTE *DEFAULT

Chapter 4. Optimizing query performance using query optimization tools 83

Table 8. QAQQINI File Records (continued). Description
QQPARM QQVAL
ASYNC_JOB_USAGE *DEFAULT
COMMITMENT_CONTROL_LOCK_LIMIT *DEFAULT
FORCE_JOIN_ORDER *DEFAULT
MESSAGES_DEBUG *DEFAULT
OPEN_CURSOR_CLOSE_COUNT *DEFAULT
OPEN_CURSOR_THRESHOLD *DEFAULT
OPTIMIZATION_GOAL *DEFAULT
OPTIMIZE_STATISTIC_LIMITATION *DEFAULT
PARALLEL_DEGREE *DEFAULT
PARAMETER_MARKER_CONVERSION *DEFAULT
QUERY_TIME_LIMIT *DEFAULT
REOPTIMIZE_ACCESS_PLAN *DEFAULT
UDF_TIME_OUT *DEFAULT

Setting the options within the query options file

The QAQQINI file query options can be modified with the INSERT, UPDATE, or DELETE SQL statements.

For the following examples, a QAQQINI file has already been created in library MyLib. To update an
existing row in MyLib/QAQQINI use the UPDATE SQL statment. This example sets MESSAGES_DEBUG
= *YES so that the query optimizer will print out the optimizer debug messages:

UPDATE MyLib/QAQQINI SET QQVAL=’*YES’

WHERE QQPARM=’MESSAGES_DEBUG’

To delete an existing row in MyLib/QAQQINI use the DELETE SQL statement. This example removes the
QUERY_TIME_LIMIT row from the QAQQINI file:
DELETE FROM MyLib/QAQQINI
WHERE QQPARM=’QUERY_TIME_LIMIT’

To insert a new row into MyLib/QAQQINI use the INSERT SQL statement. This example adds the
QUERY_TIME_LIMIT row with a value of *NOMAX to the QAQQINI file:
INSERT INTO MyLib/QAQQINI
VALUES(’QUERY_TIME_LIMIT’,’*NOMAX’,’New time limit set by DBAdmin’)

QAQQINI query options

The following table summarizes the query options that can be specified on the QAQQINI command:

84 DB2 UDB for iSeries Database Performance and Query Optimization V5R1
Table 9. Query Options Specified on QAQQINI Command
Parameter Value Description
*DEFAULT The default value is set to *NO.
The CHGQRYA attributes for the job are not applied to the
*NO remote jobs. The remote jobs will use the attributes associated
to them on their systems.
The query attributes for the job are applied to the remote jobs
APPLY_REMOTE
used in processing database queries involving distributed
tables. For attributes where *SYSVAL is specified, the system
*YES value on the remote system is used for the remote job. This
option requires that, if CHGQRYA was used for this job, the
remote jobs must have authority to use the CHGQRYA
command.
*DEFAULT The default value is set to *LOCAL.
Asynchronous jobs may be used for database queries that
involve only tables local to the system where the database
queries are being run. In addition, for queries involving
distributed tables, this option allows the communications
*LOCAL
required to be asynchronous. This allows each system
involved in the query of the distributed tables to run its portion
of the query at the same time (in parallel) as the other
ASYNC_JOB_USAGE systems.
Asynchronous jobs may be used for database queries that
*DIST
involve distributed tables.
*ANY Asynchronous jobs may be used for any database query.
No asynchronous jobs are allowed to be used for database
query processing. In addition, all processing for queries
*NONE
involving distributed tables occurs synchronously. Therefore,
no inter-system parallel processing will occur.
*DEFAULT *DEFAULT is equivalent to 500,000,000.
COMMIT_CONTROL_LOCK The maximum number of records that can be locked to a
_LIMIT Integer Value commit transaction initiated after setting the new value. The
valid integer value is 1–500,000,000.
*DEFAULT The default is set to *NO.
*NO Allow the optimizer to re-order join tables.
Only force the join order for those queries that use the SQL
*SQL JOIN syntax. This mimics the behavior for the optimizer prior
to V4R4M0.
Only force the join position for the file listed by the numeric
FORCE_JOIN_ORDER
value nnn (nnn is optional and will default to 1) into the
*PRIMARY nnn primary position (or dial) for the join. The optimizer will then
determine the join order for all of the remaining files based
upon cost.
Do not allow the query optimizer to re-order join tables as part
*YES of its optimization process. The join will occur in the order in
which the tables were specified in the query.
*DEFAULT The default is set to *NO.
MESSAGES_DEBUG *NO No debug messages are to be displayed.
*YES Issue all Query Optimizer debug messages.

Chapter 4. Optimizing query performance using query optimization tools 85

Table 9. Query Options Specified on QAQQINI Command (continued)
Parameter Value Description
*DEFAULT *DEFAULT is equivalent to 0.
The number of cursors that are hard closed when the job has
more cursors pseudo-closed than is specified in the
OPEN_CURSOR_CLOSE_COUNT OPEN_CURSOR_THRESHOLD parameter. The value for this
Integer Value parameter should be less than or equal to the number in the
OPEN_CURSOR_THREHOLD parameter. This value is
ignored if OPEN_CURSOR_THRESHOLD is *DEFAULT. The
valid values for this parameter are 1–65536.
*DEFAULT *DEFAULT is equivalent to 0.
The number of cursors that the job can have pseudo-closed
OPEN_CURSOR_THRESHOLD before the cursors are hard (full) closed, oldest first. The valid
Integer Value user entered values for this parameter are 1–65536. Having a
value of 0 (default value) would indicate that there is no
threshold and hard closes cannot be forced.
Optimization goal is determined by the interface (ODBC, SQL
*DEFAULT
precompiler options, OPTIMIZE FOR nnn ROWS clause).
All queries will be optimized with the goal of returning the first
page of output as fast as possible. This goal works well when
the control of the output is controlled by a user who is most
*FIRSTIO
likely to abort the query after viewing the first page of output
data. Queries coded with an OPTIMIZE FOR nnn ROWS
OPTIMIZATION_GOAL
clause will honor the goal specified by the clause.
All queries will be optimized with the goal of running the entire
query to completion in the shortest amount of elapsed time.
This is a good option for when the output of a query is being
*ALLIO
written to a file or report, or the interface is queuing the output
data. Queries coded with an OPTIMIZE FOR nnn ROWS
clause will honor the goal specified by the clause.
The amount of time spent in gathering index statistics is
*DEFAULT
determined by the query optimizer.
No index statistics will be gathered by the query optimizer.
*NO Default statistics will be used for optimization. (Use this option
sparingly.)
OPTIMIZE_STATISTIC_ Specifies the maximum percentage of the index that will be
LIMITATION *PERCENTAGE
searched while gathering statistics. Valid values for are 1 to
integer value
99.
*MAX_NUMBER_ Specifies the largest table size, in number of rows, for which
OF_RECORDS_ gathering statistics is allowed. For tables with more rows than
ALLOWED integer the specified value, the optimizer will not gather statistics and
value will use default values.

86 DB2 UDB for iSeries Database Performance and Query Optimization V5R1
Table 9. Query Options Specified on QAQQINI Command (continued)
Parameter Value Description
*DEFAULT The default value is set to *SYSVAL.
The processing option used is set to the current value of the
*SYSVAL
system value, QQRYDEGREE.
Any number of tasks can be used when the database query
*IO optimizer chooses to use I/O parallel processing for queries.
SMP parallel processing is not allowed.
The query optimizer can choose to use any number of tasks
for either I/O or SMP parallel processing to process the query
or database file keyed access path build, rebuild, or
maintenance. SMP parallel processing is used only if the
system feature, DB2 Symmetric Multiprocessing for OS/400, is
installed. Use of parallel processing and the number of tasks
used is determined with respect to the number of processors
*OPTIMIZE
available in the system, this job has a share of the amount of
active memory available in the pool in which the job is run,
and whether the expected elapsed time for the query or
database file keyed access path build or rebuild is limited by
CPU processing or I/O resources. The query optimizer
PARALLEL_DEGREE
chooses an implementation that minimizes elapsed time based
on the job has a share of the memory in the pool.
The query optimizer chooses to use either I/O or SMP parallel
processing to process the query. SMP parallel processing will
only be used if the system feature, DB2 Symmetric
Multiprocessing for OS/400, is installed. The choices made by
*MAX the query optimizer are similar to those made for parameter
value *OPTIMIZE except the optimizer assumes that all active
memory in the pool can be used to process the query or
database file keyed access path build, rebuild, or
maintenance.
No parallel processing is allowed for database query
*NONE processing or database table index build, rebuild, or
maintenance.
Indicates the maximum number of tasks that can be used for a
*NUMBER_OF single query. The number of tasks will be capped off at either
_TASKS nn this value or the number of disk arms associated with the
table.
*DEFAULT The default value is set to *YES.
PARAMETER_MARKER_
*NO Constants cannot be implemented as parameter markers.
CONVERSION
*YES Constants can be implemented as parameter markers.
*DEFAULT The default value is set to *SYSVAL.
The query time limit for this job will be obtained from the
*SYSVAL
system value, QQRYTIMLMT.
*NOMAX There is no maximum number of estimated elapsed seconds.
QUERY_TIME_LIMIT
Specifies the maximum value that is checked against the
estimated number of elapsed seconds required to run a query.
integer value If the estimated elapsed seconds is greater than this value, the
query is not started. Valid values range from 0 through
2147352578.

Chapter 4. Optimizing query performance using query optimization tools 87

Table 9. Query Options Specified on QAQQINI Command (continued)
Parameter Value Description
Do not force the existing query to be reoptimized. However, if
*DEFAULT the optmizer determines that optimization is necessary, the
query will be reoptimized.
Do not force the existing query to be reoptimized. However, if
*NO the optmizer determines that optimization is necessary, the
query will be reoptimized.
*YES Force the existing query to be reoptimized.
REOPTIMIZE_ACCESS_PLAN *FORCE Force the existing query to be reoptimized.
Do not allow the plan to be reoptimized for any subjective
reasons. For these cases, continue to use the existing plan
since it is still a valid workable plan. This may mean that you
may not get all of the performance benefits that a
*ONLY_REQUIRED
reoptimization plan may derive. Subjective reasons include, file
size changes, new indexes, etc. Non-subjective reasons
include, deletion of an index used by existing access plan,
query file being deleted and recreated, etc.
The amount of time to wait is determined by the database.
*DEFAULT
The default is 30 seconds.
The maximum amount of time that the database will wait for
*MAX
the UDF to finish.
UDF_TIME_OUT
Specify the number of seconds that the database should wait
for a UDF to finish. If the value given exceeds the database
integer value maximum wait time, the maximum wait time will be used by
the database. Minimum value is 1 and maximum value is
system defined.

QAQQINI query options file authority requirements

QAQQINI is shipped with a *PUBLIC *USE authority. This allows users to view the query options file, but
not change it. It is recommended that only the system or database administrator have *CHANGE authority
to the QAQQINI query options file.

The query options file, which resides in the library specified on the CHGQRYA CL command QRYOPTLIB
parameter, is always used by the query optimizer. This is true even if the user has no authority to the
query options library and file. This provides the system administrator with an additional security
mechanism.

When the QAQQINI file resides in the library QUSRSYS the query options will effect all of the query users
on the system. To prevent anyone from inserting, deleting, or updating the query options, the system
administrator should remove update authority from *PUBLIC to the file. This will prevent users from
changing the data in the file.

When the QAQQINI file resides in a user library and that library is specified on the QRYOPTLIB parameter
of the CHGQRYA command, the query options will effect all of the queries run for that user’s job. To
prevent the query options from being retrieved from a particular library the system administrator can
revoke authority to the CHGQRYA CL command.

QAQQINI file system supplied triggers

The query options file QAQQINI file uses a system-supplied trigger program in order to process any
changes made to the file. A trigger cannot be removed from or added to the file QAQQINI.

If an error occurs on the update of the QAQQINI file (an INSERT, DELETE, or UPDATE operation), the
following SQL0443 diagnostic message will be issued:

88 DB2 UDB for iSeries Database Performance and Query Optimization V5R1
Trigger program or external routine detected an error.

Control long-running queries with the DB2 UDB for iSeries Predictive
Query Governor
The DB2 Universal Database for iSeries Predictive Query Governor can stop the initiation of a query if the
estimated or predicted run time (elapsed execution time) for the query is excessive. The governor acts
before a query is run instead of while a query is run. The governor can be used in any interactive or batch
job on the iSeries. It can be used with all DB2 Universal Database for iSeries query interfaces and is not
limited to use with SQL queries.

The ability of the governor to predict and stop queries before they are started is important because:
v Operating a long-running query and abnormally ending the query before obtaining any results wastes
system resources.
v Some operations within a query cannot be interrupted by the End Request (ENDRQS) CL command.
The creation of a temporary index or a query using a column function without a GROUP BY clause are
two examples of these types of queries. It is important to not start these operations if they will take
longer than the user wants to wait.

The governor in DB2 Universal Database for iSeries is based on the estimated runtime for a query. If the
query’s estimated runtime exceeds the user defined time limit, the initiation of the query can be stopped.

To define a time limit for the governor to use, do one of the following:
v Use the Query Time Limit (QRYTIMLMT) parameter on the Change Query Attributes (CHGQRYA) CL
command. This is the first place where the query optimizer attempts to find the time limit.
v Set the Query Time Limit option in the query options file. This is the second place where the query
optimizer attempts to find the time limit.
v Set the QQRYTIMLMT system value. Allow each job to use the value *SYSVAL on the CHGQRYA CL
command, and set the query options file to *DEFAULT. This is the third place where the query optimizer
attempts to find the time limit.

How the query governor works

The governor works in conjunction with the query optimizer. When a user requests DB2 Universal
Database for iSeries to run a query, the following occurs:
1. The query access plan is evaluated by the optimizer.
As part of the evaluation, the optimizer predicts or estimates the runtime for the query. This helps
determine the best way to access and retrieve the data for the query.
2. The estimated runtime is compared against the user-defined query time limit currently in effect for the
job or user session.
3. If the predicted runtime for the query is less than or equal to the query time limit, the query governor
lets the query run without interruption and no message is sent to the user.
4. If the query time limit is exceeded, inquiry message CPA4259 is sent to the user. The message states
that the estimated query processing time of XX seconds exceeds the time limit of YY seconds.

Note: A default reply can be established for this message so that the user does not have the option to
reply to the message, and the query request is always ended.
5. If a default message reply is not used, the user chooses to do one of the following:
v End the query request before it is actually run.
v Continue and run the query even though the predicted runtime exceeds the governor time limit.

Chapter 4. Optimizing query performance using query optimization tools 89

 Cancelling a query
When a query is expected to run longer than the set time limit, the governor issues inquiry message
CPA4259. You can respond to the message in one of the following ways:
v Enter a C to cancel the query. Escape message CPF427F is issued to the SQL runtime code. SQL
returns SQLCODE -666.
v Enter an I to ignore the time limit and let the query run to completion.

Query governor implementation considerations

It is important to remember that the time limit generated by the optimizer is only an estimate. The actual
query runtime could be more or less than the estimate, but the value of the two should be about the same.
When setting the time limit for the entire system, it is usually best to set the limit to the maximum
allowable time that any query should be allowed to run. By setting the limit too low you will run the risk of
preventing some queries from completing and thus preventing the application from successfully finishing.
There are many functions that use the query component to internally perform query requests. These
requests will also be compared to the user-defined time limit.

Query governor considerations for user applications: Setting the time

limit
Setting the time limit for jobs other than the current job

You can set the time limit for a job other than the current job. You do this by using the JOB parameter on
the CHGQRYA command to specify either a query options file library to search (QRYOPTLIB) or a specific
QRYTIMLMT for that job.

Using the time limit to balance system resources

After the source job runs the CHGQRYA command, effects of the governor on the target job is not
dependent upon the source job. The query time limit remains in effect for the duration of the job or user
session, or until the time limit is changed by a CHGQRYA command. Under program control, a user could
be given different query time limits depending on the application function being performed, the time of day,
or the amount of system resources available. This provides a significant amount of flexibility when trying to
balance system resources with temporary query requirements.

| Controlling the default reply to the query governor inquiry message

| The system administrator can control whether the interactive user has the option of ignoring the database
| query inquiry message by using the CHGJOB CL command as follows:
| v If a value of *DFT is specified for the INQMSGRPY parameter of the CHGJOB CL command, the
| interactive user does not see the inquiry messages and the query is canceled immediately.
| v If a value of *RQD is specified for the INQMSGRPY parameter of the CHGJOB CL command, the
| interactive user sees the inquiry and must reply to the inquiry.
| v If a value of *SYSRPYL is specified for the INQMSGRPY parameter of the CHGJOB CL command, a
| system reply list is used to determine whether the interactive user sees the inquiry and whether a reply
| is necessary. For more information on the *SYSRPYL parameter, see the CL command information in
| the Programming category of the iSeries Information Center. The system reply list entries can be used
| to customize different default replies based on user profile name, user id, or process names. The fully
| qualified job name is available in the message data for inquiry message CPA4259. This will allow the
| keyword CMPDTA to be used to select the system reply list entry that applies to the process or user
| profile. The user profile name is 10 characters long and starts at position 51. The process name is 10
| character long and starts at position 27.
| v The following example will add a reply list element that will cause the default reply of C to cancel any
| requests for jobs whose user profile is ’QPGMR’.
| ADDRPYLE SEQNBR(56) MSGID(CPA4259) CMPDTA(QPGMR 51) RPY(C)

90 DB2 UDB for iSeries Database Performance and Query Optimization V5R1
| The following example will add a reply list element that will cause the default reply of C to cancel any
| requests for jobs whose process name is ’QPADEV0011’.
| ADDRPYLE SEQNBR(57) MSGID(CPA4259) CMPDTA(QPADEV0011 27) RPY(C)

| Testing performance with the query governor

You can use the query governor to test the performance of your queries:
| 1. Set the query time limit to zero ( QRYTIMLMT(0) ) using the CHGQRYA command or in the INI file.
| This forces an inquiry message from the governor stating that the estimated time to run the query
| exceeds the query time limit.
2. Prompt for message help on the inquiry message and find the same information that you would find by
running the PRTSQLINF (Print SQL Information) command.

The query governor lets you optimize performance without having to run through several iterations of the
query.

Additionally, if the query is canceled, the query optimizer evaluates the access plan and sends the
optimizer debug messages to the job log. This occurs even if the job is not in debug mode. You can then
review the optimizer tuning messages in the job log to see if additional tuning is needed to obtain optimal
query performance. This allows you to try several permutations of the query with different attributes,
indexes, and/or syntax to determine what performs better through the optimizer without actually running
the query to completion. This saves on system resources because the actual query of the data is never
actually done. If the tables to be queried contain a large number of rows, this represents a significant
savings in system resources.

Be careful when you use this technique for performance testing, because all query requests will be
stopped before they are run. This is especially important for a query that cannot be implemented in a
single query step. For these types of queries, separate multiple query requests are issued, and then their
results are accumulated before returning the final results. Stopping the query in one of these intermediate
steps gives you only the performance information that relates to that intermediate step, and not for the
entire query.

Examples of setting query time limits

To set the query time limit for the current job or user session using query options file QAQQINI, specify
QRYOPTLIB parameter on the CHGQRYA command to a user library where the QAQQINI file exists with
the parameter set to QUERY_TIME_LIMIT, and the value set to a valid query time limit. For more
information on setting the query options file, see “Control queries dynamically with the query options file
QAQQINI” on page 82.

To set the query time limit for 45 seconds you would use the following CHGQRYA command:
CHGQRYA JOB(*) QRYTIMLMT(45)

This sets the query time limit at 45 seconds. If the user runs a query with an estimated runtime equal to or
less than 45 seconds, the query runs without interruption. The time limit remains in effect for the duration
of the job or user session, or until the time limit is changed by the CHGQRYA command.

Assume that the query optimizer estimated the runtime for a query as 135 seconds. A message would be
sent to the user that stated that the estimated runtime of 135 seconds exceeds the query time limit of 45
seconds.

To set or change the query time limit for a job other than your current job, the CHGQRYA command is run
using the JOB parameter. To set the query time limit to 45 seconds for job 123456/USERNAME/JOBNAME
you would use the following CHGQRYA command:
CHGQRYA JOB(123456/USERNAME/JOBNAME) QRYTIMLMT(45)

Chapter 4. Optimizing query performance using query optimization tools 91

 This sets the query time limit at 45 seconds for job 123456/USERNAME/JOBNAME. If job
123456/USERNAME/JOBNAME tries to run a query with an estimated runtime equal to or less than 45
seconds the query runs without interruption. If the estimated runtime for the query is greater than 45
seconds, for example 50 seconds, a message would be sent to the user stating that the estimated runtime
of 50 seconds exceeds the query time limit of 45 seconds. The time limit remains in effect for the duration
of job 123456/USERNAME/JOBNAME, or until the time limit for job 123456/USERNAME/JOBNAME is
changed by the CHGQRYA command.

To set or change the query time limit to the QQRYTIMLMT system value, use the following CHGQRYA
command:
CHGQRYA QRYTIMLMT(*SYSVAL)

The QQRYTIMLMT system value is used for duration of the job or user session, or until the time limit is
changed by the CHGQRYA command. This is the default behavior for the CHGQRYA command.

| Note: The query time limit can also be set in the INI file, or by using the SYSVAL command.

Control parallel processing for queries

You can turn parallel processing on and off. If the DB2 UDB Symmetric Multiprocessing feature is installed,
then you can also turn symmetric multiprocessing (SMP) on and off.
v For system wide control, use the system value QQRYDEGREE.
v For job level control, use the DEGREE parameter on the CHGQRYA command, or the
PARALLEL_DEGREE option of the query options file QAQQINI.

Even though parallelism has been enabled for a system or given job, the individual queries that run in a
job might not actually use a parallel method. This might be because of functional restrictions, or the
optimizer might choose a non-parallel method because it runs faster. See the previous sections that
describe the performance characteristics and restrictions of each of the parallel access methods. The
parallel methods that are available are:
| v Parallel table scan method
| v Parallel index scan-key selection method
| v Parallel index scan-key positioning method
| v Parallel index only access method (also for non-parallel)
| v Parallel hashing method (also for non-parallel)
| v Parallel bitmap processing method

Because queries being processed with parallel access methods aggressively use main storage, CPU, and
disk resources, the number of queries that use parallel processing should be limited and controlled.

Controlling system wide parallel processing for queries

You can use the QQRYDEGREE system value to control parallel processing for a system. The current
value of the system value can be displayed or modified using the following CL commands:
v WRKSYSVAL - Work with System Value
v CHGSYSVAL - Change System Value
v DSPSYSVAL - Display System Value
v RTVSYSVAL - Retrieve System Value

The special values for QQRYDEGREE control whether parallel processing is allowed by default for all jobs
on the system. The possible values are:
*NONE
No parallel processing is allowed for database query processing.

92 DB2 UDB for iSeries Database Performance and Query Optimization V5R1
*IO
I/O parallel processing is allowed for queries.
*OPTIMIZE
The query optimizer can choose to use any number of tasks for either I/O or SMP parallel processing
to process the queries. SMP parallel processing is used only if the DB2 UDB Symmetric
Multiprocessing feature is installed. The query optimizer chooses to use parallel processing to
minimize elapsed time based on the job’s share of the memory in the pool.
*MAX
The query optimizer can choose to use either I/O or SMP parallel processing to process the query.
SMP parallel processing can be used only if the DB2 UDB Symmetric Multiprocessing feature is
installed. The choices made by the query optimizer are similar to those made for parameter value
*OPTIMIZE, except the optimizer assumes that all active memory in the pool can be used to process
the query.

The default value of the QQRYDEGREE system value is *NONE, so the value must be changed if parallel
query processing is desired as the default for jobs run on the system.

Changing this system value affects all jobs that will be run or are currently running on the system whose
DEGREE query attribute is *SYSVAL. However, queries that have already been started or queries using
reusable ODPs are not affected.

Controlling job level parallel processing for queries

You can also control query parallel processing at the job level using the DEGREE parameter of the
Change Query Attributes (CHGQRYA) command or in the INI file. The parallel processing option allowed
and, optionally, the number of tasks that can be used when running database queries in the job can be
specified. You can prompt on the CHGQRYA command in an interactive job to display the current values
of the DEGREE query attribute.

Changing the DEGREE query attribute does not affect queries that have already been started or queries
using reusable ODPs.

Job: B,I Pgm: B,I REXX: B,I Exec

(1) (2)


CHGQRYA

*
JOB( user-name/ job-name )
job-number/

*SAME *SAME
QRYTIMLMT( *NOMAX ) DEGREE ( *NONE )
*SYSVAL *IO
seconds *OPTIMIZE
*MAX
*SYSVAL
*ANY
*NBRTASKS number-of-tasks

Chapter 4. Optimizing query performance using query optimization tools 93




*SAME *SAME
ASYNCJ ( *LOCAL ) APYRMT ( *YES )
*DIST *NO
*NONE
*ANY

Notes:
1 Value *ANY is equivalent to value *IO.
2 All parameters preceding this point can be specified in positional form.

The parameter values for the DEGREE keyword are:

*SAME
The parallel degree query attribute does not change.
*NONE
No parallel processing is allowed for database query processing.
*IO
Any number of tasks can be used when the database query optimizer chooses to use I/O parallel
processing for queries. SMP parallel processing is not allowed.
*OPTIMIZE
The query optimizer can choose to use any number of tasks for either I/O or SMP parallel processing
to process the query. SMP parallel processing can be used only if the DB2 UDB Symmetric
Multiprocessing feature is installed. Use of parallel processing and the number of tasks used is
determined with respect to the number of processors available in the system, the job’s share of the
amount of active memory available in the pool in which the job is run, and whether the expected
elapsed time for the query is limited by CPU processing or I/O resources. The query optimizer
chooses an implementation that minimizes elapsed time based on the job’s share of the memory in
the pool.
*MAX
The query optimizer can choose to use either I/O or SMP parallel processing to process the query.
SMP parallel processing can be used only if the DB2 UDB Symmetric Multiprocessing feature is
installed. The choices made by the query optimizer are similar to those made for parameter value
*OPTIMIZE except the optimizer assumes that all active memory in the pool can be used to process
the query.
*NBRTASKS number-of-tasks
Specifies the number of tasks to be used when the query optimizer chooses to use SMP parallel
processing to process a query. I/O parallelism is also allowed. SMP parallel processing can be used
only if the DB2 UDB Symmetric Multiprocessing feature is installed.
Using a number of tasks less than the number of processors available on the system restricts the
number of processors used simultaneously for running a given query. A larger number of tasks
ensures that the query is allowed to use all of the processors available on the system to run the query.
Too many tasks can degrade performance because of the over commitment of active memory and the
overhead cost of managing all of the tasks.
*SYSVAL
Specifies that the processing option used should be set to the current value of the QQRYDEGREE
system value.

The initial value of the DEGREE attribute for a job is *SYSVAL.

94 DB2 UDB for iSeries Database Performance and Query Optimization V5R1
Query optimization tools: Comparison table
PRTSQLINF STRDBG or File-based monitor Memory -Based Visual Explain
CHGQRYA Monitor
Available without Only available when Only available when Only available when Only available when
running query (after the query is run the query is run the query is run the query is explained
access plan has been
created)
Displayed for all Displayed only for Displayed only for Displayed only for Displayed only for
queries in SQL those queries which those queries which those queries which those queries that are
program, whether are executed are executed are executed explained
executed or not
Information on host Limited information on All information on All information on All information on
variable the implementation of host variables, host variables, host variables,
implementation host variables implementation, and implementation, and implementation, and
values values values
Available only to SQL Available to all query Available to all query Available only to SQL Available through
users with programs, users (OPNQRYF, users (OPNQRYF, interfaces Operations Navigator
packages, or service SQL, QUERY/400) SQL, QUERY/400) Database and API
programs interface
Messages are printed Messages is Performance rows are Performance Information is
to spool file displayed in job log written to database information is displayed visually
table collected in memory through Operations
and then written to Navigator
database table
Easier to tie Difficult to tie Uniquely identifies Repeated query Easy to view
messages to query messages to query every query, subquery requests are implementation of the
with subqueries or with subqueries or and materialized view summarized query and associated
unions unions information

Chapter 4. Optimizing query performance using query optimization tools 95

96 DB2 UDB for iSeries Database Performance and Query Optimization V5R1
Chapter 5. Using indexes to speed access to large tables
DB2 Universal Database for iSeries provides two basic means for accessing tables: a table scan
(sequential) and an index-based (direct) retrieval. Index-based retrieval is usually more efficient than table
scan. However, when a very large percentage of pages are retrieved, table scan is more efficient than
index-based retrieval.

If DB2 Universal Database for iSeries cannot use an index to access the data in a table, it will have to
read all the data in the table. Very large tables present a special performance problem: the high cost of
retrieving all the data in the table. The following topics provide suggestions that will help you to design
code which allows DB2 Universal Database for iSeries to take advantage of available indexes:
v “Coding for effective indexes: Avoid numeric conversions”
v “Coding for effective indexes: Avoid arithmetic expressions” on page 98
v “Coding for effective indexes: Avoid character string padding” on page 98
v “Coding for effective indexes: Avoid the use of like patterns beginning with % or _” on page 98
v “Coding for effective indexes: Be aware of the instances where DB2 UDB for iSeries does not use an
index” on page 99

Additional information about using indexes:

See “Coding for effective indexes: Using indexes with sort sequence” on page 100 for information about
how indexes work with sort sequence tables.

See “Examples of indexes” on page 101 for coding examples of effective indexes.

Coding for effective indexes: Avoid numeric conversions

When a column value and a host variable (or constant value) are being compared, try to specify the same
data types and attributes. DB2 Universal Database for iSeries does not use an index for the named
column if the host variable or constant value has a greater precision than the precision of the column. If
the two items being compared have different data types, DB2 Universal Database for iSeries will have to
convert one or the other of the values, which can result in inaccuracies (because of limited machine
precision). To avoid problems for columns and constants being compared, use the following:
v same data type
v same scale, if applicable
v same precision, if applicable
For example, EDUCLVL is a halfword integer value (SMALLINT). When using SQL, specify:
... WHERE EDUCLVL < 11 AND
EDUCLVL >= 2

instead of
... WHERE EDUCLVL < 1.1E1 AND
EDUCLVL > 1.3

When using the OPNQRYF command, specify:

... QRYSLT(’EDUCLVL *LT 11 *AND ENUCLVL *GE 2’)

instead of
... QRYSLT(’EDUCLVL *LT 1.1E1 *AND EDUCLVL *GT 1.3’)

© Copyright IBM Corp. 2000, 2001 97

 If an index was created over the EDUCLVL column, then the optimizer does not use the index in the
second example because the precision of the constant is greater than the precision of the column. In the
first example, the optimizer considers using the index, because the precisions are equal.

Coding for effective indexes: Avoid arithmetic expressions

| Do not use an arithmetic expression as an operand to be compared to a column in a row selection
| predicate. The optimizer does not use an index on a column that is being compared to an arithmetic
| expression. While this may not cause an index over the column to become unusable, it will prevent any
| estimates and possibly the use of index scan-key positioning on the index. The primary thing that is lost is
| the ability to use and extract any statistics that might be useful in the optimization of the query. For
| example, when using SQL, specify the following:
| ... WHERE SALARY > 16500

instead of
... WHERE SALARY > 15000*1.1

Coding for effective indexes: Avoid character string padding

Try to use the same data length when comparing a fixed-length character string column value to a host
variable or constant value. DB2 Universal Database for iSeries does not use an index if the constant value
or host variable is longer than the column length. For example, EMPNO is CHAR(6) and DEPTNO is
CHAR(3). For example, when using SQL, specify the following:
... WHERE EMPNO > ’000300’ AND
DEPTNO < ’E20’

instead of
... WHERE EMPNO > ’000300 ’ AND
DEPTNO < ’E20 ’

When using the OPNQRYF command, specify:

... QRYSLT(’EMPNO *GT "000300" *AND DEPTNO *LT "E20"’)

instead of
... QRYSLT(’EMPNO *GT "000300" *AND DEPTNO *LT "E20"’)

Coding for effective indexes: Avoid the use of like patterns beginning
with % or _
The percent sign (%), and the underline (_), when used in the pattern of a LIKE (OPNQRYF %WLDCRD)
predicate, specify a character string that is similar to the column value of rows you want to select. They
can take advantage of indexes when used to denote characters in the middle or at the end of a character
string, as in the following. For example, when using SQL, specify the following:
... WHERE LASTNAME LIKE ’J%SON%’

When using the OPNQRYF command, specify the following:

... QRYSLT(’LASTNAME *EQ %WLDCRD(’’J*SON*’’)’)

| However, when used at the beginning of a character string, they can prevent DB2 Universal Database for
| iSeries from using any indexes that might be defined on the LASTNAME column to limit the number of
| rows scanned using index scan-key positioning. Index scan-key selection, however, is allowed. For
| example, in the following queries index scan-key selection could be used, but index scan-key positioning
| could not be.

98 DB2 UDB for iSeries Database Performance and Query Optimization V5R1
 In SQL:
... WHERE LASTNAME LIKE ’%SON’

In OPNQRYF:
... QRYSLT(’LASTNAME *EQ %WLDCRD(’’*SON’’)’)

| Ideally, you should avoid patterns with a % so that you can get the best performance when you perform
| key processing on the predicate. If you can exercise control over the queries or application, you should try
| to get a partial string to search so that index scan-key positioning can be used.

| For example, if you were looking for the name ″Smithers″, but you only type ″S%,″ this query will return all
| names starting with ″S.″ You would probably then adjust the query to return all names with ″Smi%″, so by
| forcing the use of partial strings, better performance would be realized in the long term.

Coding for effective indexes: Be aware of the instances where DB2

UDB for iSeries does not use an index
DB2 Universal Database for iSeries does not use indexes in the following instances:
v For a column that is expected to be updated; for example, when using SQL, your program might include
the following:
EXEC SQL
DECLARE DEPTEMP CURSOR FOR
SELECT EMPNO, LASTNAME, WORKDEPT
FROM CORPDATA.EMPLOYEE
WHERE (WORKDEPT = ’D11’ OR
WORKDEPT = ’D21’) AND
EMPNO = ’000190’
FOR UPDATE OF EMPNO, WORKDEPT
END-EXEC.

When using the OPNQRYF command, for example:

OPNQRYF FILE((CORPDATA/EMPLOYEE)) OPTION(*ALL)
QRYSLT(’(WORKDEPT *EQ ’’D11’’ *OR WORKDEPT *EQ ’’D21’’)
*AND EMPNO *EQ ’’000190’’’)

Even if you do not intend to update the employee’s department, DB2 Universal Database for iSeries
cannot use an index with a key of WORKDEPT.

DB2 Universal Database for iSeries can use an index if all of the updateable columns used within the
index are also used within the query as an isolatable selection predicate with an equal operator. In the
previous example, DB2 Universal Database for iSeries would use an index with a key of EMPNO.

DB2 Universal Database for iSeries can operate more efficiently if the FOR UPDATE OF column list
only names the column you intend to update: WORKDEPT. Therefore, do not specify a column in the
FOR UPDATE OF column list unless you intend to update the column.

If you have an updateable cursor because of dynamic SQL or the FOR UPDATE clause was not
specified and the program contains an UPDATE statement then all columns can be updated.
v For a column being compared with another column from the same row. For example, when using SQL,
your program might include the following:
EXEC SQL
DECLARE DEPTDATA CURSOR FOR
SELECT WORKDEPT, DEPTNAME
FROM CORPDATA.EMPLOYEE
WHERE WORKDEPT = ADMRDEPT
END-EXEC.

Chapter 5. Using indexes to speed access to large tables 99

 When using the OPNQRYF command, for example:
OPNQRYF FILE (EMPLOYEE) FORMAT(FORMAT1)
QRYSLT(’WORKDEPT *EQ ADMRDEPT’)

Even though there is an index for WORKDEPT and another index for ADMRDEPT, DB2 Universal
Database for iSeries will not use either index. The index has no added benefit because every row of the
table needs to be looked at.

Coding for effective indexes: Using indexes with sort sequence

The following sections provide useful information about how indexes work with sort sequence tables.
v “Coding for effective indexes: Using indexes and sort sequence with selection, joins, or grouping”
v “Coding for effective indexes: Ordering”

For more information on how sort sequence tables work, see the topic ″Sort Sequence″ in the SQL
Reference book.

Coding for effective indexes: Using indexes and sort sequence with
selection, joins, or grouping
Before using an existing index, DB2 Universal Database for iSeries ensures the attributes of the columns
(selection, join, or grouping columns) match the attributes of the key columns in the existing index. The
sort sequence table is an additional attribute that must be compared.

The sort sequence table associated with the query (specified by the SRTSEQ and LANGID parameters)
must match the sort sequence table with which the existing index was built. DB2 Universal Database for
iSeries compares the sort sequence tables. If they do not match, the existing index cannot be used.

There is an exception to this, however. If the sort sequence table associated with the query is a
unique-weight sequence table (including *HEX), DB2 Universal Database for iSeries acts as though no
sort sequence table is specified for selection, join, or grouping columns that use the following operators
and predicates:
v equal (=) operator
v not equal (^= or <>) operator
v LIKE predicate (OPNQRYF %WLDCRD and *CT)
v IN predicate (OPNQRYF %VALUES)

When these conditions are true, DB2 Universal Database for iSeries is free to use any existing index
where the key columns match the columns and either:
v The index does not contain a sort sequence table or
v The index contains a unique-weight sort sequence table

Note: The table does not have to match the unique-weight sort sequence table associated with the query.

Note: Bitmap processing has a special consideration when multiple indexes are used for a table. If two or
more indexes have a common key column between them that is also referenced in the query
selection, then those indexes must either use the same sort sequence table or use no sort
sequence table.

Coding for effective indexes: Ordering

Unless the optimizer chooses to do a sort to satisfy the ordering request, the sort sequence table
associated with the index must match the sort sequence table associated with the query.

100 DB2 UDB for iSeries Database Performance and Query Optimization V5R1
When a sort is used, the translation is done during the sort. Since the sort is handling the sort sequence
requirement, this allows DB2 Universal Database for iSeries to use any existing index that meets the
selection criteria.

Examples of indexes
For the purposes of the examples, assume that three indexes are created.

Assume that an index HEXIX was created with *HEX as the sort sequence.
CREATE INDEX HEXIX ON STAFF (JOB)

Assume that an index UNQIX was created with a unique-weight sort sequence.
CREATE INDEX UNQIX ON STAFF (JOB)

Assume that an index SHRIX was created with a shared-weight sort sequence.
CREATE INDEX SHRIX ON STAFF (JOB)

Index example: Equals selection with no sort sequence table

Equals selection with no sort sequence table (SRTSEQ(*HEX)).
SELECT * FROM STAFF
WHERE JOB = ’MGR’

When using the OPNQRYF command, specify:

OPNQRYF FILE((STAFF))
QRYSLT(’JOB *EQ ’’MGR’’’)
SRTSEQ(*HEX)

DB2 Universal Database for iSeries could use either index HEXIX or index UNQIX.

Index example: Equals selection with a unique-weight sort sequence

table
Equals selection with a unique-weight sort sequence table (SRTSEQ(*LANGIDUNQ) LANGID(ENU)).
SELECT * FROM STAFF
WHERE JOB = ’MGR’

When using the OPNQRYF command, specify:

OPNQRYF FILE((STAFF))
QRYSLT(’JOB *EQ ’’MGR’’’)
SRTSEQ(*LANGIDUNQ) LANGID(ENU)

DB2 Universal Database for iSeries could use either index HEXIX or index UNQIX.

Index example: Equal selection with a shared-weight sort sequence

table
Equal selection with a shared-weight sort sequence table (SRTSEQ(*LANGIDSHR) LANGID(ENU)).
SELECT * FROM STAFF
WHERE JOB = ’MGR’

When using the OPNQRYF command, specify:

OPNQRYF FILE((STAFF))
QRYSLT(’JOB *EQ ’’MGR’’’)
SRTSEQ(*LANGIDSHR) LANGID(ENU)

DB2 Universal Database for iSeries could only use index SHRIX.

Chapter 5. Using indexes to speed access to large tables 101

Index example: Greater than selection with a unique-weight sort
sequence table
Greater than selection with a unique-weight sort sequence table (SRTSEQ(*LANGIDUNQ) LANGID(ENU)).
SELECT * FROM STAFF
WHERE JOB > ’MGR’

When using the OPNQRYF command, specify:

OPNQRYF FILE((STAFF))
QRYSLT(’JOB *GT ’’MGR’’’)
SRTSEQ(*LANGIDUNQ) LANGID(ENU)

DB2 Universal Database for iSeries could only use index UNQIX.

Index example: Join selection with a unique-weight sort sequence

table
Join selection with a unique-weight sort sequence table (SRTSEQ(*LANGIDUNQ) LANGID(ENU)).
SELECT * FROM STAFF S1, STAFF S2
WHERE S1.JOB = S2.JOB

or the same query using the JOIN syntax.

SELECT *
FROM STAFF S1 INNER JOIN STAFF S2
ON S1.JOB = S2.JOB

When using the OPNQRYF command, specify:

OPNQRYF FILE(STAFF STAFF)
FORMAT(FORMAT1)
JFLD((1/JOB 2/JOB *EQ))
SRTSEQ(*LANGIDUNQ) LANGID(ENU)

DB2 Universal Database for iSeries could use either index HEXIX or index UNQIX for either query.

Index example: Join selection with a shared-weight sort sequence

table
Join selection with a shared-weight sort sequence table (SRTSEQ(*LANGIDSHR) LANGID(ENU)).
SELECT * FROM STAFF S1, STAFF S2
WHERE S1.JOB = S2.JOB

or the same query using the JOIN syntax.

SELECT *
FROM STAFF S1 INNER JOIN STAFF S2
ON S1.JOB = S2.JOB

When using the OPNQRYF command, specify:

OPNQRYF FILE(STAFF STAFF) FORMAT(FORMAT1)
JFLD((1/JOB 2/JOB *EQ))
SRTSEQ(*LANGIDSHR) LANGID(ENU)

DB2 Universal Database for iSeries could only use index SHRIX for either query.

Index example: Ordering with no sort sequence table

Ordering with no sort sequence table (SRTSEQ(*HEX)).

102 DB2 UDB for iSeries Database Performance and Query Optimization V5R1
 SELECT * FROM STAFF
WHERE JOB = ’MGR’
ORDER BY JOB

When using the OPNQRYF command, specify:

OPNQRYF FILE((STAFF))
QRYSLT(’JOB *EQ ’’MGR’’’)
KEYFLD(JOB)
SRTSEQ(*HEX)

DB2 Universal Database for iSeries could only use index HEXIX.

Index example: Ordering with a unique-weight sort sequence table

Ordering with a unique-weight sort sequence table (SRTSEQ(*LANGIDUNQ) LANGID(ENU)).
SELECT * FROM STAFF
WHERE JOB = ’MGR’
ORDER BY JOB

When using the OPNQRYF command, specify:

OPNQRYF FILE((STAFF))
QRYSLT(’JOB *EQ ’’MGR’’’)
KEYFLD(JOB) SRTSEQ(*LANGIDUNQ) LANGID(ENU)

DB2 Universal Database for iSeries could only use index UNQIX.

Index example: Ordering with a shared-weight sort sequence table

Ordering with a shared-weight sort sequence table (SRTSEQ(*LANGIDSHR) LANGID(ENU)).
SELECT * FROM STAFF
WHERE JOB = ’MGR’
ORDER BY JOB

When using the OPNQRYF command, specify:

OPNQRYF FILE((STAFF))
QRYSLT(’JOB *EQ ’’MGR’’’)
KEYFLD(JOB) SRTSEQ(*LANGIDSHR) LANGID(ENU)

DB2 Universal Database for iSeries could only use index SHRIX.

Index example: Ordering with ALWCPYDTA(*OPTIMIZE) and a

unique-weight sort sequence table
Ordering with ALWCPYDTA(*OPTIMIZE) and a unique-weight sort sequence table
(SRTSEQ(*LANGIDUNQ) LANGID(ENU)).
SELECT * FROM STAFF
WHERE JOB = ’MGR’
ORDER BY JOB

When using the OPNQRYF command, specify:

OPNQRYF FILE((STAFF))
QRYSLT(’JOB *EQ ’’MGR’’’)
KEYFLD(JOB)
SRTSEQ(*LANGIDUNQ) LANGID(ENU)
ALWCPYDTA(*OPTIMIZE)

DB2 Universal Database for iSeries could use either index HEXIX or index UNQIX for selection. Ordering
would be done during the sort using the *LANGIDUNQ sort sequence table.

Chapter 5. Using indexes to speed access to large tables 103

Index example: Grouping with no sort sequence table
Grouping with no sort sequence table (SRTSEQ(*HEX)).
SELECT JOB FROM STAFF
GROUP BY JOB

When using the OPNQRYF command, specify:

OPNQRYF FILE((STAFF)) FORMAT(FORMAT2)
GRPFLD((JOB))
SRTSEQ(*HEX)

DB2 Universal Database for iSeries could use either index HEXIX or index UNQIX.

Index example: Grouping with a unique-weight sort sequence table

Grouping with a unique-weight sort sequence table (SRTSEQ(*LANGIDUNQ) LANGID(ENU)).
SELECT JOB FROM STAFF
GROUP BY JOB

When using the OPNQRYF command, specify:

OPNQRYF FILE((STAFF)) FORMAT(FORMAT2)
GRPFLD((JOB))
SRTSEQ(*LANGIDUNQ) LANGID(ENU)

DB2 Universal Database for iSeries could use either index HEXIX or index UNQIX.

Index example: Grouping with a shared-weight sort sequence table

Grouping with a shared-weight sort sequence table (SRTSEQ(*LANGIDSHR) LANGID(ENU)).
SELECT JOB FROM STAFF
GROUP BY JOB

When using the OPNQRYF command, specify:

OPNQRYF FILE((STAFF)) FORMAT(FORMAT2)
GRPFLD((JOB))
SRTSEQ(*LANGIDSHR) LANGID(ENU)

DB2 Universal Database for iSeries could only use index SHRIX.

The following examples assume that 3 more indexes are created over columns JOB and SALARY. The
CREATE INDEX statements precede the examples.

Assume an index HEXIX2 was created with *HEX as the sort sequence.
CREATE INDEX HEXIX2 ON STAFF (JOB, SALARY)

Assume that an index UNQIX2 was created and the sort sequence is a unique-weight sort sequence.
CREATE INDEX UNQIX2 ON STAFF (JOB, SALARY)

Assume an index SHRIX2 was created with a shared-weight sort sequence.

CREATE INDEX SHRIX2 ON STAFF (JOB, SALARY)

Index example: Ordering and grouping on the same columns with a

unique-weight sort sequence table
Ordering and grouping on the same columns with a unique-weight sort sequence table
(SRTSEQ(*LANGIDUNQ) LANGID(ENU)).

104 DB2 UDB for iSeries Database Performance and Query Optimization V5R1
 SELECT JOB, SALARY FROM STAFF
GROUP BY JOB, SALARY
ORDER BY JOB, SALARY

When using the OPNQRYF command, specify:

OPNQRYF FILE((STAFF)) FORMAT(FORMAT3)
GRPFLD(JOB SALARY)
KEYFLD(JOB SALARY)
SRTSEQ(*LANGIDUNQ) LANGID(ENU)

DB2 Universal Database for iSeries could use UNQIX2 to satisfy both the grouping and ordering
requirements. If index UNQIX2 did not exist, DB2 Universal Database for iSeries would create an index
using a sort sequence table of *LANGIDUNQ.

Index example: Ordering and grouping on the same columns with

ALWCPYDTA(*OPTIMIZE) and a unique-weight sort sequence table
Ordering and grouping on the same columns with ALWCPYDTA(*OPTIMIZE) and a unique-weight sort
sequence table (SRTSEQ(*LANGIDUNQ) LANGID(ENU)).
SELECT JOB, SALARY FROM STAFF
GROUP BY JOB, SALARY
ORDER BY JOB, SALARY

When using the OPNQRYF command, specify:

OPNQRYF FILE((STAFF)) FORMAT(FORMAT3)
GRPFLD(JOB SALARY)
KEYFLD(JOB SALARY)
SRTSEQ(*LANGIDUNQ) LANGID(ENU)
ALWCPYDTA(*OPTIMIZE)

DB2 Universal Database for iSeries could use UNQIX2 to satisfy both the grouping and ordering
requirements. If index UNQIX2 did not exist, DB2 Universal Database for iSeries would either:
v Create an index using a sort sequence table of *LANGIDUNQ or
v Use index HEXIX2 to satisfy the grouping and to perform a sort to satisfy the ordering

Index example: Ordering and grouping on the same columns with a

shared-weight sort sequence table
Ordering and grouping on the same columns with a shared-weight sort sequence table
(SRTSEQ(*LANGIDSHR) LANGID(ENU)).
SELECT JOB, SALARY FROM STAFF
GROUP BY JOB, SALARY
ORDER BY JOB, SALARY

When using the OPNQRYF command, specify:

OPNQRYF FILE((STAFF)) FORMAT(FORMAT3)
GRPFLD(JOB SALARY)
KEYFLD(JOB SALARY)
SRTSEQ(*LANGIDSHR) LANGID(ENU)

DB2 Universal Database for iSeries could use SHRIX2 to satisfy both the grouping and ordering
requirements. If index SHRIX2 did not exist, DB2 Universal Database for iSeries would create an index
using a sort sequence table of *LANGIDSHR.

Chapter 5. Using indexes to speed access to large tables 105

Index example: Ordering and grouping on the same columns with
ALWCPYDTA(*OPTIMIZE) and a shared-weight sort sequence table
Ordering and grouping on the same columns with ALWCPYDTA(*OPTIMIZE) and a shared-weight sort
sequence table (SRTSEQ(*LANGIDSHR) LANGID(ENU).
SELECT JOB, SALARY FROM STAFF
GROUP BY JOB, SALARY
ORDER BY JOB, SALARY

When using the OPNQRYF command, specify:

OPNQRYF FILE((STAFF)) FORMAT(FORMAT3)
GRPFLD(JOB SALARY)
KEYFLD(JOB SALARY)
SRTSEQ(*LANGIDSHR) LANGID(ENU)
ALWCPYDTA(*OPTIMIZE)

DB2 Universal Database for iSeries could use SHRIX2 to satisfy both the grouping and ordering
requirements. If index SHRIX2 did not exist, DB2 Universal Database for iSeries would create an index
using a sort sequence table of *LANGIDSHR.

Index example: Ordering and grouping on different columns with a

unique-weight sort sequence table
Ordering and grouping on different columns with a unique-weight sort sequence table
(SRTSEQ(*LANGIDUNQ) LANGID(ENU)).
SELECT JOB, SALARY FROM STAFF
GROUP BY JOB, SALARY
ORDER BY SALARY, JOB

When using the OPNQRYF command, specify:

OPNQRYF FILE((STAFF)) FORMAT(FORMAT3)
GRPFLD(JOB SALARY)
KEYFLD(SALARY JOB)
SRTSEQ(*LANGIDSHR) LANGID(ENU)

DB2 Universal Database for iSeries could use index HEXIX2 or index UNQIX2 to satisfy the grouping
requirements. A temporary result would be created containing the grouping results. A temporary index
would then be built over the temporary result using a *LANGIDUNQ sort sequence table to satisfy the
ordering requirements.

Index example: Ordering and grouping on different columns with

ALWCPYDTA(*OPTIMIZE) and a unique-weight sort sequence table
Ordering and grouping on different columns with ALWCPYDTA(*OPTIMIZE) and a unique-weight sort
sequence table (SRTSEQ(*LANGIDUNQ) LANGID(ENU)).
SELECT JOB, SALARY FROM STAFF
GROUP BY JOB, SALARY
ORDER BY SALARY, JOB

When using the OPNQRYF command, specify:

OPNQRYF FILE((STAFF)) FORMAT(FORMAT3)
GRPFLD(JOB SALARY)
KEYFLD(SALARY JOB)
SRTSEQ(*LANGIDUNQ) LANGID(ENU)
ALWCPYDTA(*OPTIMIZE)

DB2 Universal Database for iSeries could use index HEXIX2 or index UNQIX2 to satisfy the grouping
requirements. A sort would be performed to satisfy the ordering requirements.

106 DB2 UDB for iSeries Database Performance and Query Optimization V5R1
Index example: Ordering and grouping on different columns with
ALWCPYDTA(*OPTIMIZE) and a shared-weight sort sequence table
Ordering and grouping on different columns with ALWCPYDTA(*OPTIMIZE) and a shared-weight sort
sequence table (SRTSEQ(*LANGIDSHR) LANGID(ENU)).
SELECT JOB, SALARY FROM STAFF
GROUP BY JOB, SALARY
ORDER BY SALARY, JOB

When using the OPNQRYF command, specify:

OPNQRYF FILE((STAFF)) FORMAT(FORMAT3)
GRPFLD(JOB SALARY)
KEYFLD(SALARY JOB)
SRTSEQ(*LANGIDSHR) LANGID(ENU)
ALWCPYDTA(*OPTIMIZE)

DB2 Universal Database for iSeries could use index SHRIX2 to satisfy the grouping requirements. A sort
would be performed to satisfy the ordering requirements.

What are encoded vector indexes?

An encoded vector index (EVI) is an index object that is used by the query optimizer and database engine
to provide fast data access in decision support and query reporting environments. EVIs are a
complementary alternative to existing index objects (binary radix tree structure - logical file or SQL index)
and are a variation on bitmap indexing. Because of their compact size and relative simplicity, EVIs provide
for faster scans of a table that can also be processed in parallel.

An EVI is a data structure that is stored as two components:

v The symbol table contains statistical and descriptive information about each distinct key value
represented in the table. Each distinct key is assigned a unique code, either 1, 2 or 4 bytes in size.
v The vector is an array of codes listed in the same ordial position as the rows in the table. The vector
does not contain any pointers to the actual rows in the table.

Advantages of EVIs
v Require less storage
v May have better build times
v Provide more accurate statistics to the query optimizer

Disadvantages of EVIs
v Cannot be used in ordering and grouping
v Have limited use in joins
v Some additional maintenance idiosyncrasies

How the EVI works

The optimizer uses the symbol table to collect the costing information about the query. If the optimizer
decides to use an EVI to process the query, the database engine uses the vector to build the dynamic
bitmap that contains one bit for each row in the table. If the row satisfies the query selection, the bit is set
on. If the row does not satisfy the query selection, the bit is set off. Like a bitmap index, intermediate
dynamic bitmaps can be AND’ed and OR’ed together to satisfy an ad hoc query. For example, if a user
wants to see sales data for a certain region during a certain time period, you can define an EVI over the
region column and the Quarter column of the database. When the query runs, the database engine builds
dynamic bitmaps using the two EVIs and then ANDs the bitmaps together to produce a bitmap that
contains only the relevant rows for both selection criteria. This AND’ing capability drastically reduces the
number of rows that the system must read and test. The dynamic bitmap(s) exists only as long as the
query is executing. Once the query is completed, the dynamic bitmap(s) are eliminated.

Chapter 5. Using indexes to speed access to large tables 107

When should EVIs be used?
Encoded vector indexes should be considered when you want to gather statistics, when full table scan is
selected, selectivity of the query is 20%-70% and using skip sequential access with dynamic bitmaps will
speed up the scan, or when a star schema join is expected to be used for star schema join queries.
Encoded vector indexes should be created with:
v Single key columns with a low number of distinct values expected
v Keys columns with a low volatililty (they don’t change often)
v Maximum number of distinct values expected using the WITH n DISTINCT VALUES clause
v Single key over foreign key columns for a star schema model
General index maintenance EVI maintenance When using EVIs, there are a unique challenges to index
maintenance. The following table shows a progression of how EVIs are maintained and the conditions
under which EVIs are most effective andto the conditions where EVIs are least effective based on the EVI
maintenance .idiosyncrasies.

General index maintenance

Whenever indexes are created and used, there is a potential for a decrease in I/O velocity due to
maintenance, therefore, it is essential that you consider the maintenance cost of creating and using
additional indexes. For radix indexes with MAINT(*IMMED) and EVIs, maintenance occurs when inserting,
updating or deleting rows.

To reduce the maintenance of your indexes consider:

v Minimizing the number of indexes over a given table
v Dropping indexes during batch inserts, updates, and deletes
v Creating indexes, one at a time, in parallel using SMP
v Creating multiple indexes simultaneously with multiple batch jobs using multiple CPUs
v Maintaining indexes in parallel using SMP

The goal of creating indexes for performance is to balance the maximum number of indexes for statistics
and implementation while minimizing the number of indexes to maintain.

EVI maintenance
When using EVIs, there are a unique challenges to index maintenance. The following table shows a
progression of how EVIs are maintained and the conditions under which EVIs are most effective and
where EVIs are least effective based on the EVI maintenance idiosyncrasies.

108 DB2 UDB for iSeries Database Performance and Query Optimization V5R1
Table 10. EVI Maintenance Considerations
Condition Characteristics
When inserting an existing distinct v Minimum overhead
key value
v Symbol table key value looked up
and statistics updated
v Vector element added for new row,
with existing byte code
When inserting a new distinct key v Minimum overhead
value - in order, within byte code
v Symbol table key value added,
range
byte code assigned, statistics
assigned
Most Effective
v Vector element added for new row,
with new byte code
When inserting a new distinct key v Minimum overhead if contained
value - out of order, within byte code within overflow area threshold
range
v Symbol table key value added to
overflow area, byte code assigned,
statistics assigned
v Vector element added for new row,
with new byte code
v Considerable overhead if overflow
area threshold reached
v Access path validated - not
available
v EVI refreshed, overflow area keys
incorporated, new byte codes
assigned (symbol table and vector
elements updated)
When inserting a new distinct key v Considerble overhead
value - out of byte code range
v Access plan invalidated - not
available
Least Effective
v EVI refreshed, next byte code size
used, new byte codes assigned
(symbol table and vector elements
updated

Recommendations for EVI use

Encoded vector indexes are a powerful tool for providing fast data access in decision support and query
reporting environments, but to ensure the effective use of EVIs, you should implement EVIs with the
following guidelines:

Create EVIs on:

v Read only tables or tables with a minimum of INSERT, UPDATE, DELETE activity.
v Key columns that are used in the WHERE clause - local selection predicates of SQL requests.
v Single key columns that have a relatively small set of distinct values.
v Multiple key columns that result in a relatively small set of distinct values.
v Key columns that have a static or relatively static set of distinct values.
v Non-unique key columns, with many duplicates.

Create EVIs with the maximum byte code size expected:

Chapter 5. Using indexes to speed access to large tables 109

v Use the ″WITH n DISTINCT VALUES″ clause on the CREATE ENCODED VECTOR INDEX statement.
v If unsure, use a number greater than 65,535 to create a 4 byte code, thus avoiding the EVI
maintenance overhead of switching byte code sizes.

When loading data:

v Drop EVIs, load data, create EVIs.
v EVI byte code size will be assigned automatically based on the number of actual distinct key values
found in the table.
v Symbol table will contain all key values, in order, no keys in overflow area.

Consider SMP and parallel index creation and maintenance:

Symmetrical Multiprocessing (SMP) is a valuable tool for building and maintaining indexes in parallel. The
results of using the optional SMP feature of OS/400 are faster index build times, and faster I/O velocities
while maintaining indexes in parallel. Using an SMP degree value of either *OPTIMIZE or *MAX, additional
multiple tasks and additional system resources are used to build or maintain the indexes. With a degree
value of *MAX, expect linear scalability on index creation. For example, creating indexes on a 4 processor
system can be 4 times as fast as a 1 processor system.

Checking values in the overflow area:

You can also use the Display Field Description (DSPFD) command (or V4R5M0 Operations Navigator -
Database) to check how many values are in the overflow area. Once the DSPFD command is issued,
check the overflow area parameter for details on the initial and actual number of distinct key values in the
overflow area.

Using CHGLF to rebuild an index’s access path:

Use the Change Logical File (CHGLF) command with the attribute Force Rebuild Access Path set to YES
(FRCRBDAP(*YES)). This command accomplishes the same thing as dropping and recreating the index,
but it does not require that you know about how the index was built. This command is especially effective
for applications where the original index definitions are not available, or for refreshing the access path.

110 DB2 UDB for iSeries Database Performance and Query Optimization V5R1
Chapter 6. Application design tips for database performance
This section contains the following design tips that you can apply when designing SQL applications to
maximize your database performance:
v “Database application design tips: Use structure parameter passing techniques”
v “Database application design tips: Use live data” on page 112
v “Database application design tips: Reduce the number of open operations” on page 114
v “Database application design tips: Retain cursor positions” on page 116

Database application design tips: Use structure parameter passing

techniques
The SQL precompiler uses two methods of passing host variables to the SQL run time.

The original method was to pass each host variable as a separate parameter. For example:
CALL QSQROUTE
(SQLCA, hostvariable1, hostvariable2, hostvariable3)

The second method is to create a data structure with an element for each host variable that is referenced
in the statement. Then that data structure could be passed as a parameter. For example:
CALL QSQROUTE
(SQLCA, hostvariable structure)

The second method will provide better performance.

Note: The structure parameter passing technique is not used for SQL statements for special cases in PL/I
and RPG for iSeries programs, which are discussed in the SQL Programming with Host Languages
book.

Database application design tips: Background information on

parameter passing
Additional code is created in the program to move the input host variable data to the structure before the
call to the SQL program. After the call to the SQL program, code is added that moves the data from the
structure to the output host variables.

The precompiler creates the structure so that the SQL header is first, followed by the input host variables,
the output host variables, the indicators for the input host variables, and the indicators for the output host
variables. Because the output host variables are created in a contiguous storage space, the SQL run-time
support can check for a match with the I/O buffer (each result column attribute is checked for a matching
host variable attribute) and move the data with one instruction if they all match.

The SQL header on the structure contains information unique to the statement so that SQL does not have
to reconstruct the information on each call. Some of this information is added and processed by the SQL
run-time support. SQL run-time support creates an SQLDA internally for each statement that uses host
variables. With the original parameter list, the host variable address could be different for each call of the
statement and, therefore, SQL run-time support rebuilds the SQLDA for each call of the statement. With
structure parameter passing, the structure is created as a static variable, and the address of the elements
will not change. SQL run-time support builds the SQLDA the first time the statement is called, and saves
the SQLDA so it can be used on future calls of the statement.

With the original type of parameter passing, the number of host variables that could be referred to in a
program was approximately 4000 because of an architecture limit of 4100 pointers in a program. Each

© Copyright IBM Corp. 2000, 2001 111

parameter required a pointer. With the structure parameter passing, only two parameters are passed for
each SQL statement; therefore, the limit of 4000 pointers does not apply.

Database application design tips: Differences in structure parameter

passing techniques
You need to be aware of the following differences in the parameter passing techniques when embedding
SQL statements into application programs:
v SQLCODE +326 is returned when the structure parameter passing technique is used and the number of
host variables specified is greater than the number of columns in the select list.
The extra output host variables are ignored by the original parameter passing technique, but with the
structure parameter passing technique, the host variable is set to the value in the SQL created structure
(blanks are used for character host variables and zeros are used for numeric host variables).
Attention: This could cause incorrect results when the extra output host variables overlap with other
host variables.

An example of this problem, using RPG for iSeries, follows:

ISTRUC DS
I 1 1 A
I 2 2 B
I 1 2 C

SELECT * INTO :STRUC FROM ATABLE

In the above example, if ATABLE has only one or two columns, the SQLCODE will be set to +326.
When the assignment to C from the SQL structure is done, the contents of A and B will be blank instead
of the value of the column corresponding to A and B.
v With the original parameter passing technique, SQLCODE -302 or -304 is returned when a conversion
error occurs (because of numeric data that is not valid) while processing the data for a host variable.
However, with the structure parameter passing technique, SQL does not detect this error. The
conversion error occurs in the host language statements that reference the host variable. For example,
if a DECIMAL(5,2) input host variable contains the invalid data ’FFFFFF’X, an error will occur in the host
language when the data is moved into the data structure.
v The structure created by SQL uses names that start with the letters SQL. If existing programs use
variable names starting with SQL, those names may conflict with the SQL-created names.
v The contents of the SQL-created data structure must not be changed by the application programs.

Database application design tips: Use live data

The term live data refers to the type of access that the database manager uses when it retrieves data
without making a copy of the data. Using this type of access, the data, which is returned to the program,
always reflects the current values of the data in the database. The programmer can control whether the
database manager uses a copy of the data or retrieves the data directly. This is done by specifying the
allow copy data (ALWCPYDTA) parameter on the precompiler commands or on the Start SQL (STRSQL)
command.

Specifying ALWCPYDTA(*NO) instructs the database manager to always use live data. Live data access
can be used as a performance advantage because the cursor does not have to be closed and opened
again to refresh the data being retrieved. An example application demonstrating this advantage is one that
produces a list on a display. If the display screen can only show 20 elements of the list at a time, then,
after the initial 20 elements are displayed, the application programmer can request that the next 20 rows
be displayed. A typical SQL application designed for an operating system other than the OS/400 operating
system, might be structured as follows:

112 DB2 UDB for iSeries Database Performance and Query Optimization V5R1
EXEC SQL
DECLARE C1 CURSOR FOR
SELECT EMPNO, LASTNAME, WORKDEPT
FROM CORPDATA.EMPLOYEE
ORDER BY EMPNO
END-EXEC.

EXEC SQL
OPEN C1
END-EXEC.

* PERFORM FETCH-C1-PARA 20 TIMES.

MOVE EMPNO to LAST-EMPNO.

EXEC SQL
CLOSE C1
END-EXEC.

* Show the display and wait for the user to indicate that
* the next 20 rows should be displayed.

EXEC SQL
DECLARE C2 CURSOR FOR
SELECT EMPNO, LASTNAME, WORKDEPT
FROM CORPDATA.EMPLOYEE
WHERE EMPNO > :LAST-EMPNO
ORDER BY EMPNO
END-EXEC.

EXEC SQL
OPEN C2
END-EXEC.

* PERFORM FETCH-C21-PARA 20 TIMES.

* Show the display with these 20 rows of data.

EXEC SQL
CLOSE C2
END-EXEC.

In the above example, notice that an additional cursor had to be opened to continue the list and to get
current data. This could result in creating an additional ODP that would increase the processing time on
the iSeries system. In place of the above example, the programmer could design the application specifying
ALWCPYDTA(*NO) with the following SQL statements:
EXEC SQL
DECLARE C1 CURSOR FOR
SELECT EMPNO, LASTNAME, WORKDEPT
FROM CORPDATA.EMPLOYEE
ORDER BY EMPNO
END-EXEC.

EXEC SQL
OPEN C1
END-EXEC.

* Display the screen with these 20 rows of data.

* PERFORM FETCH-C1-PARA 20 TIMES.

* Show the display and wait for the user to indicate that
* the next 20 rows should be displayed.

* PERFORM FETCH-C1-PARA 20 TIMES.

Chapter 6. Application design tips for database performance 113

EXEC SQL
CLOSE C1
END-EXEC.

In the above example, the query could perform better if the FOR 20 ROWS clause was used on the
multiple-row FETCH statement. Then, the 20 rows would be retrieved in one operation.

Database application design tips: Reduce the number of open

operations
The SQL data manipulation language statements must do database open operations in order to create an
open data path (ODP) to the data. An open data path is the path through which all input/output operations
for the table are performed. In a sense, it connects the SQL application to a table. The number of open
operations in a program can significantly affect performance. A database open operation occurs on:
v An OPEN statement
v SELECT INTO statement
v An INSERT statement with a VALUES clause
v An UPDATE statement with a WHERE condition
v An UPDATE statement with a WHERE CURRENT OF cursor and SET clauses that refer to operators or
functions
v SET statement that contains an expression
v VALUES INTO statement that contains an expression
v A DELETE statement with a WHERE condition

An INSERT statement with a select-statement requires two open operations. Certain forms of subqueries
may also require one open per subselect.

To minimize the number of opens, DB2 Universal Database for iSeries leaves the open data path (ODP)
open and reuses the ODP if the statement is run again, unless:
v The ODP used a host variable to build a subset temporary index. The OS/400 database support may
choose to build a temporary index with entries for only the rows that match the row selection specified
in the SQL statement. If a host variable was used in the row selection, the temporary index will not have
the entries required for a different value contained in the host variable.
v Ordering was specified on a host variable value.
v An Override Database File (OVRDBF) or Delete Override (DLTOVR) CL command has been issued
since the ODP was opened, which would affect the SQL statement execution.

Note: Only overrides that affect the name of the table being referred to will cause the ODP to be
closed within a given program invocation.
v The join is a complex join that requires temporaries to contain the intermediate steps of the join.
v Some cases involve a complex sort, where a temporary file is required, may not be reusable.
v A change to the library list since the last open has occurred, which would change the table selected by
an unqualified referral in system naming mode.
v The join was implemented using hash join.

DB2 Universal Database for iSeries only reuses ODPs opened by the same statement. An identical
statement coded later in the program does not reuse an ODP from any other statement. If the identical
statement must be run in the program many times, code it once in a subroutine and call the subroutine to
run the statement.

The ODPs opened by DB2 Universal Database for iSeries are closed when any of the following occurs:

114 DB2 UDB for iSeries Database Performance and Query Optimization V5R1
v A CLOSE, INSERT, UPDATE, DELETE, or SELECT INTO statement completes and the ODP required a
temporary result or a subset temporary index.
v The Reclaim Resources (RCLRSC) command is issued. A RCLRSC is issued when the first COBOL
program on the call stack ends or when a COBOL program issues the STOP RUN COBOL statement.
RCLRSC will not close ODPs created for programs precompiled using CLOSQLCSR(*ENDJOB). For
interaction of RCLRSC with non-default activation groups, see the following books:
– ILE C/C++ Programmer’s Guide
– ILE COBOL Programmer’s Guide
– ILE RPG Programmer’s Guide
v When the last program that contains SQL statements on the call stack exits, except for ODPs created
for programs precompiled using CLOSQLCSR(*ENDJOB) or modules precompiled using
CLOSQLCSR(*ENDACTGRP).
v When a CONNECT (Type 1) statement changes the application server for an activation group, all ODPs
created for the activation group are closed.
v When a DISCONNECT statement ends a connection to the application server, all ODPs for that
application server are closed.
v When a released connection is ended by a successful COMMIT, all ODPs for that application server are
closed.
v When the threshold for open cursors specified by the query options file (QAQQINI) parameter
OPEN_CURSOR_THRESHOLD is exceeded.

You can control whether DB2 Universal Database for iSeries keeps the ODPs open in the following ways:
v Design the application so a program that issues an SQL statement is always on the call stack
v Use the CLOSQLCSR(*ENDJOB) or CLOSQLCSR(*ENDACTGRP) parameter
v By specifying the OPEN_CURSOR_THRESHOLD and OPEN_CURSOR_CLOSE_COUNT parameters
of the query options file (QAQQINI)

DB2 Universal Database for iSeries does an open operation for the first execution of each UPDATE
WHERE CURRENT OF when any expression in the SET clause contains an operator or function. The
open can be avoided by coding the function or operation in the host language code.

For example, the following UPDATE causes DB2 Universal Database for iSeries to do an open operation:
EXEC SQL
FETCH EMPT INTO :SALARY
END-EXEC.

EXEC SQL
UPDATE CORPDATA.EMPLOYEE
SET SALARY = :SALARY + 1000
WHERE CURRENT OF EMPT
END-EXEC.

Instead, use the following coding technique to avoid opens:

EXEC SQL
FETCH EMPT INTO :SALARY
END EXEC.

ADD 1000 TO SALARY.

EXEC SQL
UPDATE CORPDATA.EMPLOYEE
SET SALARY = :SALARY
WHERE CURRENT OF EMPT
END-EXEC.

Chapter 6. Application design tips for database performance 115

You can use the CL commands Trace Job (TRCJOB) or Display Journal (DSPJRN) to determine the
number of opens that are being performed by an SQL statement.

Database application design tips: Retain cursor positions

You can improve performance by retaining cursor positions. The next two sections provide information on
retaining cursors for non-ILE and ILE program calls.

Database application design tips: Retaining cursor positions for

non-ILE program calls
For non-ILE program calls, the close SQL cursor (CLOSQLCSR) parameter allows you to specify the
scope of the following:
v The cursors
v The prepared statements
v The locks

When used properly, the CLOSQLCSR parameter can reduce the number of SQL OPEN, PREPARE, and
LOCK statements needed. It can also simplify applications by allowing you to retain cursor positions
across program calls.
*ENDPGM
This is the default for all non-ILE precompilers. With this option, a cursor remains open and
accessible only while the program that opened it is on the call stack. When the program ends, the
SQL cursor can no longer be used. Prepared statements are also lost when the program ends.
Locks, however, remain until the last SQL program on the call stack has completed.
*ENDSQL
With this option, SQL cursors and prepared statements that are created by a program remain
open until the last SQL program on the call stack has completed. They cannot be used by other
programs, only by a different call to the same program. Locks remain until the last SQL program in
the call stack completes.
*ENDJOB
This option allows you to keep SQL cursors, prepared statements, and locks active for the
duration of the job. When the last SQL program on the stack has completed, any SQL resources
created by *ENDJOB programs are still active. The locks remain in effect. The SQL cursors that
were not explicitly closed by the CLOSE, COMMIT, or ROLLBACK statements remain open. The
prepared statements are still usable on subsequent calls to the same program.

Database application design tips: Retaining cursor positions across

ILE program calls
For ILE program calls, the close SQL cursor (CLOSQLCSR) parameter allows you to specify the scope of
the following:
v The cursors
v The prepared statements
v The locks
When used properly, the CLOSQLCSR parameter can reduce the number of SQL OPEN, PREPARE, and
LOCK statements needed. It can also simplify applications by allowing you to retain cursor positions
across program calls.
*ENDACTGRP
This is the default for the ILE precompilers. With this option, SQL cursors and prepared

116 DB2 UDB for iSeries Database Performance and Query Optimization V5R1
 statements remain open until the activation group that the program is running under ends. They
cannot be used by other programs, only by a different call to the same program. Locks remain
until the activation group ends.
*ENDMOD
With this option, a cursor remains open and accessible only while the module that opened it is
active. When the module ends, the SQL cursor can no longer be used. Prepared statements will
also be lost when the module ends. Locks, however, remain until the last SQL program in the call
stack completes.

Database application design tips: General rules for retaining cursor

positions for all program calls
When using programs compiled with either CLOSQLCSR(*ENDPGM) or CLOSQLCSR(*ENDMOD), a
cursor must be opened every time the program or module is called, in order to access the data. If the SQL
program or module is going to be called several times, and you want to take advantage of a reusable
ODP, then the cursor must be explicitly closed before the program or module exits.

Using the CLOSQLCSR parameter and specifying *ENDSQL, *ENDJOB, or *ENDACTGRP, you may not
need to run an OPEN and a CLOSE statement on every call. In addition to having fewer statements to
run, you can maintain the cursor position between calls to the program or module.

The following examples of SQL statements help demonstrate the advantage of using the CLOSQLCSR
parameter:
EXEC SQL
DECLARE DEPTDATA CURSOR FOR
SELECT EMPNO, LASTNAME
FROM CORPDATA.EMPLOYEE
WHERE WORKDEPT = :DEPTNUM
END-EXEC.

EXEC SQL
OPEN DEPTDATA
END-EXEC.

EXEC SQL
FETCH DEPTDATA INTO :EMPNUM, :LNAME
END-EXEC.

EXEC SQL
CLOSE DEPTDATA
END-EXEC.

If this program is called several times from another SQL program, it will be able to use a reusable ODP.
This means that, as long as SQL remains active between the calls to this program, the OPEN statement
will not require a database open operation. However, the cursor is still positioned to the first result row
after each OPEN statement, and the FETCH statement will always return the first row.

In the following example, the CLOSE statement has been removed:

EXEC SQL
DECLARE DEPTDATA CURSOR FOR
SELECT EMPNO, LASTNAME
FROM CORPDATA.EMPLOYEE
WHERE WORKDEPT = :DEPTNUM
END-EXEC.

IF CURSOR-CLOSED IS = TRUE THEN

EXEC SQL
OPEN DEPTDATA
END-EXEC.

Chapter 6. Application design tips for database performance 117

 EXEC SQL
FETCH DEPTDATA INTO :EMPNUM, :LNAME
END-EXEC.

If this program is precompiled with the *ENDJOB option or the *ENDACTGRP option and the activation
group remains active, the cursor position is maintained. The cursor position is also maintained when the
following occurs:
v The program is precompiled with the *ENDSQL option.
v SQL remains active between program calls.

The result of this strategy is that each call to the program retrieves the next row in the cursor. On
subsequent data requests, the OPEN statement is unnecessary and, in fact, fails with a -502 SQLCODE.
You can ignore the error, or add code to skip the OPEN. You can do this by using a FETCH statement
first, and then running the OPEN statement only if the FETCH operation failed.

This technique also applies to prepared statements. A program could first try the EXECUTE, and if it fails,
perform the PREPARE. The result is that the PREPARE would only be needed on the first call to the
program, assuming the correct CLOSQLCSR option was chosen. Of course, if the statement can change
between calls to the program, it should perform the PREPARE in all cases.

The main program could also control this by sending a special parameter on the first call only. This special
parameter value would indicate that because it is the first call, the subprogram should perform the OPENs,
PREPAREs, and LOCKs.

Note: If you are using COBOL programs, do not use the STOP RUN statement. When the first COBOL
program on the call stack ends or a STOP RUN statement runs, a reclaim resource (RCLRSC)
operation is done. This operation closes the SQL cursor. The *ENDSQL option does not work as
desired.

118 DB2 UDB for iSeries Database Performance and Query Optimization V5R1
Chapter 7. Programming techniques for database
performance
The following coding tips can help you improve the performance of your SQL queries:
v “Programming techniques for database performance: Use the OPTIMIZE clause”
v “Programming techniques for database performance: Use FETCH FOR n ROWS” on page 120
v “Programming techniques for database performance: Use INSERT n ROWS” on page 121
v “Programming techniques for database performance: Control database manager blocking” on page 121
v “Programming techniques for database performance: Optimize the number of columns that are selected
with SELECT statements” on page 122
v “Programming techniques for database performance: Eliminate redundant validation with SQL
PREPARE statements” on page 123
v “Programming techniques for database performance: Page interactively displayed data with
REFRESH(*FORWARD)” on page 123

Programming techniques for database performance: Use the OPTIMIZE

clause
If an application is not going to retrieve the entire result table for a cursor, using the OPTIMIZE clause can
improve performance. The query optimizer modifies the cost estimates to retrieve the subset of rows using
the value specified on the OPTIMIZE clause.

Assume that the following query returns 1000 rows:

EXEC SQL
DECLARE C1 CURSOR FOR
SELECT EMPNO, LASTNAME, WORKDEPT
FROM CORPDATA.EMPLOYEE
WHERE WORKDEPT = ’A00’
ORDER BY LASTNAME
OPTIMIZE FOR 100 ROWS
END EXEC.

Note: The values that can be used for the OPTIMIZE clause above are 1–9999999 or ALL.

The optimizer calculates the following costs.

The optimize ratio = optimize for n rows value / estimated number of rows in answer set.
Cost using a temporarily created index:

Cost to retrieve answer set rows

+ Cost to create the index
+ Cost to retrieve the rows again
with a temporary index * optimize ratio

Cost using a SORT:

Cost to retrieve answer set rows

+ Cost for SORT input processing
+ Cost for SORT output processing * optimize ratio

Cost using an existing index:

Cost to retrieve answer set rows

using an existing index * optimize ratio

© Copyright IBM Corp. 2000, 2001 119

In the previous examples, the estimated cost to sort or to create an index is not adjusted by the optimize
ratio. This enables the optimizer to balance the optimization and preprocessing requirements. If the
optimize number is larger than the number of rows in the result table, no adjustments are made to the cost
estimates. If the OPTIMIZE clause is not specified for a query, a default value is used based on the
statement type, value of ALWCPYDTA specified, or output device.

Statement Type ALWCPYDTA(*OPTIMIZE) ALWCPYDTA(*YES or *NO)

DECLARE CURSOR The number or rows in the result 3% or the number of rows in the result
table. table.
Embedded Select 2 2
INTERACTIVE Select output to 3% or the number of rows in the result 3% or the number of rows in the result
display table. table.
INTERACTIVE Select output to printer The number of rows in the result The number of rows in the result
or database table table. table.

The OPTIMIZE clause influences the optimization of a query:

v To use an existing index (by specifying a small number).
v To enable the creation of an index or to run a sort or a hash by specifying a large number of possible
rows in the answer set.

Programming techniques for database performance: Use FETCH FOR

n ROWS
Applications that perform many FETCH statements in succession may be improved by using FETCH FOR
n ROWS. With this clause, you can retrieve multiple rows of data from a table and put them into a host
structure array or row storage area with a single FETCH. For more information on declaring arrays of host
structures or row storage areas, see the SQL Reference book or the individual programming chapters in
the SQL Programming with Host Languages book.

An SQL application that uses a FETCH statement without the FOR n ROWS clause can be improved by
using the multiple-row FETCH statement to retrieve multiple rows. After the host structure array or row
storage area has been filled by the FETCH, the application can loop through the data in the array or
storage area to process each of the individual rows. The statement runs faster because the SQL run-time
was called only once and all the data was simultaneously returned to the application program.

You can change the application program to allow the database manager to block the rows that the SQL
run-time retrieves from the tables. For more information, see “Programming techniques for database
performance: Control database manager blocking” on page 121.

In the following table, the program attempted to FETCH 100 rows into the application. Note the differences
in the table for the number of calls to SQL run-time and the database manager when blocking can be
performed.
Table 11. Number of Calls Using a FETCH Statement
Database Manager Not Using Database Manager Using Blocking
Blocking
Single-Row FETCH Statement 100 SQL calls 100 database calls 100 SQL calls 1 database call
Multiple-Row FETCH Statement 1 SQL run-time call 100 database 1 SQL run-time call 1 database call
calls

120 DB2 UDB for iSeries Database Performance and Query Optimization V5R1
Programming techniques for database performance: Improve SQL
blocking performance when using FETCH FOR n ROWS
Special performance considerations should be made for the following points when using FETCH FOR n
ROWS. You can improve SQL blocking performance with the following:
v The attribute information in the host structure array or the descriptor associated with the row storage
area should match the attributes of the columns retrieved.
v The application should retrieve as many rows as possible with a single multiple-row FETCH call. The
blocking factor for a multiple-row FETCH request is not controlled by the system page sizes or the
SEQONLY parameter on the OVRDBF command. It is controlled by the number of rows that are
requested on the multiple-row FETCH request.
v Single- and multiple-row FETCH requests against the same cursor should not be mixed within a
program. If one FETCH against a cursor is treated as a multiple-row FETCH, all fetches against that
cursor are treated as multiple-row fetches. In that case, each of the single-row FETCH requests would
be treated as a multiple-row FETCH of one row.
v The PRIOR, CURRENT, and RELATIVE scroll options should not be used with multiple-row FETCH
statements. To allow random movement of the cursor by the application, the database manager must
maintain the same cursor position as the application. Therefore, the SQL run-time treats all FETCH
requests against a scrollable cursor with these options specified as multiple-row FETCH requests.

Programming techniques for database performance: Use INSERT n

ROWS
Applications that perform many INSERT statements in succession may be improved by using INSERT n
ROWS. With this clause, you can insert one or more rows of data from a host structure array into a target
table. This array must be an array of structures where the elements of the structure correspond to columns
in the target table.

An SQL application that loops over an INSERT...VALUES statement (without the n ROWS clause) can be
improved by using the INSERT n ROWS statement to insert multiple rows into the table. After the
application has looped to fill the host array with rows, a single INSERT n ROWS statement can be run to
insert the entire array into the table. The statement runs faster because the SQL run-time was only called
once and all the data was simultaneously inserted into the target table.

In the following table, the program attempted to INSERT 100 rows into a table. Note the differences in the
number of calls to SQL run-time and to the database manager when blocking can be performed.
Table 12. Number of Calls Using an INSERT Statement
Database Manager Not Using Database Manager Using Blocking
Blocking
Single-Row INSERT Statement 100 SQL run-time calls 100 database 100 SQL run-time calls 1 database
calls call
Multiple-Row INSERT Statement 1 SQL run-time call 100 database 1 SQL run-time call 1 database call
calls

Programming techniques for database performance: Control database

manager blocking
To improve performance, the SQL run-time attempts to retrieve and insert rows from the database
manager a block at a time whenever possible.

Chapter 7. Programming techniques for database performance 121

You can control blocking, if desired. Use the SEQONLY parameter on the CL command Override Database
File (OVRDBF) before calling the application program that contains the SQL statements. You can also
specify the ALWBLK parameter on the CRTSQLxxx commands.

The database manager does not allow blocking in the following situations:
v The cursor is update or delete capable.
v The length of the row plus the feedback information is greater than 32767. The minimum size for the
feedback information is 11 bytes. The feedback size is increased by the number of bytes in the key
columns for the index used by the cursor and by the number of key columns, if any, that are null
capable.
v COMMIT(*CS) is specified, and ALWBLK(*ALLREAD) is not specified.
v COMMIT(*ALL) is specified, and the following are true:
– A SELECT INTO statement or a blocked FETCH statement is not used
– The query does not use column functions or specify group by columns.
– A temporary result table does not have to be created.
v COMMIT(*CHG) is specified, and ALWBLK(*ALLREAD) is not specified.
v The cursor contains at least one subquery and the outermost subselect provided a correlated reference
for a subquery or the outermost subselect processed a subquery with an IN, = ANY, or < > ALL
subquery predicate operator, which is treated as a correlated reference, and that subquery is not
isolatable.

The SQL run-time automatically blocks rows with the database manager in the following cases:
v INSERT
If an INSERT statement contains a select-statement, inserted rows are blocked and not actually inserted
into the target table until the block is full. The SQL run-time automatically does blocking for blocked
inserts.

Note: If an INSERT with a VALUES clause is specified, the SQL run-time might not actually close the
internal cursor that is used to perform the inserts until the program ends. If the same INSERT
statement is run again, a full open is not necessary and the application runs much faster.
v OPEN
Blocking is done under the OPEN statement when the rows are retrieved if all of the following
conditions are true:
– The cursor is only used for FETCH statements.
– No EXECUTE or EXECUTE IMMEDIATE statements are in the program, or ALWBLK(*ALLREAD)
was specified, or the cursor is declared with the FOR FETCH ONLY clause.
– COMMIT(*CHG) and ALWBLK(*ALLREAD) are specified, COMMIT(*CS) and ALWBLK(*ALLREAD)
are specified, or COMMIT(*NONE) is specified.

Programming techniques for database performance: Optimize the

number of columns that are selected with SELECT statements
The number of columns that you specify in the select list of a SELECT statement causes the database
manager to retrieve the data from the underlying tables and map the data into host variables in the
application programs. By minimizing the number of columns that are specified, processing unit resource
usage can be conserved. Even though it is convenient to code SELECT *, it is far better to explicitly code
the columns that are actually required for the application. This is especially important if index-only access
is desired or if all of the columns will participate in a sort operation (as happens for SELECT DISTINCT
and for SELECT UNION).

122 DB2 UDB for iSeries Database Performance and Query Optimization V5R1
This is also important when considering index only access, since you minimize the number of columns in a
query and thereby increase the odds that an index can be used to completely satisfy the request for all the
data.

Programming techniques for database performance: Eliminate

redundant validation with SQL PREPARE statements
The processing which occurs when an SQL PREPARE statement is run is similar to the processing which
occurs during precompile processing. The following processing occurs for the statement that is being
prepared:
v The syntax is checked.
v The statement is validated to ensure that the usage of objects are valid.
v An access plan is built.
Again when the statement is executed or opened, the database manager will revalidate that the access
plan is still valid. Much of this open processing validation is redundant with the validation which occurred
during the PREPARE processing. The DLYPRP(*YES) parameter specifies whether PREPARE statements
in this program will completely validate the dynamic statement. The validation will be completed when the
dynamic statement is opened or executed. This parameter can provide a significant performance
enhancement for programs which use the PREPARE SQL statement because it eliminates redundant
validation. Programs that specify this precompile option should check the SQLCODE and SQLSTATE after
running the OPEN or EXECUTE statement to ensure that the statement is valid. DLYPRP(*YES) will not
provide any performance improvement if the INTO clause is used on the PREPARE statement or if a
DESCRIBE statement uses the dynamic statement before an OPEN is issued for the statement.

Programming techniques for database performance: Page interactively

displayed data with REFRESH(*FORWARD)
In large tables, paging performance is usually degraded because of the REFRESH(*ALWAYS) parameter
on the STRSQL command which dynamically retrieves the latest data directly from the table. Paging
performance can be improved by specifying REFRESH(*FORWARD).

When interactively displaying data using REFRESH(*FORWARD), the results of a select-statement are
copied to a temporary table as you page forward through the display. Other users sharing the table can
make changes to the rows while you are displaying the select-statement results. If you page backward or
forward to rows that have already been displayed, the rows shown are those in the temporary table
instead of those in the updated table.

The refresh option can be changed on the Session Services display.

Chapter 7. Programming techniques for database performance 123

124 DB2 UDB for iSeries Database Performance and Query Optimization V5R1
Chapter 8. General DB2 UDB for iSeries performance
considerations
As you code your applications, the following general tips can help you optimize performance:
v “Effects on database performance when using long object names”
v “Effects of precompile options on database performance”
v “Effects of the ALWCPYDTA parameter on database performance” on page 126
v “Tips for using VARCHAR and VARGRAPHIC data types in databases” on page 127

Effects on database performance when using long object names

Long object names are converted internally to system object names when used in SQL statements. This
conversion can have some performance impacts.

Qualify the long object name with a library name, and the conversion to the short name happens at
precompile time. In this case, there is no performance impact when the statement is executed. Otherwise,
the conversion is done at execution time, and has a small performance impact.

Effects of precompile options on database performance

Several precompile options are available for creating SQL programs with improved performance. They are
only options because using them may impact the function of the application. For this reason, the default
value for these parameters is the value that will ensure successful migration of applications from prior
releases. However, you can improve performance by specifying other options. The following table shows
these precompile options and their performance impacts.

Some of these options may be suitable for most of your applications. Use the command CRTDUPOBJ to
create a copy of the SQL CRTSQLxxx command. and the CHGCMDDFT command to customize the
optimal values for the precompile parameters. The DSPPGM, DSPSRVPGM, DSPMOD, or PRTSQLINF
commands can be used to show the precompile options that are used for an existing program object.

Precompile Option Optimal Value Improvements Considerations Related Topics

ALWCPYDTA *OPTIMIZE (the Queries where the A copy of the data See “Effects of the
default) ordering or grouping may be made when ALWCPYDTA
criteria conflicts with the query is opened. parameter on
the selection criteria. database
performance” on
page 126.
ALWBLK *ALLREAD (the Additional read-only ROLLBACK HOLD See “Programming
default) cursors use blocking. may not change the techniques for
position of a read-only database
cursor.Dynamic performance: Control
processing of database manager
positioned updates or blocking” on
deletes might fail. page 121.
CLOSQLCSR *ENDJOB, *ENDSQL, Cursor position can Implicit closing of SQL See “Database
or *ENDACTGRP be retained across cursor is not done application design
program invocations. when the program tips: Retaining cursor
invocation ends. positions for non-ILE
program calls” on
page 116.

© Copyright IBM Corp. 2000, 2001 125

Precompile Option Optimal Value Improvements Considerations Related Topics
DLYPRP *YES Programs using SQL Complete validation of See “Programming
PREPARE statements the prepared techniques for
may run faster. statement is delayed database
until the statement is performance:
run or opened. Eliminate redundant
validation with SQL
PREPARE
statements” on
page 123.
TGTRLS *CURRENT (the The precompiler can The program object See “Database
default) generate code that cannot be used on a application design
will take advantage of system from a tips: Use structure
performance previous release. parameter passing
enhancements techniques” on
available in the page 111.
current release.

Effects of the ALWCPYDTA parameter on database performance

Some complex queries can perform better by using a sort or hashing method to evaluate the query instead
of using or creating an index. By using the sort or hash, the database manager is able to separate the row
selection from the ordering and grouping process. Bitmap processing can also be partially controlled
through this parameter. This separation allows the use of the most efficient index for the selection. For
example, consider the following SQL statement:
EXEC SQL
DECLARE C1 CURSOR FOR
SELECT EMPNO, LASTNAME, WORKDEPT
FROM CORPDATA.EMPLOYEE
WHERE WORKDEPT = ’A00’
ORDER BY LASTNAME
END-EXEC.

The above SQL statement would be written in the following way by using the OPNQRYF command:
OPNQRYF FILE(CORPDATA/EMPLOYEE)
FORMAT(FORMAT1)
QRYSLT(WORKDEPT *EQ ’’AOO’’)
KEYFLD(LASTNAME)

In the above example when ALWCPYDTA(*NO) or ALWCPYDTA(*YES) is specified, the database

manager may try to create an index from the first index with a column named LASTNAME, if such an
index exists. The rows in the table are scanned, using the index, to select only the rows matching the
WHERE condition.

If ALWCPYDTA(*OPTIMIZE) is specified, the database manager uses an index with the first index column
of WORKDEPT. It then makes a copy of all of the rows that match the WHERE condition. Finally, it may
sort the copied rows by the values in LASTNAME. This row selection processing is significantly more
efficient, because the index used immediately locates the rows to be selected.

ALWCPYDTA(*OPTIMIZE) optimizes the total time that is required to process the query. However, the time
required to receive the first row may be increased because a copy of the data must be made prior to
returning the first row of the result table. This initial change in response time may be important for
applications that are presenting interactive displays or that retrieve only the first few rows of the query. The
DB2 Universal Database for iSeries query optimizer can be influenced to avoid sorting by using the
OPTIMIZE clause. Refer to “Programming techniques for database performance: Use the OPTIMIZE
clause” on page 119 for more information.

126 DB2 UDB for iSeries Database Performance and Query Optimization V5R1
Queries that involve a join operation may also benefit from ALWCPYDTA(*OPTIMIZE) because the join
order can be optimized regardless of the ORDER BY specification.

Tips for using VARCHAR and VARGRAPHIC data types in databases

Variable-length column (VARCHAR or VARGRAPHIC) support allows you to define any number of columns
in a table as variable length. If you use VARCHAR or VARGRAPHIC support, the size of a table can
usually be reduced.

Data in a variable-length column is stored internally in two areas: a fixed-length or ALLOCATE area and an
overflow area. If a default value is specified, the allocated length is at least as large as the value. The
following points help you determine the best way to use your storage area.

When you define a table with variable-length data, you must decide the width of the ALLOCATE area. If
the primary goal is:
v Space saving: use ALLOCATE(0).
v Performance: the ALLOCATE area should be wide enough to incorporate at least 90% to 95% of the
values for the column.

It is possible to balance space savings and performance. In the following example of an electronic phone
book, the following data is used:
v 8600 names that are identified by: last, first, and middle name
v The Last, First, and Middle columns are variable length.
v The shortest last name is 2 characters; the longest is 22 characters.

This example shows how space can be saved by using variable-length columns. The fixed-length column
table uses the most space. The table with the carefully calculated allocate sizes uses less disk space. The
table that was defined with no allocate size (with all of the data stored in the overflow area) uses the least
disk space.

Number of Rows
Variety of Last Name First Name Middle Name Total Physical in Overflow
Support Max/Alloc Max/Alloc Max/Alloc File Size Space
Fixed Length 22 22 22 567 K 0
Variable Length 40/10 40/10 40/7 408 K 73
Variable-Length 40/0 40/0 40/0 373 K 8600
Default

In many applications, performance must be considered. If you use the default ALLOCATE(0), it will double
the disk unit traffic. ALLOCATE(0) requires two reads; one to read the fixed-length portion of the row and
one to read the overflow space. The variable-length implementation, with the carefully chosen ALLOCATE,
minimizes overflow and space and maximizes performance. The size of the table is 28% smaller than the
fixed-length implementation. Because 1% of rows are in the overflow area, the access requiring two reads
is minimized. The variable-length implementation performs about the same as the fixed-length
implementation.

To create the table using the ALLOCATE keyword:

CREATE TABLE PHONEDIR
(LAST VARCHAR(40) ALLOCATE(10),
FIRST VARCHAR(40) ALLOCATE(10),
MIDDLE VARCHAR(40) ALLOCATE(7))

Chapter 8. General DB2 UDB for iSeries performance considerations 127

If you are using host variables to insert or update variable-length columns, the host variables should be
variable length. Because blanks are not truncated from fixed-length host variables, using fixed-length host
variables would cause more rows to spill into the overflow space. This would increase the size of the table.

In this example, fixed-length host variables are used to insert a row into a table:
01 LAST-NAME PIC X(40).
...
MOVE "SMITH" TO LAST-NAME.
EXEC SQL
INSERT INTO PHONEDIR
VALUES(:LAST-NAME, :FIRST-NAME, :MIDDLE-NAME, :PHONE)
END-EXEC.

The host-variable LAST-NAME is not variable length. The string “SMITH”, followed by 35 blanks, is
inserted into the VARCHAR column LAST. The value is longer than the allocate size of 10. Thirty of
thirty-five trailing blanks are in the overflow area.

In this example, variable-length host variables are used to insert a row into a table:
01 VLAST-NAME.
49 LAST-NAME-LEN PIC S9(4) BINARY.
49 LAST-NAME-DATA PIC X(40).
...
MOVE "SMITH" TO LAST-NAME-DATA.
MOVE 5 TO LAST-NAME-LEN.
EXEC SQL
INSERT INTO PHONEDIR
VALUES(:VLAST-NAME, :VFIRST-NAME, :VMIDDLE-NAME, :PHONE)
END-EXEC.

The host variable VLAST-NAME is variable length. The actual length of the data is set to 5. The value is
shorter than the allocated length. It can be placed in the fixed portion of the column.

For more information about using variable-length host variables, see the SQL Programming with Host
Languages book.

Running the RGZPFM command against tables that contain variable-length columns can improve
performance. The fragments in the overflow area that are not in use are compacted by the RGZPFM
command. This reduces the read time for rows that overflow, increases the locality of reference, and
produces optimal order for serial batch processing.

Choose the appropriate maximum length for variable-length columns. Selecting lengths that are too long
increases the process access group (PAG). A large PAG slows performance. A large maximum length
makes SEQONLY(*YES) less effective. Variable-length columns longer than 2000 bytes are not eligible as
key columns.

128 DB2 UDB for iSeries Database Performance and Query Optimization V5R1
Appendix A. Database monitor: DDS
This appendix contains the DDS that is used to create the database monitor physical and logical files:
v “Database monitor physical file DDS”
v “Optional database monitor logical file DDS” on page 136

Database monitor physical file DDS

The following figure shows the DDS that is used to create the QSYS/QAQQDBMN performance statistics
physical file.
|...+....1....+....2....+....3....+....4....+....5....+....6....+....7....+....8
A*
A* Database Monitor physical file row format
A*
A R QQQDBMN TEXT(’Database Monitor +
Base Table’)
A QQRID 15P TEXT(’Row +
ID’) +
EDTCDE(4) +
COLHDG(’Row’ ’ID’)
A QQTIME Z TEXT(’Time row was +
created’) +
COLHDG(’Time’ +
’Row’ +
’Created’)
A QQJFLD 46H TEXT(’Join Column’) +
COLHDG(’Join’ ’Column’)
A QQRDBN 18A TEXT(’Relational +
Database Name’) +
COLHDG(’Relational’ +
’Database’ ’Name’)
A QQSYS 8A TEXT(’System Name’) +
COLHDG(’System’ ’Name’)
A QQJOB 10A TEXT(’Job Name’) +
COLHDG(’Job’ ’Name’)
A QQUSER 10A TEXT(’Job User’) +
COLHDG(’Job’ ’User’)
A QQJNUM 6A TEXT(’Job Number’) +
COLHDG(’Job’ ’Number’)
A QQUCNT 15P TEXT(’Unique Counter’) +
ALWNULL +
COLHDG(’Unique’ ’Counter’)
A QQUDEF 100A VARLEN TEXT(’User Defined +
Column’) +
ALWNULL +
COLHDG(’User’ ’Defined’ +
’Column’)
A QQSTN 15P TEXT(’Statement Number’) +
ALWNULL +
COLHDG(’Statement’ +
’Number’)
A QQQDTN 15P TEXT(’Subselect Number’) +
ALWNULL +
COLHDG(’Subselect’ +
’Number’)
A QQQDTL 15P TEXT(’Nested level of +
subselect’) +
ALWNULL +
COLHDG(’Nested’ +
’Level of’ +
’Subselect’)
A QQMATN 15P TEXT(’Subselect of +

© Copyright IBM Corp. 2000, 2001 129

 materialized view’) +
ALWNULL +
COLHDG(’Subselect’ +
’Number of’ +
’Materialized View’)
A QQMATL 15P TEXT(’Nested level of +
Views subselect’) +
ALWNULL +
COLHDG(’Nested Level’ +
’of View’’s’ +
’Subselect’)
A QQTLN 10A TEXT(’Library of +
Table Queried’) +
ALWNULL +
COLHDG(’Library of’ +
’Table’ +
’Queried’)
A QQTFN 10A TEXT(’Name of +
Table Queried’) +
ALWNULL +
COLHDG(’Name of’ +
’Table’ +
’Queried’)
A QQTMN 10A TEXT(’Member of +
Table Queried’) +
ALWNULL +
COLHDG(’Member of’ +
’Table’ +
’Queried’)
A QQPTLN 10A TEXT(’Base Library’) +
ALWNULL +
COLHDG(’Library of’ +
’Base’ +
’Table’)
A QQPTFN 10A TEXT(’Base Table’) +
ALWNULL +
COLHDG(’Name of’ +
’Base’ +
’Table’)
A QQPTMN 10A TEXT(’Base Member’) +
ALWNULL +
COLHDG(’Member of’ +
’Base’ +
’Table’)
A QQILNM 10A TEXT(’Library of +
Index Used’) +
ALWNULL +
COLHDG(’Library of’ +
’Index’ +
’Used’)
A QQIFNM 10A TEXT(’Name of +
Index Used’) +
ALWNULL +
COLHDG(’Name of’ +
’Index’ +
’Used’)
A QQIMNM 10A TEXT(’Member of +
Index Used’) +
ALWNULL +
COLHDG(’Member of’ +
’Index’ +
’Used’)
A QQNTNM 10A TEXT(’NLSS Table’) +
ALWNULL +
COLHDG(’NLSS’ ’Table’)
A QQNLNM 10A TEXT(’NLSS Library’) +
ALWNULL +

130 DB2 UDB for iSeries Database Performance and Query Optimization V5R1
 COLHDG(’NLSS’ ’Library’)
A QQSTIM Z TEXT(’Start timestamp’) +
ALWNULL +
COLHDG(’Start’ ’Time’)
A QQETIM Z TEXT(’End timestamp’) +
ALWNULL +
COLHDG(’End’ ’Time’)
A QQKP 1A TEXT(’Index scan-key positioning’) +
ALWNULL +
COLHDG(’Key’ ’Positioning’)
A QQKS 1A TEXT(’Key selection’) +
ALWNULL +
COLHDG(’Key’ ’Selection’)
A QQTOTR 15P TEXT(’Total rows in table’) +
ALWNULL +
COLHDG(’Total’ +
’Rows in’ +
’Table’)
A QQTMPR 15P TEXT(’Number of rows in +
temporary’) +
ALWNULL +
COLHDG(’Number’ +
’of Rows’ +
’in Temporary’)
A QQJNP 15P TEXT(’Join Position’) +
ALWNULL +
COLHDG(’Join’ ’Position’)
A QQEPT 15P TEXT(’Estimated processing +
time’) +
ALWNULL +
COLHDG(’Estimated’ +
’Processing’ +
’Time’)
A QQDSS 1A TEXT(’Data space +
Selection’)
ALWNULL +
COLHDG(’Data’ ’Space’ +
’Selection’)
A QQIDXA 1A TEXT(’Index advised’) +
ALWNULL +
COLHDG(’Index’ ’Advised’)
A QQORDG 1A TEXT(’Ordering’) +
ALWNULL +
COLHDG(’Ordering’)
A QQGRPG 1A TEXT(’Grouping’) +
ALWNULL +
COLHDG(’Grouping’)
A QQJNG 1A TEXT(’Join’) +
ALWNULL +
COLHDG(’Join’)
A QQUNIN 1A TEXT(’Union’) +
ALWNULL +
COLHDG(’Union’)
A QQSUBQ 1A TEXT(’Subquery’) +
ALWNULL +
COLHDG(’Subquery’)
A QQHSTV 1A TEXT(’Host Variables’) +
ALWNULL +
COLHDG(’Host’ ’Variables’)
A QQRCDS 1A TEXT(’Row Selection’) +
ALWNULL +
COLHDG(’Row’ ’Selection’)
A QQRCOD 2A TEXT(’Reason Code’) +
ALWNULL +
COLHDG(’Reason’ ’Code’)
A QQRSS 15P TEXT(’Number of rows +
selected or sorted’) +

Appendix A. Database monitor: DDS 131

 ALWNULL +
COLHDG(’Number of’ +
’Rows’ +
’Selected’)
A QQREST 15P TEXT(’Estimated number +
of rows selected’) +
ALWNULL +
COLHDG(’Estimated’ +
’Rows’ +
’Selected’)
A QQRIDX 15P TEXT(’Number of entries +
in index created’) +
ALWNULL +
COLHDG(’Entries in’ +
’Index’ +
’Created’)
A QQFKEY 15P TEXT(’Estimated keys for +
index scan-key positioning’) +
ALWNULL +
COLHDG(’Estimated’ +
’Entries for’ +
’index scan-key positioning’)
A QQKSEL 15P TEXT(’Estimated keys for +
key selection’) +
ALWNULL +
COLHDG(’Estimated’ +
’Entries for’ +
’Key Selection’)
A QQAJN 15P TEXT(’Estimated number +
of joined rows’) +
ALWNULL +
COLHDG(’Estimated’ +
’Joined’ +
’Rows’)
A QQIDXD 1000A VARLEN(48) +
TEXT(’Columns +
for the index advised’) +
ALWNULL +
COLHDG(’Advised’ ’Key’ +
’Columns’)
A QQC11 1A ALWNULL
A QQC12 1A ALWNULL
A QQC13 1A ALWNULL
A QQC14 1A ALWNULL
A QQC15 1A ALWNULL
A QQC16 1A ALWNULL
A QQC18 1A ALWNULL
A QQC21 2A ALWNULL
A QQC22 2A ALWNULL
A QQC23 2A ALWNULL
A QQI1 15P ALWNULL
A QQI2 15P ALWNULL
A QQI3 15P ALWNULL
A QQI4 15P ALWNULL
A QQI5 15P ALWNULL
A QQI6 15P ALWNULL
A QQI7 15P ALWNULL
A QQI8 15P ALWNULL
A QQI9 15P TEXT(’Thread +
Identifier’) +
ALWNULL +
COLHDG(’Thread’ +
’Identifier’)
A QQIA 15P ALWNULL
A QQF1 15P ALWNULL
A QQF2 15P ALWNULL
A QQF3 15P ALWNULL

132 DB2 UDB for iSeries Database Performance and Query Optimization V5R1
A QQC61 6A ALWNULL
A QQC81 8A ALWNULL
A QQC82 8A ALWNULL
A QQC83 8A ALWNULL
A QQC84 8A ALWNULL
A QQC101 10A ALWNULL
A QQC102 10A ALWNULL
A QQC103 10A ALWNULL
A QQC104 10A ALWNULL
A QQC105 10A ALWNULL
A QQC106 10A ALWNULL
A QQC181 18A ALWNULL
A QQC182 18A ALWNULL
A QQC183 18A ALWNULL
A QQC301 30A VARLEN(10) ALWNULL
A QQC302 30A VARLEN(10) ALWNULL
A QQC303 30A VARLEN(10) ALWNULL
A QQ1000 1000A VARLEN(48) ALWNULL
A QQTIM1 Z ALWNULL
A QQTIM2 Z ALWNULL
A*
A* New columns added for Visual Explain
A*
A QVQTBL 128A VARLEN(10) +
TEXT(’Queried Table, +
Long Name’) +
ALWNULL +
COLHDG(’Queried’ +
’Table’ +
’Long Name’)
A QVQLIB 128A VARLEN(10) +
TEXT(’Queried Library, +
Long Name’) +
ALWNULL +
COLHDG(’Queried’ +
’Library’ +
’Long Name’)
A QVPTBL 128A VARLEN(10) +
TEXT(’Base Table, +
Long Name’) +
ALWNULL +
COLHDG(’Base’ +
’Table’ +
’Long Name’)
A QVPLIB 128A VARLEN(10) +
TEXT(’Base Library, +
Long Name’) +
ALWNULL +
COLHDG(’Base’ +
’Library’ +
’Long Name’)
A QVINAM 128A VARLEN(10) +
TEXT(’Index Used, +
Long Name’) +
ALWNULL +
COLHDG(’Index’ +
’Used’ +
’Long Name’)
A QVILIB 128A VARLEN(10) +
TEXT(’Index Used, +
Libary Name’) +
ALWNULL +
COLHDG(’Index’ +
’Used’ +
’Library’ +
’Name’)
A QVQTBLI 1A TEXT(’Table Long +

Appendix A. Database monitor: DDS 133

 Required’)
ALWNULL +
COLHDG(’Table’ +
’Long’ +
’Required’)
A QVPTBLI 1A TEXT(’Base Long +
Required’)
ALWNULL +
COLHDG(’Base’ +
’Long’ +
’Required’)
A QVINAMI 1A TEXT(’Index Long +
Required’)
ALWNULL +
COLHDG(’Index’ +
’Long’ +
’Required’)
A QVBNDY 1A TEXT(’I/O or CPU +
Bound’) +
ALWNULL +
COLHDG(’I/O or CPU’ +
’Bound’)
A QVJFANO 1A TEXT(’Join +
Fan out’) +
ALWNULL +
COLHDG(’Join’ +
’Fan’ +
’Out’)
A QVPARPF 1A TEXT(’Parallel +
Pre-Fetch’) +
ALWNULL +
COLHDG(’Parallel’ +
’Pre-Fetch’)
A QVPARPL 1A TEXT(’Parallel +
Preload’) +
ALWNULL +
COLHDG(’Parallel’ +
’Preload’)
A QVC11 1A ALWNULL
A QVC12 1A ALWNULL
A QVC13 1A ALWNULL
A QVC14 1A ALWNULL
A QVC15 1A ALWNULL
A QVC16 1A ALWNULL
A QVC17 1A ALWNULL
A QVC18 1A ALWNULL
A QVC19 1A ALWNULL
A QVC1A 1A ALWNULL
A QVC1B 1A ALWNULL
A QVC1C 1A ALWNULL
A QVC1D 1A ALWNULL
A QVC1E 1A ALWNULL
A QVC1F 1A ALWNULL
A QWC11 1A ALWNULL
A QWC12 1A ALWNULL
A QWC13 1A ALWNULL
A QWC14 1A ALWNULL
A QWC15 1A ALWNULL
A QWC16 1A ALWNULL
A QWC17 1A ALWNULL
A QWC18 1A ALWNULL
A QWC19 1A ALWNULL
A QWC1A 1A ALWNULL
A QWC1B 1A ALWNULL
A QWC1C 1A ALWNULL
A QWC1D 1A ALWNULL
A QWC1E 1A ALWNULL

134 DB2 UDB for iSeries Database Performance and Query Optimization V5R1
A QWC1F 1A ALWNULL
A QVC21 2A ALWNULL
A QVC22 2A ALWNULL
A QVC23 2A ALWNULL
A QVC24 2A ALWNULL
A QVCTIM 15P TEXT(’Cumulative +
Time’) +
ALWNULL +
COLHDG(’Estimated’ +
’Cumulative’ +
’Time’)
A QVPARD 15P TEXT(’Parallel Degree, +
Requested’) +
ALWNULL +
COLHDG(’Parallel’ +
’Degree’ +
’Requested’)
A QVPARU 15P TEXT(’Parallel Degree, +
Used’) +
ALWNULL +
COLHDG(’Parallel’ +
’Degree’ +
’Used’)
A QVPARRC 15P TEXT(’Parallel Limited, +
Reason Code’) +
ALWNULL +
COLHDG(’Parallel’ +
’Limited’ +
’Reason Code’)
A QVRCNT 15P TEXT(’Refresh Count’) +
ALWNULL +
COLHDG(’Refresh’ +
’Count’)
A QVFILES 15P TEXT(’Number of, +
Tables Joined’)
ALWNULL +
COLHDG(’Number of’ +
’Tables’ +
’Joined’)
A QVP151 15P ALWNULL
A QVP152 15P ALWNULL
A QVP153 15P ALWNULL
A QVP154 15P ALWNULL
A QVP155 15P ALWNULL
A QVP156 15P ALWNULL
A QVP157 15P ALWNULL
A QVP158 15P ALWNULL
A QVP159 15P ALWNULL
A QVP15A 15P TEXT(’Decomposed’ +
’Subselect Number’) +
ALWNULL +
COLHDG(’Decomposed’
’Subselect’ +
’Number’)
A QVP15B 15P TEXT(’Number of’ +
’Decomposed + Subselects’) +
ALWNULL +
COLHDG(’Number of’ +
’Decomposed’ + Subselects’)
A QVP15C 15P TEXT(’Decomposed’ +
’Subselect + Reason code’) +
ALWNULL +
COLHDG(’Decomposed’ +
’Reason’ +
’Code’)
A QVP15D 15P TEXT(’Number of first’ +
’Decomposed + Subselect’) +

Appendix A. Database monitor: DDS 135

 ALWNULL +
COLHDG(’Starting’ + ’Decomposed’ +
’Subselect’
A QVP15E 15P ALWNULL
A QVP15F 15P ALWNULL
A QVC41 4A ALWNULL
A QVC42 4A ALWNULL
A QVC43 4A ALWNULL
A QVC44 4A ALWNULL
A QVC81 8A ALWNULL
A QVC82 8A ALWNULL
A QVC83 8A ALWNULL
A QVC84 8A ALWNULL
A QVC85 8A ALWNULL
A QVC86 8A ALWNULL
A QVC87 8A ALWNULL
A QVC88 8A ALWNULL
A QVC101 10A ALWNULL
A QVC102 10A ALWNULL
A QVC103 10A ALWNULL
A QVC104 10A ALWNULL
A QVC105 10A ALWNULL
A QVC106 10A ALWNULL
A QVC107 10A ALWNULL
A QVC108 10A ALWNULL
A QVC1281 128A VARLEN(10) ALWNULL
A QVC1282 128A VARLEN(10) ALWNULL
A QVC1283 128A VARLEN(10) ALWNULL
A QVC1284 128A VARLEN(10) ALWNULL
A QVC3001 300A VARLEN(32) ALWNULL
A QVC3002 300A VARLEN(32) ALWNULL
A QVC3003 300A VARLEN(32) ALWNULL
A QVC3004 300A VARLEN(32) ALWNULL
A QVC3005 300A VARLEN(32) ALWNULL
A QVC3006 300A VARLEN(32) ALWNULL
A QVC3007 300A VARLEN(32) ALWNULL
A QVC3008 300A VARLEN(32) ALWNULL
A QVC5001 500A VARLEN(32) ALWNULL
A QVC5002 500A VARLEN(32) ALWNULL
A QVC1000 1000A VARLEN(48) ALWNULL
A QWC1000 1000A VARLEN(48) ALWNULL

Optional database monitor logical file DDS

| The following examples show the different optional logical files that you can create with the DDS shown.
| The column descriptions are explained in the tables following each example. These tables are not shipped
| with the system, and you must create them, if you choose to do so. These files are optional and are not
| required for analyzing monitor data.

Database monitor logical table 1000 - Summary Row for SQL

Information
|...+....1....+....2....+....3....+....4....+....5....+....6....+....7....+....8
A*
A*
A* DB Monitor logical table 1000 - Summary Row for SQL Information
A*
A R QQQ1000 PTABLE(*CURLIB/QAQQDBMN)
A QQRID
A QQTIME
A QQJFLD
A QQRDBN
A QQSYS
A QQJOB
A QQUSER

136 DB2 UDB for iSeries Database Performance and Query Optimization V5R1
A QQJNUM
A QQTHRD RENAME(QQI9) +
COLHDG(’Thread’ +
’Identifier’)
A QQUCNT
A QQRCNT RENAME(QQI5) +
COLHDG(’Refresh’ +
’Counter’)
A QQUDEF
A*
A* Information about the SQL statement executed
A*
A QQSTN
A QQSTF RENAME(QQC11) +
COLHDG(’Statement’ +
’Function’)
A QQSTOP RENAME(QQC21) +
COLHDG(’Statement’ +
’Operation’)
A QQSTTY RENAME(QQC12) +
COLHDG(’Statement’ ’Type’)
A QQPARS RENAME(QQC13) +
COLHDG(’Parse’ ’Required’)
A QQPNAM RENAME(QQC103) +
COLHDG(’Package’ ’Name’)
A QQPLIB RENAME(QQC104) +
COLHDG(’Package’ ’Library’)
A QQCNAM RENAME(QQC181) +
COLHDG(’Cursor’ ’Name’)
A QQSNAM RENAME(QQC182) +
COLHDG(’Statement’ ’Name’)
A QQSTIM
A QQSTTX RENAME(QQ1000) +
COLHDG(’Statement’ ’Text’)
A QQSTOC RENAME(QQC14) +
COLHDG(’Statement’ +
’Outcome’)
A QQROWR RENAME(QQI2) +
COLHDG(’Rows’ ’Returned’)
A QQDYNR RENAME(QQC22) +
COLHDG(’Dynamic’ ’Replan’)
A QQDACV RENAME(QQC16) +
COLHDG(’Data’ ’Conversion’)
A QQTTIM RENAME(QQI4) +
COLHDG(’Total’ ’Time’ +
’Milliseconds’)
A QQROWF RENAME(QQI3) +
COLHDG(’Rows’ ’Fetched’)
A QQETIM
A QQTTIMM RENAME(QQI6) +
COLHDG(’Total’ ’Time’)
’Microseconds’)
A QQSTMTLN RENAME(QQI7) +
COLHDG(’Total’ +
’Statement’ +
’Length’)
A QQIUCNT RENAME(QQI1) +
COLHDG(’Insert’ ’Unique’)
’Count’) A*
A QQADDTXT RENAME(QWC14) +
COLHDG(’Additional’ ’SQL’)
’Text’)
A* Columns added for Visual Explain
A*
A*
A* Additional information about the SQL statement executed
A*

Appendix A. Database monitor: DDS 137

 A QVSQCOD RENAME(QQI8) +
COLHDG(’SQL’ +
’Return’ +
’Code’)
A QVSQST RENAME(QQC81) +
COLHDG(’SQLSTATE’)
A QVCLSCR RENAME(QVC101) +
COLHDG(’CLOSQLCSR’ +
’Setting’)
A QVALWCY RENAME(QVC11) +
COLHDG(’ALWCPYDTA’ +
’Setting’)
A QVPSUDO RENAME(QVC12) +
COLHDG(’Pseudo’ +
’Open’)
A QVPSUDC RENAME(QVC13) +
COLHDG(’Pseudo’ +
’Close’)
A QVODPI RENAME(QVC14) +
COLHDG(’ODP’ +
’Implementation’)
A QVDYNSC RENAME(QVC21) +
COLHDG(’Dynamic’ +
’Replan’ +
’Subtype Code’)
A QVCMMT RENAME(QVC41) +
COLHDG(’Commit’ +
’Level’)
A QVBLKE RENAME(QVC15) +
COLHDG(’Blocking’ +
’Enabled’)
A QVDLYPR RENAME(QVC16) +
COLHDG(’Delay’ +
’Prep’)
A QVEXPLF RENAME(QVC1C) +
COLHDG(’SQL’ +
’Statement’ +
’Explainable’)
A QVNAMC RENAME(QVC17) +
COLHDG(’Naming’ +
’Convention’)
A QVDYNTY RENAME(QVC18) +
COLHDG(’Type of’ +
’Dynamic’ +
’Processing’)
A QVOLOB RENAME(QVC19) +
COLHDG(’Optimize’ +
’LOB’ +
’Data Types’)
A QVUSRP RENAME(QVC1A) +
COLHDG(’User’ +
’Profile’)
A QVDUSRP RENAME(QVC1B) +
COLHDG(’Dynamic’ +
’User’ +
’Profile’)
A QVDFTCL RENAME(QVC1281) +
COLHDG(’Default’ +
’Collection’)
A QVPROCN RENAME(QVC1282) +
COLHDG(’Procedure’ +
’Name on’ +
’CALL’)
A QVPROCL RENAME(QVC1283) +
COLHDG(’Procedure’ +
’Library on’ +
’CALL’)

138 DB2 UDB for iSeries Database Performance and Query Optimization V5R1
A QVSPATH RENAME(QVC1000) +
COLHDG(’SQL’ +
’Path’)
A QVSPATHB RENAME(QwC1000) +
COLHDG(’SQL’ +
’Path’ +
’Continued’)
A QVSPATHC RENAME(QVC5001) +
COLHDG(’SQL’ +
’Path’ +
’Continued’)
A QVSPATHD RENAME(QVC5002) +
COLHDG(’SQL’ +
’Path’ +
’Continued’)
A QVSPATHE RENAME(QVC3001) +
COLHDG(’SQL’ +
’Path’ +
’Continued’)
A QVSPATHF RENAME(QVC3002) +
COLHDG(’SQL’ +
’Path’ +
’Continued’)
A QVSPATHG RENAME(QVC30013) +
COLHDG(’SQL’ +
’Path’ +
’Continued’)
A*
A* Environmental information about the SQL statement executed
A*
A QVDFMT RENAME(QVC42) +
COLHDG(’Date’ +
’Format’)
A QVDSEP RENAME(QWC11) +
COLHDG(’Date’ +
’Separator’)
A QVTFMT RENAME(QVC43) +
COLHDG(’Time’ +
’Format’)
A QVTSEP RENAME(QWC12) +
COLHDG(’Time’ +
’Separator’)
A QVDPNT RENAME(QWC13) +
COLHDG(’Decimal’ +
’Point’)
A QVSRTSQ RENAME(QVC104) +
COLHDG(’Sort’ +
’Sequence’ +
’Table’)
A QVSRTSL RENAME(QVC105) +
COLHDG(’Sort’ +
’Sequence’ +
’Library’)
A QVLNGID RENAME(QVC44) +
COLHDG(’Language’ +
’ID’)
A QVCNTID RENAME(QVC23) +
COLHDG(’Country’ +
’ID’)
A K QQJFLD
A S QQRID CMP(EQ 1000)

Appendix A. Database monitor: DDS 139

Table 13. QQQ1000 - Summary row for SQL Information
Logical Column Physical Column
Name Name Description
QQRID QQRID Row identification
QQTIME QQTIME Time row was created
QQJFLD QQJFLD Join column (unique per job)
QQRDBN QQRDBN Relational database name
QQSYS QQSYS System name
QQJOB QQJOB Job name
QQUSER QQUSER Job user
QQJNUM QQJNUM Job number
QQTHRD QQI9 Thread identifier
QQUCNT QQUCNT Unique count (unique per query)
QQRCNT QQI5 Unique refresh counter
QQUDEF QQUDEF User defined column
QQSTN QQSTN Statement number (unique per statement)
QQSTF QQC11 Statement function:
v S - Select
v U - Update
v I - Insert
v D - Delete
v L - Data definition language
v O - Other

140 DB2 UDB for iSeries Database Performance and Query Optimization V5R1
Table 13. QQQ1000 - Summary row for SQL Information (continued)
Logical Column Physical Column
Name Name Description
QQSTOP QQC21 Statement operation:
v AL - Alter table
v CA - Call
v CC - Create collection
v CD - Create type
v CF - Create function
v CI - Create index
v CL - Close
v CM - Commit
v CN - Connect
v CO - Comment on
v CP - Create procedure
v CS - Create alias/synonym
v CT - Create table
v CV - Create view
v DE - Describe
v DI - Disconnect
v DL - Delete
v DM - Describe parameter marker
v DP - Declare procedure
v DR - Drop
v DT - Describe table
v EI - Execute immediate
v EX - Execute
v FE - Fetch
v FL - Free locator
v GR - Grant
v HC - Hard close
v IN - Insert
v JR - Server job reused
v LK - Lock
v LO - Label on
v MT - More text
v OP - Open
v PD - Prepare and describe
v PR - Prepare
v RE - Release
v RO - Rollback
v RT - Rename table
v RV - Revoke
v SC - Set connection

Appendix A. Database monitor: DDS 141

Table 13. QQQ1000 - Summary row for SQL Information (continued)
Logical Column Physical Column
Name Name Description
QQSTOP (continued) QQC21 v SI - Select into
v SP - Set path
v SR - Set result set
v ST - Set transaction
v SV - Set variable
v UP - Update
v VI - Values into
QQSTTY QQC12 Statement type:
v D - Dynamic statement
v S - Static statement
QQPARS QQC13 Parse required (Y/N)
QQPNAM QQC103 Name of the package or name of the program that contains the
current SQL statement
QQPLIB QQC104 Name of the library containing the package
QQCNAM QQC181 Name of the cursor corresponding to this SQL statement, if
applicable
QQSNAM QQC182 Name of statement for SQL statement, if applicable
QQSTIM QQSTIM Time this statement entered
QQSTTX QQ1000 Statement text
QQSTOC QQC14 Statement outcome
v S - Successful
v U - Unsuccessful

QQROWR QQI2 Number of result rows returned

142 DB2 UDB for iSeries Database Performance and Query Optimization V5R1
Table 13. QQQ1000 - Summary row for SQL Information (continued)
Logical Column Physical Column
Name Name Description
QQDYNR QQC22 Dynamic replan (access plan rebuilt)
v NA - No replan.
v NR - SQL QDT rebuilt for new release.
v A1 - A table or member is not the same object as the one
referenced when the access plan was last built. Some reasons
why they could be different are:
– Object was deleted and recreated.
– Object was saved and restored.
– Library list was changed.
– Object was renamed.
– Object was moved.
– Object was overridden to a different object.
– This is the first run of thisquery after the object containing the
query has been restored.
v A2 - Access plan was built to use a reusable Open Data Path
(ODP) and the optimizer chose to use a non-reusable ODP for this
call.
v A3 - Access plan was built to use a non-reusable Open Data Path
(ODP) and the optimizer chose to use a reusable ODP for this
call.
v A4 - The number of rows in the table member has changed by
more than 10% since the access plan was last built.
v A5 - A new index exists over one of the tables in the query.
v A6 - An index that was used for this access plan no longer exists
or is no longer valid.
v A7 - OS/400 Query requires the access plan to be rebuilt because
of system programming changes.
v A8 - The CCSID of the current job is different than the CCSID of
the job that last created the access plan.
v A9 - The value of one or more of the following is different for the
current job than it was for the job that last created this access
plan:
– date format
– date separator
– time format
– time separator
v AA - The sort sequence table specified is different than the sort
sequence table that was used when this access plan was created.
v AB - Storage pool changed or DEGREE parameter of CHGQRYA
command changed.
v AC - The system feature DB2 multisystem has been installed or
removed.
v AD - The value of the degree query attribute has changed.
v AE - A view is either being opened by a high level language or a
view is being materialized.
v AF - A user-defined type or user-defined function is not the same
object as the one referred to in the access plan.

Appendix A. Database monitor: DDS 143

Table 13. QQQ1000 - Summary row for SQL Information (continued)
Logical Column Physical Column
Name Name Description
QQDYNR (continued) QQC22 v B0 - The options specified have changed as a result of the query
options file.
v B1 - The access plan was generated with a commitment control
level that is different in the current job.
v B2 - The access plan was generated with a static cursor answer
set size that is different than the previous access plan.
QQDACV QQC16 Data conversion
v N - No.
v 0 - Not applicable.
v 1 - Lengths do not match.
v 2 - Numeric types do not match.
v 3 - C host variable is NUL-terminated.
v 4 - Host variable or column is variable length and the other is not
variable length.
v 5 - CCSID conversion.
v 6 - DRDA and NULL capable, variable length, contained in a
partial row, derived expression, or blocked fetch with not enough
host variables.
v 7 - Data, time, or timestamp column.
v 8 - Too many host variables.
v 9 - Target table of an insert is not an SQL table.
QQTTIM QQI4 Total time for this statement, in milliseconds. For fetches, this
includes all fetches for this OPEN of the cursor.
QQROWF QQI3 Total rows fetched for cursor
QQETIM QQETIM Time SQL request completed
QQTTIMM QQI6 Total time for this statement, in microseconds. For fetches, this
includes all fetches for this OPEN of the cursor.
QQSTMTLN QQI7 Length of SQL Statement
QQIUCNT QQI1 Unique query count for the QDT associated with the INSERT.
QQUCNT contains the unique query count for the QDT associated
with the WHERE part of the statement.
QVSQCOD QQI8 SQL return code
QVSQST QQC81 SQLSTATE
QVCLSCR QVC101 Close Cursor. Possible values are:
v *ENDJOB - SQL cursors are closed when the job ends.
v *ENDMOD - SQL cursors are closed when the module ends
v *ENDPGM - SQL cursors are closed when the program ends.
v *ENDSQL - SQL cursors are closed when the first SQL program
on the call stack ends.
v *ENDACTGRP - SQL cursors are closed when the activation
group ends.
QVALWCY QVC11 ALWCPYDTA setting (Y/N/O)
v Y - A copy of the data may be used.
v N - Cannot use a copy of the data.
v O - The optimizer can choose to use a copy of the data for
performance.

144 DB2 UDB for iSeries Database Performance and Query Optimization V5R1
Table 13. QQQ1000 - Summary row for SQL Information (continued)
Logical Column Physical Column
Name Name Description
QVPSUDO QVC12 Pseudo Open (Y/N) for SQL operations that can trigger opens.
v OP - Open
v IN - Insert
v UP - Update
v DL - Delete
v SI - Select Into
v SV - Set
v VI - Values into
For all operations it can be blank.
QVPSUDC QVC13 Pseudo Close (Y/N) for SQL operations that can trigger a close.
v CL - Close
v IN - Insert
v UP - Update
v DL - Delete
v SI - Select Into
v SV - Set
v VI - Values into
For all operations it can be blank.
QVODPI QVC14 ODP implementation
v R - Reusable ODP
v N - Nonreusable ODP
v ’ ’ - Column not used
QQDYNSC QVC21 Dynamic replan, subtype reason code
QVCMMT QVC41 Commitment control level. Possible values are:
v NC
v UR
v CS
v CSKL
v RS
v RR
QVBLKE QVC15 Type of blocking . Possible value are:
v S - Single row, ALWBLK(*READ)
v F - Force one row, ALWBLK(*NONE)
v L - Limited block, ALWBLK(*ALLREAD)
QVDLYPR QVC16 Delay Prep (Y/N)
QVEXPLF QVC1C The SQL statement is explainable (Y/N).
QVNAMC QVC17 Naming convention. Possibles values:
v N - System naming convention
v S - SQL naming convention
QVDYNTY QVC18 Type of dynamic processing.
v E - Extended dynamic
v S - System wide cache
v L - Local prepared statement

Appendix A. Database monitor: DDS 145

Table 13. QQQ1000 - Summary row for SQL Information (continued)
Logical Column Physical Column
Name Name Description
QVOLOB QVC19 Optimize LOB data types (Y/N)
QVUSRP QVC1A User profile used when compiled programs are executed. Possible
values are:
v N = User Profile is determined by naming conventions. For *SQL,
USRPRF(*OWNER) is used. For *SYS, USRPRF(*USER) is used.
v U = USRPRF(*USER) is used.
v O = USRPRF(*OWNER) is used.
QVDUSRP QVC1B User profile used for dynamic SQL statements.
v U = USRPRF(*USER) is used.
v O = USRPRF(*OWNER) is used.
QVDFTCL QVC1281 Name of the default collection.
QVPROCN QVC1282 Procedure name on CALL to SQL.
QVPROCL QVC1283 Procedure library on CALL to SQL.
QVSPATH QVC1000 Path used to find procedures, functions, and user defined types for
static SQL statements.
QVSPATHB QWC1000 Continuation of SQL path, if needed. Contains bytes 1001-2000 of
the SQL path.
QVSPATHC QWC5001 Continuation of SQL path, if needed. Contains bytes 2001-2500 of
the SQL path.
QVSPATHD QWC5002 Continuation of SQL path, if needed. Contains bytes 2501-3000 of
the SQL path.
QVSPATHE QWC3001 Continuation of SQL path, if needed. Contains bytes 3001-3300 of
the SQL path.
QVSPATHF QWC3002 Continuation of SQL path, if needed. Contains bytes 3301-3600 of
the SQL path.
QVSPATHG QWC3003 Continuation of SQL path, if needed. Contains bytes 3601-3900 of
the SQL path.
QVDFMT QVC42 Date Format. Possible values are:
v ISO
v USA
v EUR
v JIS
v MDY
v DMY
v YMD
QVDSEP QWC11 Date Separator. Possible values are:
v ″/″
v ″.″
v ″,″
v ″-″
v ″″

146 DB2 UDB for iSeries Database Performance and Query Optimization V5R1
Table 13. QQQ1000 - Summary row for SQL Information (continued)
Logical Column Physical Column
Name Name Description
QVTFMT QVC43 Time Format. Possible values are:
v ISO
v USA
v EUR
v JIS
v HMS
QVTSEP QWC12 Time Separator. Possible values are:
v ″:″
v ″.″
v ″,″
v ″″
QVDPNT QWC13 Decimal Point. Possible values are:
v ″.″
v ″,″
QVSRTSQ QVC104 Sort Sequence Table
QVSRTSL QVC105 Sort Sequence Library
QVLNGID QVC44 Language ID
QVCNTID QVC23 Country ID

Database monitor logical table 3000 - Summary Row for Table Scan
|...+....1....+....2....+....3....+....4....+....5....+....6....+....7....+....8
A*
A* DB Monitor logical table 3000 - Summary Row for Table Scan
A*
A R QQQ3000 PTABLE(*CURLIB/QAQQDBMN)
A QQRID
A QQTIME
A QQJFLD
A QQRDBN
A QQSYS
A QQJOB
A QQUSER
A QQJNUM
A QQTHRD RENAME(QQI9) +
COLHDG(’Thread’ +
’Identifier’)
A QQUCNT
A QQUDEF
A QQQDTN
A QQQDTL
A QQMATN
A QQMATL
A QDQDTN RENAME(QVP15A) +
COLHDG(’Decomposed’ +
’Subselect’ +
’Number’)
A QDQDTT RENAME(QVP15B) +
COLHDG(’Number of’ +
’Decomposed’ +
’Subselects’)
A QDQDTR RENAME(QVP15C) +
COLHDG(’Decomposed’ +

Appendix A. Database monitor: DDS 147

 ’Reason’ +
’Code’)
A QDQDTS RENAME(QVP15D) +
COLHDG(’Starting’ +
’Decomposed’ +
’Subselect’)
A QQTLN
A QQTFN
A QQTMN
A QQPTLN
A QQPTFN
A QQPTMN
A QQTOTR
A QQREST
A QQAJN
A QQEPT
A QQJNP
A QQJNDS RENAME(QQI1) +
COLHDG(’Data Space’ +
’Number’)
A QQJNMT RENAME(QQC21) +
COLHDG(’Join’ ’Method’)
A QQJNTY RENAME(QQC22) +
COLHDG(’Join’ ’Type’)
A QQJNOP RENAME(QQC23) +
COLHDG(’Join’ ’Operator’)
A QQIDXK RENAME(QQI2) +
COLHDG(’Advised’ +
’Primary’ +
’Keys’)
A QQDSS
A QQIDXA
A QQRCOD
A QQIDXD
A*
A* Columns added for Visual Explain
A*
A QVQTBL
A QVQLIB
A QVPTBL
A QVPLIB
A QVBNDY
A QVRCNT
A QVJFANO
A QVFILES
A QVPARPF
A QVPARPL
A QVPARD
A QVPARU
A QVPARRC
A QVCTIM
A QVSKIPS RENAME(QQC11) +
COLHDG(’Skip’ +
’Sequential’)
A QVTBLSZ RENAME(QQI3) +
COLHDG(’Actual’ +
’Table’ ’Size’)
A QVTSFLDS RENAME(QVC3001) +
COLHDG(’Columns for’ +
’Data Space’ +
’Selection’)
A QVDVFLD RENAME(QQC14) +
COLHDG(’Derived’ +
’Column’ +
’Selection’)
A QVDVFLDS RENAME(QVC3002) +
COLHDG(’Columns for’ +

148 DB2 UDB for iSeries Database Performance and Query Optimization V5R1
 ’Derived’ +
’Selection’)
A K QQJFLD
A S QQRID CMP(EQ 3000)

Table 14. QQQ3000 - Summary Row for Table Scan

Logical Column Physical Column
Name Name Description
QQRID QQRID Row identification
QQTIME QQTIME Time row was created
QQJFLD QQJFLD Join column (unique per job)
QQRDBN QQRDBN Relational database name
QQSYS QQSYS System name
QQJOB QQJOB Job name
QQUSER QQUSER Job user
QQJNUM QQJNUM Job number
QQTHRD QQI9 Thread identifier
QQUCNT QQUCNT Unique count (unique per query)
QQUDEF QQUDEF User defined column
QQQDTN QQQDTN Unique subselect number
QQQDTL QQQDTL Subselect nested level
QQMATN QQMATN Materialized view subselect number
QQMATL QQMATL Materialized view nested level
QDQDTN QVP15A Decomposed query subselect number, unique across all
decomposed subselects
QDQDTT QVP15B Total number of decomposed subselects
QDQDTR QVP15C Decomposed query subselect reason code
QDQDTS QVP15D Decomposed query subselect number for the first decomposed
subselect
QQTLN QQTLN Library of table queried
QQTFN QQTFN Name of table queried
QQTMN QQTMN Member name of table queried
QQPTLN QQPTLN Library name of base table
QQPTFN QQPTFN Name of base table for table queried
QQPTMN QQPTMN Member name of base table
QQTOTR QQTOTR Total rows in table
QQREST QQREST Estimated number of rows selected
QQAJN QQAJN Estimated number of joined rows
QQEPT QQEPT Estimated processing time, in seconds
QQJNP QQJNP Join position - when available
QQJNDS QQI1 dataspace number
QQJNMT QQC21 Join method - when available
v NL - Nested loop
v MF - Nested loop with selection
v HJ - Hash join

Appendix A. Database monitor: DDS 149

Table 14. QQQ3000 - Summary Row for Table Scan (continued)
Logical Column Physical Column
Name Name Description
QQJNTY QQC22 Join type - when available
v IN - Inner join
v PO - Left partial outer join
v EX - Exception join
QQJNOP QQC23 Join operator - when available
v EQ - Equal
v NE - Not equal
v GT - Greater than
v GE - Greater than or equal
v LT - Less than
v LE - Less than or equal
v CP - Cartesian product
QQIDXK QQI2 Number of advised columns that use index scan-key positioning
QQDSS QQDSS dataspace selection
v Y - Yes
v N - No
QQIDXA QQIDXA Index advised
v Y - Yes
v N - No
QQRCOD QQRCOD Reason code
v T1 - No indexes exist.
v T2 - Indexes exist, but none could be used.
v T3 - Optimizer chose table scan over available indexes.
QQIDXD QQIDXD Columns for the index advised
QVQTBL QVQTBL Queried table, long name
QVQLIB QVQLIB Library of queried table, long name
QVPTBL QVPTBL Base table, long name
QVPLIB QVPLIB Library of base table, long name
QVBNDY QVBNDY I/O or CPU bound. Possible values are:
v I - I/O bound
v C - CPU bound
QVRCNT QVRCNT Unique refresh counter
QVJFANO QVJFANO Join fan out. Possible values are:
v N - Normal join situation where fanout is allowed and each
matching row of the join fanout is returned.
v D - Distinct fanout. Join fanout is allowed however none of the join
fanout rows are returned.
v U - Unique fanout. Join fanout is not allowed. Error situation if join
fanout occurs.
QVFILES QVFILES Number of tables joined
QVPARPF QVPARPF Parallel Prefetch (Y/N)
QVPARPL QVPARPL Parallel Preload (Y/N)
QVPARD QVPARD Parallel degree requested

150 DB2 UDB for iSeries Database Performance and Query Optimization V5R1
Table 14. QQQ3000 - Summary Row for Table Scan (continued)
Logical Column Physical Column
Name Name Description
QVPARU QVPARU Parallel degree used
QVPARRC QVPARRC Reason parallel processing was limited
QVCTIM QVCTIM Estimated cumulative time, in seconds
QVSKIPS QQC11 Skip sequential table scan (Y/N)
QVTBLSZ QQI3 Size of table being queried
QVTSFLDS QVC3001 Columns used for dataspace selection
QVDVFLD QQC14 Derived column selection (Y/N)
QVDVFLDS QVC3002 Columns used for derived column selection

Database monitor logical table 3001 - Summary Row for Index Used
|...+....1....+....2....+....3....+....4....+....5....+....6....+....7....+....8
A*
A* DB Monitor logical table 3001 - Summary Row for Index Used
A*
A R QQQ3001 PTABLE(*CURLIB/QAQQDBMN)
A QQRID
A QQTIME
A QQJFLD
A QQRDBN
A QQSYS
A QQJOB
A QQUSER
A QQJNUM
A QQTHRD RENAME(QQI9) +
COLHDG(’Thread’ +
’Identifier’)
A QQUCNT
A QQUDEF
A QQQDTN
A QQQDTL
A QQMATN
A QQMATL
A QDQDTN RENAME(QVP15A) +
COLHDG(’Decomposed’ +
’Subselect’ +
’Number’)
A QDQDTT RENAME(QVP15B) +
COLHDG(’Number of’ +
’Decomposed’ +
’Subselects’)
A QDQDTR RENAME(QVP15C) +
COLHDG(’Decomposed’ +
’Reason’ +
’Code’)
A QDQDTS RENAME(QVP15D) +
COLHDG(’Starting’ +
’Decomposed’ +
’Subselect’)
A QQTLN
A QQTFN
A QQTMN
A QQPTLN
A QQPTFN
A QQPTMN
A QQILNM
A QQIFNM

Appendix A. Database monitor: DDS 151

 A QQIMNM
A QQTOTR
A QQREST
A QQFKEY
A QQKSEL
A QQAJN
A QQEPT
A QQJNP
A QQJNDS RENAME(QQI1) +
COLHDG(’Data Space’ +
’Number’)
A QQJNMT RENAME(QQC21) +
COLHDG(’Join’ ’Method’)
A QQJNTY RENAME(QQC22) +
COLHDG(’Join’ ’Type’)
A QQJNOP RENAME(QQC23) +
COLHDG(’Join’ ’Operator’)
A QQIDXPK RENAME(QQI2) +
COLHDG(’Advised’ +
’Primary’ +
’Keys’)
A QQKP
A QQKPN RENAME(QQI3) +
COLHDG(’Number of’ ’Key’ +
’Positioning’ +
’Columns’)
A QQKS
A QQDSS
A QQIDXA
A QQRCOD
A QQIDXD
A QQCST RENAME(QQC11) +
COLHDG(’Index’ +
’Is a’ +
’Constraint’)
A QQCSTN RENAME(QQ1000) +
COLHDG(’Constraint’ +
’Name’)
A*
A* Columns added for Visual Explain
A*
A QVQTBL
A QVQLIB
A QVPTBL
A QVPLIB
A QVINAM
A QVILIB
A QVBNDY
A QVRCNT
A QVJFANO
A QVFILES
A QVPARPF
A QVPARPL
A QVPARD
A QVPARU
A QVPARRC
A QVCTIM
A QVKOA RENAME(QVC14) +
COLHDG(’Index’ +
’Only’ +
’Access’)
A QVIDXM RENAME(QQC12) +
COLHDG(’Index’ +
’fits in’ +
’Memory’)
A QVIDXTY RENAME(QQC15) +
COLHDG(’Index’ +

152 DB2 UDB for iSeries Database Performance and Query Optimization V5R1
 ’Type’)
A QVIDXUS RENAME(QVC12) +
COLHDG(’Index’ +
’Usage’)
A QVIDXN RENAME(QQI4) +
COLHDG(’Number’ +
’Index’ +
’Entries’)
A QVIDXUQ RENAME(QQI5) +
COLHDG(’Number’ +
’Unique’ +
’Values’)
A QVIDXPO RENAME(QQI6) +
COLHDG(’Percent’ +
’Overflow’)
A QVIDXVZ RENAME(QQI7) +
COLHDG(’Vector’ +
’Size’)
A QVIDXSZ RENAME(QQI8) +
COLHDG(’Index’ +
’Size’)
A QVIDXPZ RENAME(QQIA) +
COLHDG(’Index’ +
’Page’ +
’Size’)
A QQPSIZ RENAME(QVP154) +
COLHDG(’Pool’ +
’Size’)
A QQPID RENAME(QVP155) +
COLHDG(’Pool’ +
’ID’)
A QVTBLSZ RENAME(QVP156) +
COLHDG(’Base’ +
’Table’ +
’Size’)
A QVSKIPS RENAME(QQC16) +
COLHDG(’Skip’ +
’Sequential’)
A QVIDXTR RENAME(QVC13) +
COLHDG(’Tertiary’ +
’Indexes’
’Exist’)
A QVTSFLDS RENAME(QVC3001) +
COLHDG(’Columns for’ +
’Data Space’ +
’Selection’)
A QVDVFLD RENAME(QVC12 +
COLHDG(’Derived’ +
’Column’ +
’Selection’)
A QVDVFLDS RENAME(QVC3002) +
COLHDG(’Columns for’ +
’Derived’ +
’Selection’)
A QVSKEYP RENAME(QVC3003) +
COLHDG(’Key’ +
’Positioning’ +
’Columns’)
A QVSKEYS RENAME(QVC3004) +
COLHDG(’Key’ +
’Selection’ +
’Columns’)
A QVJKEYS RENAME(QVC3005) +
COLHDG(’Join’ +
’Selection’ +
’Columns’)
A QVOKEYS RENAME(QVC3006) +

Appendix A. Database monitor: DDS 153

 COLHDG(’Ordering’ +
’Columns’)
A QVGKEYS RENAME(QVC3007) +
COLHDG(’Grouping’ +
’Columns’)
A K QQJFLD
A S QQRID CMP(EQ 3001)

Table 15. QQQ3001 - Summary Row for Index Used

Logical Column Physical Column
Name Name Description
QQRID QQRID Row identification
QQTIME QQTIME Time row was created
QQJFLD QQJFLD Join column (unique per job)
QQRDBN QQRDBN Relational database name
QQSYS QQSYS System name
QQJOB QQJOB Job name
QQUSER QQUSER Job user
QQJNUM QQJNUM Job number
QQTHRD QQI9 Thread identifier
QQUCNT QQUCNT Unique count (unique per query)
QQUDEF QQUDEF User defined column
QQQDTN QQQDTN Unique subselect number
QQQDTL QQQDTL Subselect nested level
QQMATN QQMATN Materialized view subselect number
QQMATL QQMATL Materialized view nested level
QDQDTN QVP15A Decomposed query subselect number, unique across all
decomposed subselects
QDQDTT QVP15B Total number of decomposed subselects
QDQDTR QVP15C Decomposed query subselect reason code
QDQDTS QVP15D Decomposed query subselect number for the first decomposed
subselect
QQTLN QQTLN Library of table queried
QQTFN QQTFN Name of table queried
QQTMN QQTMN Member name of table queried
QQPTLN QQPTLN Library name of base table
QQPTFN QQPTFN Name of base table for table queried
QQPTMN QQPTMN Member name of base table
QQILNM QQILNM Library name of index used for access
QQIFNM QQIFNM Name of index used for access
QQIMNM QQIMNM Member name of index used for access
QQTOTR QQTOTR Total rows in base table
QQREST QQREST Estimated number of rows selected
QQFKEY QQFKEY Columns selected thru index scan-key positioning
QQKSEL QQKSEL Columns selected thru index scan-key selection
QQAJN QQAJN Estimated number of joined rows

154 DB2 UDB for iSeries Database Performance and Query Optimization V5R1
Table 15. QQQ3001 - Summary Row for Index Used (continued)
Logical Column Physical Column
Name Name Description
QQEPT QQEPT Estimated processing time, in seconds
QQJNP QQJNP Join position - when available
QQJNDS QQI1 dataspace number
QQJNMT QQC21 Join method - when available
v NL - Nested loop
v MF - Nested loop with selection
v HJ - Hash join
QQJNTY QQC22 Join type - when available
v IN - Inner join
v PO - Left partial outer join
v EX - Exception join
QQJNOP QQC23 Join operator - when available
v EQ - Equal
v NE - Not equal
v GT - Greater than
v GE - Greater than or equal
v LT - Less than
v LE - Less than or equal
v CP - Cartesian product
QQIDXPK QQI2 Number of advised key columns that use index scan-key positioning
QQKP QQKP Index scan-key positioning
v Y - Yes
v N - No
QQKPN QQI3 Number of columns that use index scan-key positioning for the index
used
QQKS QQKS Index scan-key selection
v Y - Yes
v N - No
QQDSS QQDSS dataspace selection
v Y - Yes
v N - No
QQIDXA QQIDXA Index advised
v Y - Yes
v N - No
QQRCOD QQRCOD Reason code
v I1 - Row selection
v I2 - Ordering/Grouping
v I3 - Row selection and Ordering/Grouping
v I4 - Nested loop join
v I5 - Row selection using bitmap processing
QQIDXD QQIDXD Columns for index advised
QQCST QQC11 Index is a constraint (Y/N)
QQCSTN QQ1000 Constraint name

Appendix A. Database monitor: DDS 155

Table 15. QQQ3001 - Summary Row for Index Used (continued)
Logical Column Physical Column
Name Name Description
QVQTBL QVQTBL Queried table, long name
QVQLIB QVQLIB Library of queried table, long name
QVPTBL QVPTBL Base table, long name
QVPLIB QVPLIB Library of base table, long name
QVINAM QVINAM Name of index (or constraint) used, long name
QVILIB QVILIB Library of index used, long name
QVBNDY QVBNDY I/O or CPU bound. Possible values are:
v I - I/O bound
v C - CPU bound
QVRCNT QVRCNT Unique refresh counter
QVJFANO QVJFANO Join fan out. Possible values are:
v N - Normal join situation where fanout is allowed and each
matching row of the join fanout is returned.
v D - Distinct fanout. Join fanout is allowed however none of the join
fanout rows are returned.
v U - Unique fanout. Join fanout is not allowed. Error situation if join
fanout occurs.
QVFILES QVFILES Number of tables joined
QVPARPF QVPARPF Parallel Prefetch (Y/N)
QVPARPL QVPARPL Parallel Preload (Y/N)
QVPARD QVPARD Parallel degree requested
QVPARU QVPARU Parallel degree used
QVPARRC QVPARRC Reason parallel processing was limited
QVCTIM QVCTIM Estimated cumulative time, in seconds
QVKOA QVC14 Index only access (Y/N)
QVIDXM QQC12 Index fits in memory (Y/N)
QVIDXTY QQC15 Type of Index. Possible values are:
v B - Binary Radix Index
v C - Constraint (Binary Radix)
v E - Encoded Vector Index (EVI)
v X - Query created temporary index
QVIDXUS QVC12 Index Usage. Possible values are:
v P - Primary Index
v T - Tertiary (AND/OR) Index
QVIDXN QQI4 Number of index entries
QVIDXUQ QQI5 Number of unique key values
QVIDXPO QQI6 Percent overflow
QVIDXVZ QQI7 Vector size
QVIDXSZ QQI8 Index size
QVIDXPZ QQIA Index page size
QQPSIZ QVP154 Pool size

156 DB2 UDB for iSeries Database Performance and Query Optimization V5R1
Table 15. QQQ3001 - Summary Row for Index Used (continued)
Logical Column Physical Column
Name Name Description
QQPID QVP155 Pool id
QVTBLSZ QVP156 Table size
QVSKIPS QQC16 Skip sequential table scan (Y/N)
QVIDXTR QVC13 Tertiary indexes exist (Y/N)
QVTSFLDS QVC3001 Columns used for dataspace selection
QVDVFLD QQC14 Derived column selection (Y/N)
QVDVFLDS QVC3002 Columns used for derived column selection
QVSKEYP QVC3003 Columns used for index scan-key positioning
QVSKEYS QVC3004 Columns used for index scan-key selection
QVJKEYS QVC3005 Columns used for Join selection
QVOKEYS QVC3006 Columns used for Ordering
QVGKEYS QVC3007 Columns used for Grouping

Database monitor logical table 3002 - Summary Row for Index Created
|...+....1....+....2....+....3....+....4....+....5....+....6....+....7....+....8
A*
A* DB Monitor logical table 3002 - Summary Row for Index Created
A*
A R QQQ3002 PTABLE(*CURLIB/QAQQDBMN)
A QQRID
A QQTIME
A QQJFLD
A QQRDBN
A QQSYS
A QQJOB
A QQUSER
A QQJNUM
A QQTHRD RENAME(QQI9) +
COLHDG(’Thread’ +
’Identifier’)
A QQUCNT
A QQUDEF
A QQQDTN
A QQQDTL
A QQMATN
A QQMATL
A QDQDTN RENAME(QVP15A) +
COLHDG(’Decomposed’ +
’Subselect’ +
’Number’)
A QDQDTT RENAME(QVP15B) +
COLHDG(’Number of’ +
’Decomposed’ +
’Subselects’)
A QDQDTR RENAME(QVP15C) +
COLHDG(’Decomposed’ +
’Reason’ +
’Code’)
A QDQDTS RENAME(QVP15D) +
COLHDG(’Starting’ +
’Decomposed’ +
’Subselect’)
A QQTLN
A QQTFN

Appendix A. Database monitor: DDS 157

 A QQTMN
A QQPTLN
A QQPTFN
A QQPTMN
A QQILNM
A QQIFNM
A QQIMNM
A QQNTNM
A QQNLNM
A QQSTIM
A QQETIM
A QQTOTR
A QQRIDX
A QQREST
A QQFKEY
A QQKSEL
A QQAJN
A QQEPT
A QQJNP
A QQJNDS RENAME(QQI1) +
COLHDG(’Data Space’ +
’Number’)
A QQJNMT RENAME(QQC21) +
COLHDG(’Join’ ’Method’)
A QQJNTY RENAME(QQC22) +
COLHDG(’Join’ ’Type’)
A QQJNOP RENAME(QQC23) +
COLHDG(’Join’ ’Operator’)
A QQIDXK RENAME(QQI2) +
COLHDG(’Advised’ +
’Primary’ ’Keys’)
A QQKP
A QQKPN RENAME(QQI3) +
COLHDG(’Number’ ’Key’ +
’Positioning’ +
’Columns’)
A QQKS
A QQDSS
A QQIDXA
A QQRCOD
A QQIDXD
A QQCRTK RENAME(QQ1000) +
COLHDG(’Key Columns’ +
’of Index’ +
’Created’)
A*
A* Columns added for Visual Explain
A*
A QVQTBL
A QVQLIB
A QVPTBL
A QVPLIB
A QVINAM
A QVILIB
A QVBNDY
A QVRCNT
A QVJFANO
A QVFILES
A QVPARPF
A QVPARPL
A QVPARD
A QVPARU
A QVPARRC
A QVCTIM
A QVTIXN RENAME(QQC101) +
COLHDG(’Name of’ +
’Index’ +

158 DB2 UDB for iSeries Database Performance and Query Optimization V5R1
 ’Created’)
A QVTIXL RENAME(QQC102) +
COLHDG(’Library of’ +
’Index’ +
’Created’)
A QVTIXPZ RENAME(QQI4) +
COLHDG(’Page Size’ +
’of Index’ +
’Created’)
A QVTIXRZ RENAME(QQI5) +
COLHDG(’Row Size’ +
’of Index’ +
’Created’)
A QVTIXACF RENAME(QQC14) +
COLHDG(’ACS’ +
’Table’ +
’Used’)
A QVTIXACS RENAME(QQC103) +
COLHDG(’Alternate’ +
’Collating’ +
’Sequence’ +
’Table’)
A QVTIXACL RENAME(QQC104) +
COLHDG(’Alternate’ +
’Collating’ +
’Sequence’ +
’Library’)
A QVTIXRU RENAME(QVC13) +
COLHDG(’Index +
’Created is’ +
’Reusable’)
A QVTIXSP RENAME(QVC14) +
COLHDG(’Index +
’Created is’ +
’Sparse’)
A QVTIXTY RENAME(QVC1F) +
COLHDG(’Type of’ +
’Index’ +
’Created’)
A QVTIXUQ RENAME(QVP15F) +
COLHDG(’Number of’ +
’Unique Values’ +
’Index Created’)
A QVTIXPO RENAME(QVC15) +
COLHDG(’Permanent’ +
’Index’ +
’Created’)
A QVTIXFX RENAME(QVC16) +
COLHDG(’Index’ +
’From’ +
’Index’)
A QVTIXPD RENAME(QVP151) +
COLHDG(’Parallel’ +
’Degree ’ +
’Requested’)
A QVTIXPU RENAME(QVP152) +
COLHDG(’Parallel’ +
’Degree ’ +
’Used’)
A QVTIXPRC RENAME(QVP153) +
COLHDG(’Parallel’ +
’Degree ’ +
’Limited’)
A QVKOA RENAME(QVC17) +
COLHDG(’Index’ +
’Only’ +
’Access’)

Appendix A. Database monitor: DDS 159

 A QVIDXM RENAME(QVC18) +
COLHDG(’Index’ +
’fits in’ +
’Memory’)
A QVIDXTY RENAME(QVC1B) +
COLHDG(’Index’ +
’Type’)
A QVIDXN RENAME(QQI6) +
COLHDG(’Entries in’ +
’Index’ +
’Used’)
A QVIDXUQ RENAME(QQI7) +
COLHDG(’Number’ +
’Unique’ +
’Values’)
A QVIDXPO RENAME(QVP158) +
COLHDG(’Percent’ +
’Overflow’)
A QVIDXVZ RENAME(QVP159) +
COLHDG(’Vector’ +
’Size’)
A QVIDXSZ RENAME(QQI8) +
COLHDG(’Size of’ +
’Index’ +
’Used’)
A QVIDXPZ RENAME(QVP156) +
COLHDG(’Page Size’ +
’Index’ +
’Used’)
A QQPSIZ RENAME(QVP154) +
COLHDG(’Pool’ +
’Size’)
A QQPID RENAME(QVP155) +
COLHDG(’Pool’ +
’ID’)
A QVTBLSZ RENAME(QVP157) +
COLHDG(’Table’ +
’Size’)
A QVSKIPS RENAME(QVC1C) +
COLHDG(’Skip’ +
’Sequential’)
A QVTSFLDS RENAME(QVC3001) +
COLHDG(’Columns for’ +
’Data Space’ +
’Selection’)
A QVDVFLD RENAME(QVC1E +
COLHDG(’Derived’ +
’Column’ +
’Selection’)
A QVDVFLDS RENAME(QVC3002) +
COLHDG(’Columns for’ +
’Derived’ +
’Selection’)
A QVSKEYP RENAME(QVC3003) +
COLHDG(’Columns Used’ +
’for Key’ +
’Positioning’)
A QVSKEYS RENAME(QVC3004) +
COLHDG(’Columns Used’ +
’for Key’ +
’Selection’)
A K QQJFLD
A S QQRID CMP(EQ 3002)

160 DB2 UDB for iSeries Database Performance and Query Optimization V5R1
Table 16. QQQ3002 - Summary row for Index Created
Logical Column Physical Column
Name Name Description
QQRID QQRID Row identification
QQTIME QQTIME Time row was created
QQJFLD QQJFLD Join column (unique per job)
QQRDBN QQRDBN Relational database name
QQSYS QQSYS System name
QQJOB QQJOB Job name
QQUSER QQUSER Job user
QQJNUM QQJNUM Job number
QQTHRD QQI9 Thread identifier
QQUCNT QQUCNT Unique count (unique per query)
QQUDEF QQUDEF User defined column
QQQDTN QQQDTN Unique subselect number
QQQDTL QQQDTL Subselect nested level
QQMATN QQMATN Materialized view subselect number
QQMATL QQMATL Materialized view nested level
QDQDTN QVP15A Decomposed query subselect number, unique across all
decomposed subselects
QDQDTT QVP15B Total number of decomposed subselects
QDQDTR QVP15C Decomposed query subselect reason code
QDQDTS QVP15D Decomposed query subselect number for the first decomposed
subselect
QQTLN QQTLN Library of table queried
QQTFN QQTFN Name of table queried
QQTMN QQTMN Member name of table queried
QQPTLN QQPTLN Library name of base table
QQPTFN QQPTFN Name of base table for table queried
QQPTMN QQPTMN Member name of base table
QQILNM QQILNM Library name of index used for access
QQIFNM QQIFNM Name of index used for access
QQIMNM QQIMNM Member name of index used for access
QQNTNM QQNTNM NLSS library
QQNLNM QQNLNM NLSS table
QQSTIM QQSTIM Start timestamp
QQETIM QQETIM End timestamp
QQTOTR QQTOTR Total rows in table
QQRIDX QQRIDX Number of entries in index created
QQREST QQREST Estimated number of rows selected
QQFKEY QQFKEY Keys selected thru index scan-key positioning
QQKSEL QQKSEL Keys selected thru index scan-key selection
QQAJN QQAJN Estimated number of joined rows

Appendix A. Database monitor: DDS 161

Table 16. QQQ3002 - Summary row for Index Created (continued)
Logical Column Physical Column
Name Name Description
QQEPT QQEPT Estimated processing time, in seconds
QQJNP QQJNP Join position - when available
QQJNDS QQI1 dataspace number
QQJNMT QQC21 Join method - when available
v NL - Nested loop
v MF - Nested loop with selection
v HJ - Hash join
QQJNTY QQC22 Join type - when available
v IN - Inner join
v PO - Left partial outer join
v EX - Exception join
QQJNOP QQC23 Join operator - when available
v EQ - Equal
v NE - Not equal
v GT - Greater than
v GE - Greater than or equal
v LT - Less than
v LE - Less than or equal
v CP - Cartesian product
QQIDXK QQI2 Number of advised key columns that use index scan-key positioning
QQKP QQKP Index scan-key positioning
v Y - Yes
v N - No
QQKPN QQI3 Number of columns that use index scan-key positioning for the index
used
QQKS QQKS Index scan-key selection
v Y - Yes
v N - No
QQDSS QQDSS dataspace selection
v Y - Yes
v N - No
QQIDXA QQIDXA Index advised
v Y - Yes
v N - No
QQRCOD QQRCOD Reason code
v I1 - Row selection
v I2 - Ordering/Grouping
v I3 - Row selection and Ordering/Grouping
v I4 - Nested loop join
QQIDXD QQIDXD Key columns for index advised
QQCRTK QQ1000 Key columns for index created
QVQTBL QVQTBL Queried table, long name

162 DB2 UDB for iSeries Database Performance and Query Optimization V5R1
Table 16. QQQ3002 - Summary row for Index Created (continued)
Logical Column Physical Column
Name Name Description
QVQLIB QVQLIB Library of queried table, long name
QVPTBL QVPTBL Base table, long name
QVPLIB QVPLIB Library of base table, long name
QVINAM QVINAM Name of index (or constraint) used, long name
QVILIB QVILIB Library of index used, long name
QVBNDY QVBNDY I/O or CPU bound. Possible values are:
v I - I/O bound
v C - CPU bound
QVRCNT QVRCNT Unique refresh counter
QVJFANO QVJFANO Join fan out. Possible values are:
v N - Normal join situation where fanout is allowed and each
matching row of the join fanout is returned.
v D - Distinct fanout. Join fanout is allowed however none of the join
fanout rows are returned.
v U - Unique fanout. Join fanout is not allowed. Error situation if join
fanout occurs.
QVFILES QVFILES Number of tables joined
QVPARPF QVPARPF Parallel Prefetch (Y/N)
QVPARPL QVPARPL Parallel Preload (index used)
QVPARD QVPARD Parallel degree requested (index used)
QVPARU QVPARU Parallel degree used (index used)
QVPARRC QVPARRC Reason parallel processing was limited (index used)
QVCTIM QVCTIM Estimated cumulative time, in seconds
QVTIXN QQC101 Name of index created
QVTIXL QQC102 Library of index created
QVTIXPZ QQI4 Page size of index created
QVTIXRZ QQI5 Row size of index created
QVTIXACS QQC103 Alternate Collating Sequence table of index created.
QVTIXACL QQC104 Alternate Collating Sequence library of index created.
QVTIXRU QVC13 Index created is reusable (Y/N)
QVTIXSP QVC14 Index created is sparse index (Y/N)
QVTIXTY QVC1F Type of index created. Possible values:
v B - Binary Radix Index
v E - Encoded Vector Index (EVI)
QVTIXUQ QVP15A Number of unique values of index created if index created is an EVI
index.
QVTIXPO QVC15 Permanent index created (Y/N)
QVTIXFX QVC16 Index from index (Y/N)
QVTIXPD QVP151 Parallel degree requested (index created)
QVTIXPU QVP152 Parallel degree used (index created)
QVTIXPRC QVP153 Reason parallel processing was limited (index created)

Appendix A. Database monitor: DDS 163

Table 16. QQQ3002 - Summary row for Index Created (continued)
Logical Column Physical Column
Name Name Description
QVKOA QVC17 Index only access (Y/N)
QVIDXM QVC18 Index fits in memory (Y/N)
QVIDXTY QVC1B Type of Index. Possible values are:
v B - Binary Radix Index
v C - Constraint (Binary Radix)
v E - Encoded Vector Index (EVI)
v T - Tertiary (AND/OR) Index
QVIDXN QQI6 Number of index entries, index used
QVIDXUQ QQI7 Number of unique key values, index used
QVIDXPO QVP158 Percent overflow, index used
QVIDXVZ QVP159 Vector size, index used
QVIDXSZ QQI8 Size of index used.
QVIDXPZ QVP156 Index page size
QQPSIZ QVP154 Pool size
QQPID QVP155 Pool id
QVTBLSZ QVP157 Table size
QVSKIPS QVC1C Skip sequential table scan (Y/N)
QVTSFLDS QVC3001 Columns used for dataspace selection
QVDVFLD QVC1E Derived column selection (Y/N)
QVDVFLDS QVC3002 Columns used for derived column seleciton
QVSKEYP QVC3003 Columns used for index scan-key positioning
QVSKEYS QVC3004 Columns used for index scan-key selection

Database monitor logical table 3003 - Summary Row for Query Sort
|...+....1....+....2....+....3....+....4....+....5....+....6....+....7....+....8
A*
A* DB Monitor logical table 3003 - Summary Row for Query Sort
A*
A R QQQ3003 PTABLE(*CURLIB/QAQQDBMN)
A QQRID
A QQTIME
A QQJFLD
A QQRDBN
A QQSYS
A QQJOB
A QQUSER
A QQJNUM
A QQTHRD RENAME(QQI9) +
COLHDG(’Thread’ +
’Identifier’)
A QQUCNT
A QQUDEF
A QQQDTN
A QQQDTL
A QQMATN
A QQMATL
A QDQDTN RENAME(QVP15A) +
COLHDG(’Decomposed’ +
’Subselect’ +

164 DB2 UDB for iSeries Database Performance and Query Optimization V5R1
 ’Number’)
A QDQDTT RENAME(QVP15B) +
COLHDG(’Number of’ +
’Decomposed’ +
’Subselects’)
A QDQDTR RENAME(QVP15C) +
COLHDG(’Decomposed’ +
’Reason’ +
’Code’)
A QDQDTS RENAME(QVP15D) +
COLHDG(’Starting’ +
’Decomposed’ +
’Subselect’)
A QQSTIM
A QQETIM
A QQRSS
A QQSSIZ RENAME(QQI1) +
COLHDG(’Size of’ +
’Sort’ +
’Space’)
A QQPSIZ RENAME(QQI2) +
COLHDG(’Pool’ +
’Size’)
A QQPID RENAME(QQI3) +
COLHDG(’Pool’ +
’ID’)
A QQIBUF RENAME(QQI4) +
COLHDG(’Internal’ +
’Buffer’ +
’Length’)
A QQEBUF RENAME(QQI5) +
COLHDG(’External’ +
’Buffer’ +
’Length’)
A QQRCOD
A*
A* Columns added for Visual Explain
A*
A QVBNDY
A QVRCNT
A QVPARPF
A QVPARPL
A QVPARD
A QVPARU
A QVPARRC
A QQEPT
A QVCTIM
A QQAJN
A QQJNP
A QQJNDS RENAME(QQI6) +
COLHDG(’Data Space’ +
’Number’)
A QQJNMT RENAME(QQC21) +
COLHDG(’Join’ ’Method’)
A QQJNTY RENAME(QQC22) +
COLHDG(’Join’ ’Type’)
A QQJNOP RENAME(QQC23) +
COLHDG(’Join’ ’Operator’)
A QVJFANO
A QVFILES
A K QQJFLD
A S QQRID CMP(EQ 3003)

Appendix A. Database monitor: DDS 165

Table 17. QQQ3003 - Summary Row for Query Sort
Logical Column Physical Column
Name Name Description
QQRID QQRID Row identification
QQTIME QQTIME Time row was created
QQJFLD QQJFLD Join column (unique per job)
QQRDBN QQRDBN Relational database name
QQSYS QQSYS System name
QQJOB QQJOB Job name
QQUSER QQUSER Job user
QQJNUM QQJNUM Job number
QQTHRD QQI9 Thread identifier
QQUCNT QQUCNT Unique count (unique per query)
QQUDEF QQUDEF User defined column
QQQDTN QQQDTN Unique subselect number
QQQDTL QQQDTL Subselect nested level
QQMATN QQMATN Materialized view subselect number
QQMATL QQMATL Materialized view nested level
QDQDTN QVP15A Decomposed query subselect number, unique across all
decomposed subselects
QDQDTT QVP15B Total number of decomposed subselects
QDQDTR QVP15C Decomposed query subselect reason code
QDQDTS QVP15D Decomposed query subselect number for the first decomposed
subselect
QQSTIM QQSTIM Start timestamp
QQETIM QQETIM End timestamp
QQRSS QQRSS Number of rows selected or sorted
QQSSIZ QQI1 Size of sort space
QQPSIZ QQI2 Pool size
QQPID QQI3 Pool id
QQIBUF QQI4 Internal sort buffer length
QQEBUF QQI5 External sort buffer length

166 DB2 UDB for iSeries Database Performance and Query Optimization V5R1
Table 17. QQQ3003 - Summary Row for Query Sort (continued)
Logical Column Physical Column
Name Name Description
QQRCOD QQRCOD Reason code
v F1 - Query contains grouping columns (GROUP BY) from more
that one table, or contains grouping columns from a secondary
table of a join query that cannot be reordered.
v F2 - Query contains ordering columns (ORDER BY) from more
that one table, or contains ordering columns from a secondary
table of a join query that cannot be reordered.
v F3 - The grouping and ordering columns are not compatible.
v F4 - DISTINCT was specified for the query.
v F5 - UNION was specified for the query.
v F6 - Query had to be implemented using a sort. Key length of
more than 2000 bytes or more than 120 key columns specified for
ordering.
v F7 - Query optimizer chose to use a sort rather than an index to
order the results of the query.
v F8 - Perform specified row selection to minimize I/O wait time.
v FC - The query contains grouping fields and there is a read trigger
on at least one of the physical files in the query.
QVBNDY QVBNDY I/O or CPU bound. Possible values are:
v I - I/O bound
v C - CPU bound
QVRCNT QVRCNT Unique refresh counter
QVPARPF QVPARPF Parallel Prefetch (Y/N)
QVPARPL QVPARPL Parallel Preload (index used)
QVPARD QVPARD Parallel degree requested (index used)
QVPARU QVPARU Parallel degree used (index used)
QVPARRC QVPARRC Reason parallel processing was limited (index used)
QQEPT QQEPT Estimated processing time, in seconds
QVCTIM QVCTIM Estimated cumulative time, in seconds
QQAJN QQAJN Estimated number of joined rows
QQJNP QQJNP Join position - when available
QQJNDS QQI6 dataspace number
QQJNMT QQC21 Join method - when available
v NL - Nested loop
v MF - Nested loop with selection
v HJ - Hash join
QQJNTY QQC22 Join type - when available
v IN - Inner join
v PO - Left partial outer join
v EX - Exception join

Appendix A. Database monitor: DDS 167

Table 17. QQQ3003 - Summary Row for Query Sort (continued)
Logical Column Physical Column
Name Name Description
QQJNOP QQC23 Join operator - when available
v EQ - Equal
v NE - Not equal
v GT - Greater than
v GE - Greater than or equal
v LT - Less than
v LE - Less than or equal
v CP - Cartesian product
QVJFANO QVJFANO Join fan out. Possible values are:
v N - Normal join situation where fanout is allowed and each
matching row of the join fanout is returned.
v D - Distinct fanout. Join fanout is allowed however none of the join
fanout rows are returned.
v U - Unique fanout. Join fanout is not allowed. Error situation if join
fanout occurs.
QVFILES QVFILES Number of tables joined

Database monitor logical table 3004 - Summary Row for Temp Table
|...+....1....+....2....+....3....+....4....+....5....+....6....+....7....+....8
A*
A* DB Monitor logical table 3004 - Summary Row for Temp Table
A*
A R QQQ3004 PTABLE(*CURLIB/QAQQDBMN)
A QQRID
A QQTIME
A QQJFLD
A QQRDBN
A QQSYS
A QQJOB
A QQUSER
A QQJNUM
A QQTHRD RENAME(QQI9) +
COLHDG(’Thread’ +
’Identifier’)
A QQUCNT
A QQUDEF
A QQQDTN
A QQQDTL
A QQMATN
A QQMATL
A QDQDTN RENAME(QVP15A) +
COLHDG(’Decomposed’ +
’Subselect’ +
’Number’)
A QDQDTT RENAME(QVP15B) +
COLHDG(’Number of’ +
’Decomposed’ +
’Subselects’)
A QDQDTR RENAME(QVP15C) +
COLHDG(’Decomposed’ +
’Reason’ +
’Code’)
A QDQDTS RENAME(QVP15D) +
COLHDG(’Starting’ +
’Decomposed’ +

168 DB2 UDB for iSeries Database Performance and Query Optimization V5R1
 ’Subselect’)
A QQTLN
A QQTFN
A QQTMN
A QQPTLN
A QQPTFN
A QQPTMN
A QQSTIM
A QQETIM
A QQDFVL RENAME(QQC11) +
COLHDG(’Default’ +
’Values’)
A QQTMPR
A QQRCOD
A*
A* Columns added for Visual Explain
A*
A QVQTBL
A QVQLIB
A QVPTBL
A QVPLIB
A QVTTBLN RENAME(QQC101) +
COLHDG(’Temporary’ +
’Table’ +
’Name’)
A QVTTBLL RENAME(QQC102) +
COLHDG(’Temporary’ +
’Table’ +
’Library’)
A QVBNDY
A QVRCNT
A QVPARPF
A QVPARPL
A QVPARD
A QVPARU
A QVPARRC
A QQEPT
A QVCTIM
A QQAJN
A QQJNP
A QQJNDS RENAME(QQI6) +
COLHDG(’Data Space’ +
’Number’)
A QQJNMT RENAME(QQC21) +
COLHDG(’Join’ ’Method’)
A QQJNTY RENAME(QQC22) +
COLHDG(’Join’ ’Type’)
A QQJNOP RENAME(QQC23) +
COLHDG(’Join’ ’Operator’)
A QVJFANO
A QVFILES
A QVTTRSZ RENAME(QQI2) +
COLHDG(’Row Size’ +
’Temporary’ +
’Table’)
A QVTTSIZ RENAME(QQI3) +
COLHDG(’Table Size’ +
’Temporary’ +
’Table’)
A QVTTRST RENAME(QQC12) +
COLHDG(’Temporary’ +
’Result’)
A QVTTDST RENAME(QQC13) +
COLHDG(’Distributed’ +
’Table’)
A QVTTNOD RENAME(QVC3001) +
COLHDG(’Data’ +

Appendix A. Database monitor: DDS 169

 ’Nodes’)
A K QQJFLD
A S QQRID CMP(EQ 3004)

Table 18. QQQ3004 - Summary Row for Temp Table

Logical Column Physical Column
Name Name Description
QQRID QQRID Row identification
QQTIME QQTIME Time row was created
QQJFLD QQJFLD Join column (unique per job)
QQRDBN QQRDBN Relational database name
QQSYS QQSYS System name
QQJOB QQJOB Job name
QQUSER QQUSER Job user
QQJNUM QQJNUM Job number
QQTHRD QQI9 Thread identifier
QQUCNT QQUCNT Unique count (unique per query)
QQUDEF QQUDEF User defined column
QQQDTN QQQDTN Unique subselect number
QQQDTL QQQDTL Subselect nested level
QQMATN QQMATN Materialized view subselect number
QQMATL QQMATL Materialized view nested level
QDQDTN QVP15A Decomposed query subselect number, unique across all
decomposed subselects
QDQDTT QVP15B Total number of decomposed subselects
QDQDTR QVP15C Decomposed query subselect reason code
QDQDTS QVP15D Decomposed query subselect number for the first decomposed
subselect
QQTLN QQTLN Library of table queried
QQTFN QQTFN Name of table queried
QQTMN QQTMN Member name of table queried
QQPTLN QQPTLN Library name of base table
QQPTFN QQPTFN Name of base table for table queried
QQPTMN QQPTMN Member name of base table
QQSTIM QQSTIM Start timestamp
QQETIM QQETIM End timestamp
QQDFVL QQC11 Default values may be present in temporary
v Y - Yes
v N - No
QQTMPR QQTMPR Number of rows in the temporary

170 DB2 UDB for iSeries Database Performance and Query Optimization V5R1
Table 18. QQQ3004 - Summary Row for Temp Table (continued)
Logical Column Physical Column
Name Name Description
QQRCOD QQRCOD Reason code. Possible values are:
v F1 - Query contains grouping columns (GROUP BY) from more
that one table, or contains grouping columns from a secondary
table of a join query that cannot be reordered.
v F2 - Query contains ordering columns (ORDER BY) from more
that one table, or contains ordering columns from a secondary
table of a join query that cannot be reordered.
v F3 - The grouping and ordering columns are not compatible.
v F4 - DISTINCT was specified for the query.
v F5 - UNION was specified for the query.
v F6 - Query had to be implemented using a sort. Key length of
more than 2000 bytes or more than 120 key columns specified for
ordering.
v F7 - Query optimizer chose to use a sort sort rather than an index
to order the results of the query.
v F8 - Perform specified row selection to minimize I/O wait time.
v F9 - The query optimizer chose to use a hashing algorithm rather
than an index to perform the grouping.
v FA - The query contains a join condition that requires a temporary
table
v FB - The query optimizer creates a run-time temporary file in order
to implement certain correlated group by queries.
v FC - The query contains grouping fields and there is a read trigger
on at least one of the physical files in the query.
v H1 - Table is a join logical file and its join type does not match the
join type specified in the query.
v H2 - Format specified for the logical table references more than
one base table.
v H3 - Table is a complex SQL view requiring a temporary table to
contain the the results of the SQL view.
v H4 - For an update-capable query, a subselect references a
column in this table which matches one of the columns being
updated.
v H5 - For an update-capable query, a subselect references an SQL
view which is based on the table being updated.
v H6 - For a delete-capable query, a subselect references either the
table from which rows are to be deleted, an SQL view, or an index
based on the table from which rows are to be deleted
QVQTBL QVQTBL Queried table, long name
QVQLIB QVQLIB Library of queried table, long name
QVPTBL QVPTBL Base table, long name
QVPLIB QVPLIB Library of base table, long name
QVTTBLN QQC101 Temporary table name
QVTTBLL QQC102 Temporary table library
QVBNDY QVBNDY I/O or CPU bound. Possible values are:
v I - I/O bound
v C - CPU bound

Appendix A. Database monitor: DDS 171

Table 18. QQQ3004 - Summary Row for Temp Table (continued)
Logical Column Physical Column
Name Name Description
QVRCNT QVRCNT Unique refresh counter
QVJFANO QVJFANO Join fan out. Possible values are:
v N - Normal join situation where fanout is allowed and each
matching row of the join fanout is returned.
v D - Distinct fanout. Join fanout is allowed however none of the join
fanout rows are returned.
v U - Unique fanout. Join fanout is not allowed. Error situation if join
fanout occurs.
QVFILES QVFILES Number of tables joined
QVPARPF QVPARPF Parallel Prefetch (Y/N)
QVPARPL QVPARPL Parallel Preload (Y/N)
QVPARD QVPARD Parallel degree requested
QVPARU QVPARU Parallel degree used
QVPARRC QVPARRC Reason parallel processing was limited
QQEPT QQEPT Estimated processing time, in seconds
QVCTIM QVCTIM Estimated cumulative time, in seconds
QQAJN QQAJN Estimated number of joined rows
QQJNP QQJNP Join position - when available
QQJNDS QQI6 dataspace number
QQJNMT QQC21 Join method - when available
v NL - Nested loop
v MF - Nested loop with selection
v HJ - Hash join
QQJNTY QQC22 Join type - when available
v IN - Inner join
v PO - Left partial outer join
v EX - Exception join
QQJNOP QQC23 Join operator - when available
v EQ - Equal
v NE - Not equal
v GT - Greater than
v GE - Greater than or equal
v LT - Less than
v LE - Less than or equal
v CP - Cartesian product
QVTTRSZ QQI2 Row size of temporary table, in bytes
QVTTSIZ QQI3 Size of temporary table, in bytes
QVTTRST QQC12 Temporary result table that contains the results of the query. (Y/N)
QVTTDST QQC13 Distributed Table (Y/N)
QVTTNOD QVC3001 Data nodes of temporary table

172 DB2 UDB for iSeries Database Performance and Query Optimization V5R1
Database monitor logical table 3005 - Summary Row for Table Locked
|...+....1....+....2....+....3....+....4....+....5....+....6....+....7....+....8
A*
A* DB Monitor logical table 3005 - Summary Row for Table Locked
A*
A R QQQ3005 PTABLE(*CURLIB/QAQQDBMN)
A QQRID
A QQTIME
A QQJFLD
A QQRDBN
A QQSYS
A QQJOB
A QQUSER
A QQJNUM
A QQTHRD RENAME(QQI9) +
COLHDG(’Thread’ +
’Identifier’)
A QQUCNT
A QQUDEF
A QQQDTN
A QQQDTL
A QQMATN
A QQMATL
A QDQDTN RENAME(QVP15A) +
COLHDG(’Decomposed’ +
’Subselect’ +
’Number’)
A QDQDTT RENAME(QVP15B) +
COLHDG(’Number of’ +
’Decomposed’ +
’Subselects’)
A QDQDTR RENAME(QVP15C) +
COLHDG(’Decomposed’ +
’Reason’ +
’Code’)
A QDQDTS RENAME(QVP15D) +
COLHDG(’Starting’ +
’Decomposed’ +
’Subselect’)
A QQTLN
A QQTFN
A QQTMN
A QQPTLN
A QQPTFN
A QQPTMN
A QQLCKF RENAME(QQC11) +
COLHDG(’Lock’ +
’Indicator’)
A QQULCK RENAME(QQC12) +
COLHDG(’Unlock’ +
’Request’)
A QQRCOD
A*
A* Columns added for Visual Explain
A*
A QVQTBL
A QVQLIB
A QVPTBL
A QVPLIB
A QQJNP
A QQJNDS RENAME(QQI6) +
COLHDG(’Data Space’ +
’Number’)
A QQJNMT RENAME(QQC21) +
COLHDG(’Join’ ’Method’)
A QQJNTY RENAME(QQC22) +

Appendix A. Database monitor: DDS 173

 COLHDG(’Join’ ’Type’)
A QQJNOP RENAME(QQC23) +
COLHDG(’Join’ ’Operator’)
A QVJFANO
A QVFILES
A QVRCNT
A K QQJFLD
A S QQRID CMP(EQ 3005)
Table 19. QQQ3005 - Summary Row for Table Locked
Logical Column Physical Column
Name Name Description
QQRID QQRID Row identification
QQTIME QQTIME Time row was created
QQJFLD QQJFLD Join column (unique per job)
QQRDBN QQRDBN Relational database name
QQSYS QQSYS System name
QQJOB QQJOB Job name
QQUSER QQUSER Job user
QQJNUM QQJNUM Job number
QQTHRD QQI9 Thread identifier
QQUCNT QQUCNT Unique count (unique per query)
QQUDEF QQUDEF User defined column
QQQDTN QQQDTN Unique subselect number
QQQDTL QQQDTL Subselect nested level
QQMATN QQMATN Materialized view subselect number
QQMATL QQMATL Materialized view nested level
QDQDTN QVP15A Decomposed query subselect number, unique across all
decomposed subselects
QDQDTT QVP15B Total number of decomposed subselects
QDQDTR QVP15C Decomposed query subselect reason code
QDQDTS QVP15D Decomposed query subselect number for the first decomposed
subselect
QQTLN QQTLN Library of table queried
QQTFN QQTFN Name of table queried
QQTMN QQTMN Member name of table queried
QQPTLN QQPTLN Library name of base table
QQPTFN QQPTFN Name of base table for table queried
QQPTMN QQPTMN Member name of base table
QQLCKF QQC11 Successful lock indicator
v Y - Yes
v N - No
QQULCK QQC12 Unlock request
v Y - Yes
v N - No

174 DB2 UDB for iSeries Database Performance and Query Optimization V5R1
Table 19. QQQ3005 - Summary Row for Table Locked (continued)
Logical Column Physical Column
Name Name Description
QQRCOD QQRCOD Reason code
v L1 - UNION with *ALL or *CS with Keep Locks
v L2 - DISTINCT with *ALL or *CS with Keep Locks
v L3 - No duplicate keys with *ALL or *CS with Keep Locks
v L4 - Temporary needed with *ALL or *CS with Keep Locks
v L5 - System Table with *ALL or *CS with Keep Locks
v L6 - Orderby > 2000 bytes with *ALL or *CS with Keep Locks
v L9 - Unknown
QVQTBL QVQTBL Queried table, long name
QVQLIB QVQLIB Library of queried table, long name
QVPTBL QVPTBL Base table, long name
QVPLIB QVPLIB Library of base table, long name
QQJNP QQJNP Join position - when available
QQJNDS QQI6 dataspace number
QQJNMT QQC21 Join method - when available
v NL - Nested loop
v MF - Nested loop with selection
v HJ - Hash join
QQJNTY QQC22 Join type - when available
v IN - Inner join
v PO - Left partial outer join
v EX - Exception join
QQJNOP QQC23 Join operator - when available
v EQ - Equal
v NE - Not equal
v GT - Greater than
v GE - Greater than or equal
v LT - Less than
v LE - Less than or equal
v CP - Cartesian product
QVJFANO QVJFANO Join fan out. Possible values are:
v N - Normal join situation where fanout is allowed and each
matching row of the join fanout is returned.
v D - Distinct fanout. Join fanout is allowed however none of the join
fanout rows are returned.
v U - Unique fanout. Join fanout is not allowed. Error situation if join
fanout occurs.
QVFILES QVFILES Number of tables joined
QVRCNT QVRCNT Unique refresh counter

Appendix A. Database monitor: DDS 175

Database monitor logical table 3006 - Summary Row for Access Plan
Rebuilt
|...+....1....+....2....+....3....+....4....+....5....+....6....+....7....+....8
A*
A* DB Monitor logical table 3006 - Summary Row for Access Plan Rebuilt
A*
A R QQQ3006 PTABLE(*CURLIB/QAQQDBMN)
A QQRID
A QQTIME
A QQJFLD
A QQRDBN
A QQSYS
A QQJOB
A QQUSER
A QQJNUM
A QQTHRD RENAME(QQI9) +
COLHDG(’Thread’ +
’Identifier’)
A QQUCNT
A QQUDEF
A QQQDTN
A QQQDTL
A QQMATN
A QQMATL
A QDQDTN RENAME(QVP15A) +
COLHDG(’Decomposed’ +
’Subselect’ +
’Number’)
A QDQDTT RENAME(QVP15B) +
COLHDG(’Number of’ +
’Decomposed’ +
’Subselects’)
A QDQDTR RENAME(QVP15C) +
COLHDG(’Decomposed’ +
’Reason’ +
’Code’)
A QDQDTS RENAME(QVP15D) +
COLHDG(’Starting’ +
’Decomposed’ +
’Subselect’)
A QQINLN
A QQINFN
A QQRCOD
A*
A* Columns added for Visual Explain
A*
A QVSUBRC RENAME(QQC21) +
COLHDG(’Subtype’ +
’Reason’ +
’Code’)
A QVRCNT
A QVRPTS RENAME(QQTIM1) +
COLHDG(’Timestamp’ +
’Last’ +
’Rebuild’)
A K QQJFLD
A S QQRID CMP(EQ 3006)

Table 20. QQQ3006 - Summary Row for Access Plan Rebuilt

Logical Physical Column
Column Name Name Description
QQRID QQRID Row identification
QQTIME QQTIME Time row was created

176 DB2 UDB for iSeries Database Performance and Query Optimization V5R1
Table 20. QQQ3006 - Summary Row for Access Plan Rebuilt (continued)
Logical Physical Column
Column Name Name Description
QQJFLD QQJFLD Join column (unique per job)
QQRDBN QQRDBN Relational database name
QQSYS QQSYS System name
QQJOB QQJOB Job name
QQUSER QQUSER Job user
QQJNUM QQJNUM Job number
QQTHRD QQI9 Thread identifier
QQUCNT QQUCNT Unique count (unique per query)
QQUDEF QQUDEF User defined column
QQQDTN QQQDTN Unique subselect number
QQQDTL QQQDTL Subselect nested level
QQMATN QQMATN Materialized view subselect number
QQMATL QQMATL Materialized view nested level
QDQDTN QVP15A Decomposed query subselect number, unique across all decomposed subselects
QDQDTT QVP15B Total number of decomposed subselects
QDQDTR QVP15C Decomposed query subselect reason code
QDQDTS QVP15D Decomposed query subselect number for the first decomposed subselect

Appendix A. Database monitor: DDS 177

Table 20. QQQ3006 - Summary Row for Access Plan Rebuilt (continued)
Logical Physical Column
Column Name Name Description
QQRCOD QQRCOD Reason code why access plan was rebuilt
v A1 - A table or member is not the same object as the one referenced when
the access plan was last built. Some reasons they could be different are:
– Object was deleted and recreated.
– Object was saved and restored.
– Library list was changed.
– Object was renamed.
– Object was moved.
– Object was overridden to a different object.
– This is the first run of this query after the object containing the query has
been restored.
v A2 - Access plan was built to use a reusable Open Data Path (ODP) and the
optimizer chose to use a non-reusable ODP for this call.
v A3 - Access plan was built to use a non-reusable Open Data Path (ODP) and
the optimizer chose to use a reusable ODP for this call.
v A4 - The number of rows in the table has changed by more than 10% since
the access plan was last built.
v A5 - A new index exists over one of the tables in the query
v A6 - An index that was used for this access plan no longer exists or is no
longer valid.
v A7 - OS/400 Query requires the access plan to be rebuilt because of system
programming changes.
v A8 - The CCSID of the current job is different than the CCSID of the job that
last created the access plan.
v A9 - The value of one or more of the following is different for the current job
than it was for the job that last created this access plan:
– date format
– date separator
– time format
– time separator.
v AA - The sort sequence table specified is different than the sort sequence
table that was used when this access plan was created.
v AB - Storage pool changed or DEGREE parameter of CHGQRYA command
changed.
v AC - The system feature DB2 multisystem has been installed or removed.
v AD - The value of the degree query attribute has changed.
v AE - A view is either being opened by a high level language or a view is being
materialized.
v AF - A user-defined type or user-defined function is not the same object as the
one referred to in the access plan.
v B0 - The options specified have changed as a result of the query options file.
v B1 - The access plan was generated with a commitment control level that is
different in the current job.
v B2 - The access plan was generated with a static cursor answer set size that
is different than the previous access plan.

178 DB2 UDB for iSeries Database Performance and Query Optimization V5R1
Table 20. QQQ3006 - Summary Row for Access Plan Rebuilt (continued)
Logical Physical Column
Column Name Name Description
QVSUBRC QQC21 If the access plan rebuild reason code was A7 this reason code identifies which
specific reason for A7 forced a rebuild. This value is a 2-byte hex value. The
following values force a reoptimization.
v X’0001’ - Update specified and must check at run time or user forced
reoptimizing.
v X’0002’ - If the query uses a select/omit index and also uses host variable
values processed with V2 then we must reoptimize.
v X’0003’ - If the user intends to insert, update, or delete through a join cursor,
make sure that the first table specified is the first dial in the join cursor. This
check was not done when the access plan was created.
v X’0004’ - If the access plan indicates the query sort is to be used but the run
time options do not allow it, then reoptimize.
v X’0005’ - Check to see if a fast sort is being attempted but one isn’t allowed. If
so reoptimize so the fast sort isn’t used.
v X’0006’ - Need to reoptimize the query when the QDT requests the output to
be forced to a temporary and the access plan does not already do this.
v X’0007’ - Reoptimize if we have a change in commitiment control level.
v X’0008’ - Reoptimize if we have a change in key feedback.
v X’0009’ - Reoptimize if pool size changed.
v X’000A’ - Reoptimize SMP PRODUCT status has changed.
v X’000B’ - Reoptimize if SMP parallelism or I/O parallelism is being used but
the DEGREE query attribute is now *NONE.
v X’000C’ - Reoptimize if SMP parallel is being used but the DEGREE query
attribute is now *IO.
v X’000D’ - Reoptimize if access plan was created for a view or updated during
the direct open of a view, and it is now being materialized, or visa versa.
v X’000E’ - Reoptimize if SRVQRY is currently active and/or the access plan
was created with SRVQRY active.
v X’000F’ - Reoptimize if it is a unique query, but table scan is used for the first
dial.
v X’0010’ - Reoptimize if the query is set for hash join, but the last dial is not a
hash join dial.
v X’0011’ - Reoptimize if access plan was created using index AND/OR and now
index AND/OR is not allowed.
v X’0012’ - Reoptimize if an index used in the access plan is no longer usable.
v X’0013’ - Reoptimize if one, or more, of the index AND/OR indexes used in
the access plan is no longer usable.
v X’0014’ - Reoptimize if the number rows is greater than 1000 and number of
rows has changed by 10% (+ or -).
v X’0015’ - Reoptimize if a new permanent index has been created since the
access plan was created.
v X’0016’ - Reoptimize if an index that was invalid when the access plan was
created is now valid.
v X’0017’ and higher (excluding values starting with X’80xx’) PTF changes that
require the access plan to be reoptimized.

Appendix A. Database monitor: DDS 179

Table 20. QQQ3006 - Summary Row for Access Plan Rebuilt (continued)
Logical Physical Column
Column Name Name Description

v X’8001’ - SQ requests a forced rebuild.

v X’8002’ - Revalidate if need to perform CHAR to ZONED conversions.
v X’8003’ - Revalidate if the query uses XLAT operators with CCSID conversion.
v X’8004’ - Revalidate if query uses subtraction or addition.
v X’8005’ - Revalidate if query uses maximum or minimum.
v X’8006’ - Revalidate if query uses at least one table that specifies NULL
values to be returned.
v X’8007’ - Revalidate if query forced a revalidation.
v X’8008’ - Revalidate if the job default format separators for DATE or TIME
columns have changed since the access plan was created.
v X’8009’ - Revalidate if default CCSID or final mapping CCSID is different now
than when the access plan was created.
v X’800A’ - Revalidate if the current NLSS is different than the one saved in the
access plan.
v X’800B’ - Revalidate if the SRTSEQ passed in is different that the SRTSEQ
used when the access plan was created.
v X’800C’ - Revalidate if divide by zero preference has changed.
v X’800D’ - Revalidate if query uses a DDM table.
v X’800E’ - Revalidate if tables used in the access plan have been overridden to
different tables or members.
v X’800F’ - Revalidate if table, member or dataspace objects have changed from
when the access plan was created.
v X’8010’ - Revalidate if the format level identifiers of the tables do not match
thosed saved in the access plan.
v X’8011’ - Revalidate if CCSIDs of the tables have changed.
v X’8012’ and higher - PTF changes that require the access plan to be rebuilt.
QVRCNT QVRCNT Unique refresh counter
QVRPTS QQTIM1 Timestamp of last access plan rebuild

Database monitor logical table 3007 - Summary Row for Optimizer

Timed Out
|...+....1....+....2....+....3....+....4....+....5....+....6....+....7....+....8
A*
A* DB Monitor logical table 3007 - Summary Row for Optimizer Timed Out
A*
A R QQQ3007 PTABLE(*CURLIB/QAQQDBMN)
A QQRID
A QQTIME
A QQJFLD
A QQRDBN
A QQSYS
A QQJOB
A QQUSER
A QQJNUM
A QQTHRD RENAME(QQI9) +
COLHDG(’Thread’ +
’Identifier’)
A QQUCNT
A QQUDEF
A QQQDTN
A QQQDTL

180 DB2 UDB for iSeries Database Performance and Query Optimization V5R1
 A QQMATN
A QQMATL
A QDQDTN RENAME(QVP15A) +
COLHDG(’Decomposed’ +
’Subselect’ +
’Number’)
A QDQDTT RENAME(QVP15B) +
COLHDG(’Number of’ +
’Decomposed’ +
’Subselects’)
A QDQDTR RENAME(QVP15C) +
COLHDG(’Decomposed’ +
’Reason’ +
’Code’)
A QDQDTS RENAME(QVP15D) +
COLHDG(’Starting’ +
’Decomposed’ +
’Subselect’)
A QQTLN
A QQTFN
A QQTMN
A QQPTLN
A QQPTFN
A QQPTMN
A QQIDXN RENAME(QQ1000) +
COLHDG(’Index’ +
’Names’)
A QQTOUT RENAME(QQC11) +
COLHDG(’Optimizer’ +
’Timed Out’)
A QQISRN RENAME(QQC301) +
COLHDG(’Index’ +
’Reason’ +
’Codes’)
A*
A* Columns added for Visual Explain
A*
A QVQTBL
A QVQLIB
A QVPTBL
A QVPLIB
A QQJNP
A QQJNDS RENAME(QQI6) +
COLHDG(’Data Space’ +
’Number’)
A QQJNMT RENAME(QQC21) +
COLHDG(’Join’ ’Method’)
A QQJNTY RENAME(QQC22) +
COLHDG(’Join’ ’Type’)
A QQJNOP RENAME(QQC23) +
COLHDG(’Join’ ’Operator’)
A QVJFANO
A QVFILES
A QVRCNT
A K QQJFLD
A S QQRID CMP(EQ 3007)

Table 21. QQQ3007 - Summary Row for Optimizer Timed Out

Logical Column Physical Column
Name Name Description
QQRID QQRID Row identification
QQTIME QQTIME Time row was created
QQJFLD QQJFLD Join column (unique per job)
QQRDBN QQRDBN Relational database name

Appendix A. Database monitor: DDS 181

Table 21. QQQ3007 - Summary Row for Optimizer Timed Out (continued)
Logical Column Physical Column
Name Name Description
QQSYS QQSYS System name
QQJOB QQJOB Job name
QQUSER QQUSER Job user
QQJNUM QQJNUM Job number
QQTHRD QQI9 Thread identifier
QQUCNT QQUCNT Unique count (unique per query)
QQUDEF QQUDEF User defined column
QQQDTN QQQDTN Unique subselect number
QQQDTL QQQDTL Subselect nested level
QQMATN QQMATN Materialized view subselect number
QQMATL QQMATL Materialized view nested level
QDQDTN QVP15A Decomposed query subselect number, unique across all
decomposed subselects
QDQDTT QVP15B Total number of decomposed subselects
QDQDTR QVP15C Decomposed query subselect reason code
QDQDTS QVP15D Decomposed query subselect number for the first decomposed
subselect
QQTLN QQTLN Library of table queried
QQTFN QQTFN Name of table queried
QQTMN QQTMN Member name of table queried
QQPTLN QQPTLN Library name of base table
QQPTFN QQPTFN Name of base table for table queried
QQPTMN QQPTMN Member name of base table
QQIDXN QQ1000 Index names
QQTOUT QQC11 Optimizer timed out
v Y - Yes
v N - No
QQISRN QQC301 List of unique reason codes used by the indexes that timed out (each
index has a corresponding reason code associated with it)
QVQTBL QVQTBL Queried table, long name
QVQLIB QVQLIB Library of queried table, long name
QVPTBL QVPTBL Base table, long name
QVPLIB QVPLIB Library of base table, long name
QQJNP QQJNP Join position - when available
QQJNDS QQI6 dataspace number
QQJNMT QQC21 Join method - when available
v NL - Nested loop
v MF - Nested loop with selection
v HJ - Hash join

182 DB2 UDB for iSeries Database Performance and Query Optimization V5R1
Table 21. QQQ3007 - Summary Row for Optimizer Timed Out (continued)
Logical Column Physical Column
Name Name Description
QQJNTY QQC22 Join type - when available
v IN - Inner join
v PO - Left partial outer join
v EX - Exception join
QQJNOP QQC23 Join operator - when available
v EQ - Equal
v NE - Not equal
v GT - Greater than
v GE - Greater than or equal
v LT - Less than
v LE - Less than or equal
v CP - Cartesian product
QVJFANO QVJFANO Join fan out. Possible values are:
v N - Normal join situation where fanout is allowed and each
matching row of the join fanout is returned.
v D - Distinct fanout. Join fanout is allowed however none of the join
fanout rows are returned.
v U - Unique fanout. Join fanout is not allowed. Error situation if join
fanout occurs.
QVFILES QVFILES Number of tables joined
QVRCNT QVRCNT Unique refresh counter

Database monitor logical table 3008 - Summary Row for Subquery

Processing
|...+....1....+....2....+....3....+....4....+....5....+....6....+....7....+....8
A*
A* DB Monitor logical table 3008 - Summary Row for Subquery Processing
A*
A R QQQ3008 PTABLE(*CURLIB/QAQQDBMN)
A QQRID
A QQTIME
A QQJFLD
A QQRDBN
A QQSYS
A QQJOB
A QQUSER
A QQJNUM
A QQTHRD RENAME(QQI9) +
COLHDG(’Thread’ +
’Identifier’)
A QQUCNT
A QQUDEF
A QQQDTN
A QQQDTL
A QQMATN
A QQMATL
A QDQDTN RENAME(QVP15A) +
COLHDG(’Decomposed’ +
’Subselect’ +
’Number’)
A QDQDTT RENAME(QVP15B) +
COLHDG(’Number of’ +

Appendix A. Database monitor: DDS 183

 ’Decomposed’ +
’Subselects’)
A QDQDTR RENAME(QVP15C) +
COLHDG(’Decomposed’ +
’Reason’ +
’Code’)
A QDQDTS RENAME(QVP15D) +
COLHDG(’Starting’ +
’Decomposed’ +
’Subselect’)
A QQORGQ RENAME(QQI1) +
COLHDG(’Original’ +
’Number’ +
’of QDTs’)
A QQMRGQ RENAME(QQI2) +
COLHDG(’Number’ +
’of QDTs’ +
’Merged’)
A*
A* Columns added for Visual Explain
A*
A QQFNLQ RENAME(QQI3) +
COLHDG(’Final’ +
’Number’ +
’of QDTs’)
A QVRCNT
A K QQJFLD
A S QQRID CMP(EQ 3008)

Table 22. QQQ3008 - Summary Row for Subquery Processing

Logical Column Physical Column
Name Name Description
QQRID QQRID Row identification
QQTIME QQTIME Time row was created
QQJFLD QQJFLD Join column (unique per job)
QQRDBN QQRDBN Relational database name
QQSYS QQSYS System name
QQJOB QQJOB Job name
QQUSER QQUSER Job user
QQJNUM QQJNUM Job number
QQTHRD QQI9 Thread identifier
QQUCNT QQUCNT Unique count (unique per query)
QQUDEF QQUDEF User defined column
QQQDTN QQQDTN Unique subselect number
QQQDTL QQQDTL Subselect nested level
QQMATN QQMATN Materialized view subselect number
QQMATL QQMATL Materialized view nested level
QDQDTN QVP15A Decomposed query subselect number, unique across all
decomposed subselects
QDQDTT QVP15B Total number of decomposed subselects
QDQDTR QVP15C Decomposed query subselect reason code
QDQDTS QVP15D Decomposed query subselect number for the first decomposed
subselect
QQORGQ QQI1 Original number of QDTs

184 DB2 UDB for iSeries Database Performance and Query Optimization V5R1
Table 22. QQQ3008 - Summary Row for Subquery Processing (continued)
Logical Column Physical Column
Name Name Description
QQMRGQ QQI2 Number of QDTs merged
QQFNLQ QQI3 Final number of QDTs
QVRCNT QVRCNT Unique refresh counter

Database monitor logical table 3010 - Summary for HostVar & ODP
Implementation
|...+....1....+....2....+....3....+....4....+....5....+....6....+....7....+....8
A*
A* DB Monitor logical table 3010 - Summary for HostVar & ODP Implementation
A*
A R QQQ3010 PTABLE(*CURLIB/QAQQDBMN)
A QQRID
A QQTIME
A QQJFLD
A QQRDBN
A QQSYS
A QQJOB
A QQUSER
A QQJNUM
A QQTHRD RENAME(QQI9) +
COLHDG(’Thread’ +
’Identifier’)
A QQUCNT
A QQRCNT RENAME(QQI5) +
COLHDG(’Refresh’ +
’Counter’)
A QQUDEF
A QQODPI RENAME(QQC11) +
COLHDG(’ODP’ +
’Implementation’)
A QQHVI RENAME(QQC12) +
COLHDG(’Host Variable’ +
’Implementation’)
A QQHVAR RENAME(QQ1000) +
COLHDG(’Host Variable’ +
’Values’)
A K QQJFLD
A S QQRID CMP(EQ 3010)

Table 23. QQQ3010 - Summary for HostVar & ODP Implementation

Logical Column Physical Column
Name Name Description
QQRID QQRID Row identification
QQTIME QQTIME Time row was created
QQJFLD QQJFLD Join column (unique per job)
QQRDBN QQRDBN Relational database name
QQSYS QQSYS System name
QQJOB QQJOB Job name
QQUSER QQUSER Job user
QQJNUM QQJNUM Job number
QQTHRD QQI9 Thread identifier
QQUCNT QQUCNT Unique count (unique per query)

Appendix A. Database monitor: DDS 185

Table 23. QQQ3010 - Summary for HostVar & ODP Implementation (continued)
Logical Column Physical Column
Name Name Description
QQRCNT QQI5 Unique refresh counter
QQUDEF QQUDEF User defined column
QQODPI QQC11 ODP implementation
v R - Reusable ODP
v N - Nonreusable ODP
v ’ ’ - Column not used
QQHVI QQC12 Host variable implementation
v I - Interface supplied values (ISV)
v V - Host variables treated as constants (V2)
v U - Table management row positioning (UP)
QQHVAR QQ1000 Host variable values

Database monitor logical table 3014 - Summary Row for Generic QQ

Information
|...+....1....+....2....+....3....+....4....+....5....+....6....+....7....+....8
A*
A* DB Monitor logical table 3014 - Summary Row for Generic QQ Information
A*
A R QQQ3014 PTABLE(*CURLIB/QAQQDBMN)
A QQRID
A QQTIME
A QQJFLD
A QQRDBN
A QQSYS
A QQJOB
A QQUSER
A QQJNUM
A QQTHRD RENAME(QQI9) +
COLHDG(’Thread’ +
’Identifier’)
A QQUCNT
A QQUDEF
A QQQDTN
A QQQDTL
A QQMATN
A QQMATL
A QDQDTN RENAME(QVP15A) +
COLHDG(’Decomposed’ +
’Subselect’ +
’Number’)
A QDQDTT RENAME(QVP15B) +
COLHDG(’Number of’ +
’Decomposed’ +
’Subselects’)
A QDQDTR RENAME(QVP15C) +
COLHDG(’Decomposed’ +
’Reason’ +
’Code’)
A QDQDTS RENAME(QVP15D) +
COLHDG(’Starting’ +
’Decomposed’ +
’Subselect’)
A QQREST
A QQEPT
A QQQTIM RENAME(QQI1) +

186 DB2 UDB for iSeries Database Performance and Query Optimization V5R1
 COLHDG(’ODP’ +
’Open’ ’Time’)
A QQORDG
A QQGRPG
A QQJNG
A QQJNTY RENAME(QQC22) +
COLHDG(’Join’ +
’Type’)
A QQUNIN
A QQSUBQ
A QQHSTV
A QQRCDS
A QQGVNE RENAME(QQC11) +
COLHDG(’Query’ +
’Governor’ +
’Enabled’)
A QQGVNS RENAME(QQC12) +
COLHDG(’Stopped’ +
’by Query’ +
’Governor’)
A QQOPID RENAME(QQC101) +
COLHDG(’Query’ +
’Open ID’)
A QQINLN RENAME(QQC102) +
COLHDG(’Query’ +
’Options’ +
’Library’)
A QQINFN RENAME(QQC103) +
COLHDG(’Query’ +
’Options’ +
’File’)
A QQEE RENAME(QQC13) +
COLHDG(’Early’ +
’Exit’ +
’Indicator’)
A*
A* Columns added for Visual Explain
A*
A QVRCNT
A QVOPTIM RENAME(QQI5) +
COLHDG(’Optimization’ +
’Time’)
A QVAPRT RENAME(QQTIM1) +
COLHDG(’Access Plan’ +
’Rebuild’
’Timestamp’)
A QVOBYIM RENAME(QVC11) +
COLHDG(’Ordering’ +
’Implementation’)
A QVGBYIM RENAME(QVC12) +
COLHDG(’Grouping’ +
’Implementation’)
A QVJONIM RENAME(QVC13) +
COLHDG(’Join’ +
’Implementation’)
A QVDIST RENAME(QVC14) +
COLHDG(’Distinct’ +
’Query’)
A QVDSTRB RENAME(QVC15) +
COLHDG(’Distributed’ +
’Query’)
A QVDSTND RENAME(QVC3001) +
COLHDG(’Distributed’ +
’Nodes’)
A QVNLSST RENAME(QVC105) +
COLHDG(’Sort’ +
’Sequence’ +

Appendix A. Database monitor: DDS 187

 ’Table’)
A QVNLSSL RENAME(QVC106) +
COLHDG(’Sort’ +
’Sequence’ +
’Library’)
A QVALWCY RENAME(QVC16) +
COLHDG(’ALWCPYDTA’ +
’Setting’)
A QVVAPRC RENAME(QVC21) +
COLHDG(’Access Plan’ +
’Rebuilt’ +
’Code’)
A QVVAPSC RENAME(QVC22) +
COLHDG(’Access Plan’ +
’Rebuilt’ +
’Subcode’)
A QVIMPLN RENAME(QVC3002) +
COLHDG(’Implementation’ +
’Summary’)
A QVUNIONL RENAME(QWC16) +
COLHDG(’Last’ +
’Part of’ +
’Union’)
A DCMPFNLBLT RENAME(QQC14) +
COLHDG(’Decomposed’ +
’Final Cursor’ +
’was Built’)
A DCMPFNLTMP RENAME(QQC15) +
COLHDG(’This is’ +
’Decomposed’ +
’Final Cursor’)
A QQPSIZ RENAME(QVP154) +
COLHDG(’Pool’ +
’Size’)
A QQPID RENAME(QVP155) +
COLHDG(’Pool’ +
’ID’)
A*
A* CHGQRYA or INI environment attributes used during execution of query
A*
A QVMAXT RENAME(QQI2) +
COLHDG(’Query’ +
’Time’ +
’Limit’)
A QVPARA RENAME(QVC81) +
COLHDG(’Specified’ +
’Parallel’ +
’Option’)
A QVTASKN RENAME(QQI3) +
COLHDG(’Mamimum’ +
’Number of’ +
’Tasks’)
A QVAPLYR RENAME(QVC17) +
COLHDG(’Apply’ +
’CHGQRYA’ +
’Remotely’)
A QVASYNC RENAME(QVC82) +
COLHDG(’Asynchronous’ +
’Remote’ +
’Job Usage’)
A QVFRCJO RENAME(QVC18) +
COLHDG(’Join’ +
’Order’ +
’Forced’)
A QVDMSGS RENAME(QVC19) +
COLHDG(’Display’ +
’DEBUG’ +

188 DB2 UDB for iSeries Database Performance and Query Optimization V5R1
 ’Messages’)
A QVPMCNV RENAME(QVC1A) +
COLHDG(’Parameter’ +
’Marker’ +
’Conversion’)
A QVUDFTL RENAME(QQI4) +
COLHDG(’UDF’ +
’Time’ +
’Limit’)
A QVOLMTS RENAME(QVC1283) +
COLHDG(’Query’ +
’Optimizer’ +
’Limitations’)
A QVSQIMP RENAME(QVC1B) +
COLHGD(’Subquery’ +
’Implementation’ +
’Requested’)
A QVHVIMP RENAME(QVC1C) +
COLHDG(’HostVar’ +
’Implementation’ +
’Requested’)
A QVORIMP RENAME(QVC84) +
COLHGD(’Ordering’ +
’Implementation’ +
’Requested’)
A QVGPIMP RENAME(QVC85) +
COLHDG(’Grouping’ +
’Implementation’ +
’Requested’)
A QVJNIMP RENAME(QVC86) +
COLHDG(’Join’ +
’Implementation’ +
’Requested’)
A QVREVAL RENAME(QVC1D) +
COLHGD(’Rebuild’ +
’Access’ ’Plan’)
A QVREOPT RENAME(QVC1E) +
COLHDG(’Reoptimize’ +
’Access’ ’Plan’)
A QVOPALL RENAME(QVC87) +
COLHGD(’Optimize’ +
’All’ ’Indexes’)
A K QQJFLD
A S QQRID CMP(EQ 3014)

Table 24. QQQ3014 - Summary Row for Generic QQ Information

Logical Column Physical Column
Name Name Description
QQRID QQRID Row identification
QQTIME QQTIME Time row was created
QQJFLD QQJFLD Join column (unique per job)
QQRDBN QQRDBN Relational database name
QQSYS QQSYS System name
QQJOB QQJOB Job name
QQUSER QQUSER Job user
QQJNUM QQJNUM Job number
QQTHRD QQI9 Thread identifier
QQUCNT QQUCNT Unique count (unique per query)
QQUDEF QQUDEF User defined column

Appendix A. Database monitor: DDS 189

Table 24. QQQ3014 - Summary Row for Generic QQ Information (continued)
Logical Column Physical Column
Name Name Description
QQQDTN QQQDTN Unique subselect number
QQQDTL QQQDTL Subselect nested level
QQMATN QQMATN Materialized view subselect number
QQMATL QQMATL Materialized view nested level
QDQDTN QVP15A Decomposed query subselect number, unique across all
decomposed subselects
QDQDTT QVP15B Total number of decomposed subselects
QDQDTR QVP15C Decomposed query subselect reason code
QDQDTS QVP15D Decomposed query subselect number for the first decomposed
subselect
QQREST QQREST Estimated number of rows selected
QQEPT QQEPT Estimated processing time, in seconds
QQQTIM QQI1 Time spent to open cursor, in milliseconds
QQORDG QQORDG Ordering (Y/N)
QQGRPG QQGRPG Grouping (Y/N)
QQJNG QQJNG Join Query (Y/N)
QQJNTY QQC22 Join type - when available
v IN - Inner join
v PO - Left partial outer join
v EX - Exception join
QQUNIN QQUNIN Union Query (Y/N)
QQSUBQ QQSUBQ Subquery (Y/N)
QQHSTV QQHSTV Host variables (Y/N)
QQRCDS QQRCDS Row selection (Y/N)
QQGVNE QQC11 Query governor enabled (Y/N)
QQGVNS QQC12 Query governor stopped the query (Y/N)
QQOPID QQC101 Query open ID
QVINLN QQC102 Query Options library name
QVINFN QQC103 Query Options file name
QQEE QQC13 Query early exit value
QVRCNT QVRCNT Unique refresh counter
QVOPTIM QQI5 Time spent in optimizer, in milliseconds
QVAPRT QQTIM1 Access Plan rebuilt timestamp, last time access plan was rebuilt.
QVOBYIM QVC11 Ordering implementation. Possible values are:
v I - Index
v S - Sort
QVGBYIM QVC12 Grouping implementation. Possible values are:
v I - Index
v H - Hash grouping

190 DB2 UDB for iSeries Database Performance and Query Optimization V5R1
Table 24. QQQ3014 - Summary Row for Generic QQ Information (continued)
Logical Column Physical Column
Name Name Description
QVJONIM QVC13 Join Implementation. Possible values are:
v N - Nested Loop join
v H - Hash join
v C - Combination of Nested Loop and Hash
QVDIST QVC14 Distinct query (Y/N)
QVDSTRB QVC15 Distributed query (Y/N)
QVDSTND QVC3001 Distributed nodes
QVNLSST QVC105 Sort Sequence Table
QVNLSSL QVC106 Sort Sequence Library
QVALWC QVC16 ALWCPYDTA setting
QVVAPRC QVC21 Reason code why access plan was rebuilt
QVVAPSC QVC22 Subcode why access plan was rebuilt
QVIMPLN QVC3002 Summary of query implementation. Shows dataspace number and
name of index used for each table being queried.
QVUNIONL QWC16 Last part (last QDT) of Union (Y/N)
DCMPFNLBLT QWC14 A decomposed final temporary cursor was built (Y/N)
DCMPFNLTMP QWC15 This is the decomposed final temporary cursor (final temporary
QDT). (Y/N)
QVMAXT QQI2 Query time limit
QVPARA QVC81 Parallel Degree
v *SAME - Don’t change current setting
v *NONE - No parallel processing is allowed
v *I/O - Any number of tasks may be used for I/O processing. SMP
parallel processing is not allowed.
v *OPTIMIZE - The optimizer chooses the number of tasks to use
for either I/O or SMP parallel processing.
v *MAX - The optimizer chooses to use either I/O or SMP parallel
processing.
v *SYSVAL - Use the current system value to process the query.
v *ANY - Has the same meaning as *I/O.
v *NBRTASKS - The number of tasks for SMP parallel processing is
specified in column QVTASKN.
QVTASKN QQI3 Max number of tasks
QVAPLYR QVC17 Apply CHGQRYA remotely (Y/N)
QVASYNC QVC82 Asynchronous job usage
v *SAME - Don’t change current setting
v *DIST - Asynchronous jobs may be used for queries with
distributed tables
v *LOCAL - Asynchronous jobs may be used for queries with local
tables only
v *ANY - Asynchronous jobs may be used for any database query
v *NONE - No asynchronous jobs are allowed
QVFRCJO QVC18 Force join order (Y/N)
QVDMSGS QVC19 Print debug messages (Y/N)

Appendix A. Database monitor: DDS 191

Table 24. QQQ3014 - Summary Row for Generic QQ Information (continued)
Logical Column Physical Column
Name Name Description
QVPMCNV QVC1A Parameter marker conversion (Y/N)
QVUDFTL QQI4 User Defined Function time limit
QVOLMTS QVC1281 Optimizer limitations. Possible values:
v *PERCENT followed by 2 byte integer containing the percent
value
v *MAX_NUMBER_OF_RECORDS followed by an integer value that
represents the maximum number of rows
QVSQIMP QVC1B Subquery implementation requested
v I - Use Interface Supplied Values (ISVs)
v J - Implement subquery as a join
QVHVIMP QVC1C Host Variable implementation requested
v I - Use Interface Supplied Values (ISVs)
v V - Convert host variables to constants
QVORIMP QVC84 Ordering implementation requested
v *SAME - Use current implementation
v *NOFRC - Allow optimizer to choose
v *SORT - Implement ordering using a sort
v *INDEX - Implement ordering using an index
QVGPIMP QVC85 Grouping implementation requested
v *SAME - Use current implementation
v *NOFRC - Allow optimizer to choose
v *HASH - Force use of a hash function
v *INDEX - Force use of an index
QVJNIMP QVC86 Join implementation requested
v *SAME - Use current implementation
v *NOFRC - Don’t any particular setting
v *NESTLOOP - Force Nested Loop join
v *HASHLOOP - Force Hash join implementation
QVREVAL QVC1D Rebuild access plan requested (Y/N)
QVREOPT QVC1E Reoptimize access plan requested (Y/N)
QVOPALL QVC87 Optimize all indexes requested
v *SAME - Don’t change current setting
v *YES - Examine all indexes
v *NO - Allow optimizer to time-out
v *TIMEOUT - Force optimizer to time-out

Database monitor logical table 3018 - Summary Row for

STRDBMON/ENDDBMON
|...+....1....+....2....+....3....+....4....+....5....+....6....+....7....+....8
A*
A* DB Monitor logical table 3018 - Summary Row for STRDBMON/ENDDBMON
A*
A R QQQ3018 PTABLE(*CURLIB/QAQQDBMN)
A QQRID
A QQTIME

192 DB2 UDB for iSeries Database Performance and Query Optimization V5R1
 A QQJFLD
A QQRDBN
A QQSYS
A QQJOB
A QQUSER
A QQJNUM
A QQTHRD RENAME(QQI9) +
COLHDG(’Thread’ +
’Identifier’)
A QQJOBT RENAME(QQC11)+
COLHDG(’Job’ +
’Type’)
A QQCMDT RENAME(QQC12) +
COLHDG(’Command’ +
’Type’)
A QQJOBI RENAME(QQC301) +
COLHDG(’Job’ +
’Info’)
A K QQJFLD
A S QQRID CMP(EQ 3018)
Table 25. QQQ3018 - Summary Row for STRDBMON/ENDDBMON
Logical Column Physical Column
Name Name Description
QQRID QQRID Row identification
QQTIME QQTIME Time row was created
QQJFLD QQJFLD Join column (unique per job)
QQRDBN QQRDBN Relational database name
QQSYS QQSYS System name
QQJOB QQJOB Job name
QQUSER QQUSER Job user
QQJNUM QQJNUM Job number
QQTHRD QQI9 Thread identifier
QQJOBT QQC11 Type of job monitored
v C - Current
v J - Job name
v A - All
QQCMDT QQC12 Command type
v S - STRDBMON
v E - ENDDBMON
QQJOBI QQC301 Monitored job information
v * - Current job
v Job number/User/Job name
v *ALL - All jobs

Database monitor logical table 3019 - Detail Row for Rows Retrieved
|...+....1....+....2....+....3....+....4....+....5....+....6....+....7....+....8
A*
A* DB Monitor logical table 3019 - Detail Row for Rows Retrieved
A*
A R QQQ3019 PTABLE(*CURLIB/QAQQDBMN)
A QQRID
A QQTIME
A QQJFLD

Appendix A. Database monitor: DDS 193

 A QQRDBN
A QQSYS
A QQJOB
A QQUSER
A QQJNUM
A QQTHRD RENAME(QQI9) +
COLHDG(’Thread’ +
’Identifier’)
A QQUCNT
A QQUDEF
A QQQDTN
A QQQDTL
A QQMATN
A QQMATL
A QDQDTN RENAME(QVP15A) +
COLHDG(’Decomposed’ +
’Subselect’ +
’Number’)
A QDQDTT RENAME(QVP15B) +
COLHDG(’Number of’ +
’Decomposed’ +
’Subselects’)
A QDQDTR RENAME(QVP15C) +
COLHDG(’Decomposed’ +
’Reason’ +
’Code’)
A QDQDTS RENAME(QVP15D) +
COLHDG(’Starting’ +
’Decomposed’ +
’Subselect’)
A QQCPUT RENAME(QQI1) +
COLHDG(’Row’ +
’Retrieval’ +
’CPU Time’)
A QQCLKT RENAME(QQI2) +
COLHDG(’Row’ +
’Retrieval’ +
’Clock Time’)
A QQSYNR RENAME(QQ13) +
COLHDG(’Synch’ +
’Reads’)
A QQSYNW RENAME(QQ14) +
COLHDG(’Synch’ +
’Writes’)
A QQASYR RENAME(QQ15) +
COLHDG(’Asynch’ +
’Reads’)
A QQASYW RENAME(QQ16) +
COLHDG(’Asynch’ +
’Writes’)
A QQRCDR RENAME(QQ17) +
COLHDG(’Rows’ +
’Returned’)
A QQGETC RENAME(QQ18) +
COLHDG(’Number’ +
’of Gets’)
A K QQJFLD
A S QQRID CMP(EQ 3019)

Table 26. QQQ3019 - Detail Row for Rows Retrieved

Logical Column Physical Column
Name Name Description
QQRID QQRID Row identification
QQTIME QQTIME Time row was created

194 DB2 UDB for iSeries Database Performance and Query Optimization V5R1
Table 26. QQQ3019 - Detail Row for Rows Retrieved (continued)
Logical Column Physical Column
Name Name Description
QQJFLD QQJFLD Join column (unique per job)
QQRDBN QQRDBN Relational database name
QQSYS QQSYS System name
QQJOB QQJOB Job name
QQUSER QQUSER Job user
QQJNUM QQJNUM Job number
QQTHRD QQI9 Thread identifier
QQUCNT QQUCNT Unique count (unique per query)
QQUDEF QQUDEF User defined column
QQQDTN QQQDTN Unique subselect number
QQQDTL QQQDTL Subselect nested level
QQMATN QQMATN Materialized view subselect number
QQMATL QQMATL Materialized view nested level
QDQDTN QVP15A Decomposed query subselect number, unique across all
decomposed subselects
QDQDTT QVP15B Total number of decomposed subselects
QDQDTR QVP15C Decomposed query subselect reason code
QDQDTS QVP15D Decomposed query subselect number for the first decomposed
subselect
QQCPUT QQI1 CPU time to return all rows, in milliseconds
QQCLKT QQI2 Clock time to return all rows, in milliseconds
QQSYNR QQI3 Number of synchronous database reads
QQSYNW QQI4 Number of synchronous database writes
QQASYR QQI5 Number of asynchronous database reads
QQASYW QQI6 Number of asynchronous database writes
QQRCDR QQI7 Number of rows returned
QQGETC QQI8 Number of calls to retrieve rows returned

Database monitor logical table 3021 - Summary Row for Bitmap

Created
|...+....1....+....2....+....3....+....4....+....5....+....6....+....7....+....8
A*
A* DB Monitor logical table 3021 - Summary Row for Bitmap Created
A*
A* New row added for Visual Explain
A*
A R QQQ3021 PTABLE(*CURLIB/QAQQDBMN)
A QQRID
A QQTIME
A QQJFLD
A QQRDBN
A QQSYS
A QQJOB
A QQUSER
A QQJNUM

Appendix A. Database monitor: DDS 195

 A QQTHRD RENAME(QQI9) +
COLHDG(’Thread’ +
’Identifier’)
A QQUCNT
A QQUDEF
A QQQDTN
A QQQDTL
A QQMATN
A QQMATL
A QDQDTN RENAME(QVP15A) +
COLHDG(’Decomposed’ +
’Subselect’ +
’Number’)
A QDQDTT RENAME(QVP15B) +
COLHDG(’Number of’ +
’Decomposed’ +
’Subselects’)
A QDQDTR RENAME(QVP15C) +
COLHDG(’Decomposed’ +
’Reason’ +
’Code’)
A QDQDTS RENAME(QVP15D) +
COLHDG(’Starting’ +
’Decomposed’ +
’Subselect’)
A QVRCNT
A QVPARPF
A QVPARPL
A QVPARD
A QVPARU
A QVPARRC
A QQEPT
A QVCTIM
A QQREST
A QQAJN
A QQJNP
A QQJNDS RENAME(QQI6) +
COLHDG(’Data Space’ +
’Number’)
A QQJNMT RENAME(QQC21) +
COLHDG(’Join’ ’Method’)
A QQJNTY RENAME(QQC22) +
COLHDG(’Join’ ’Type’)
A QQJNOP RENAME(QQC23) +
COLHDG(’Join’ ’Operator’)
A QVJFANO
A QVFILES
A QVBMSIZ RENAME(QQI2) +
COLHDG(’Bitmap’ +
’Size’)
A QVBMCNT RENAME(QVP151) +
COLHDG(’Number of’ +
’Bitmaps’
’Created’)
A QVBMIDS RENAME(QVC3001) +
COLHDG(’Internal’ +
’Bitmap’ ’IDs’)
A K QQJFLD
A S QQRID CMP(EQ 3021)

Table 27. QQQ3021 - Summary Row for Bitmap Created

Logical Column Physical Column
Name Name Description
QQRID QQRID Row identification
QQTIME QQTIME Time row was created

196 DB2 UDB for iSeries Database Performance and Query Optimization V5R1
Table 27. QQQ3021 - Summary Row for Bitmap Created (continued)
Logical Column Physical Column
Name Name Description
QQJFLD QQJFLD Join column (unique per job)
QQRDBN QQRDBN Relational database name
QQSYS QQSYS System name
QQJOB QQJOB Job name
QQUSER QQUSER Job user
QQJNUM QQJNUM Job number
QQTHRD QQI9 Thread identifier
QQUCNT QQUCNT Unique count (unique per query)
QQUDEF QQUDEF User defined column
QQQDTN QQQDTN Unique subselect number
QQQDTL QQQDTL Subselect nested level
QQMATN QQMATN Materialized view subselect number
QQMATL QQMATL Materialized view nested level
QDQDTN QVP15A Decomposed query subselect number, unique across all
decomposed subselects
QDQDTT QVP15B Total number of decomposed subselects
QDQDTR QVP15C Decomposed query subselect reason code
QDQDTS QVP15D Decomposed query subselect number for the first decomposed
subselect
QVRCNT QVRCNT Unique refresh counter
QVPARPF QVPARPF Parallel Prefetch (Y/N)
QVPARPL QVPARPL Parallel Preload (index used)
QVPARD QVPARD Parallel degree requested (index used)
QVPARU QVPARU Parallel degree used (index used)
QVPARRC QVPARRC Reason parallel processing was limited (index used)
QQEPT QQEPT Estimated processing time, in seconds
QVCTIM QVCTIM Estimated cumulative time, in seconds
QQREST QQREST Estimated rows selected
QQAJN QQAJN Estimated number of joined rows
QQJNP QQJNP Join position - when available
QQJNDS QQI6 dataspace number/Original table position
QQJNMT QQC21 Join method - when available
v NL - Nested loop
v MF - Nested loop with selection
v HJ - Hash join
QQJNTY QQC22 Join type - when available
v IN - Inner join
v PO - Left partial outer join
v EX - Exception join

Appendix A. Database monitor: DDS 197

Table 27. QQQ3021 - Summary Row for Bitmap Created (continued)
Logical Column Physical Column
Name Name Description
QQJNOP QQC23 Join operator - when available
v EQ - Equal
v NE - Not equal
v GT - Greater than
v GE - Greater than or equal
v LT - Less than
v LE - Less than or equal
v CP - Cartesian product
QVJFANO QVJFANO Join fan out. Possible values are:
v N - Normal join situation where fanout is allowed and each
matching row of the join fanout is returned.
v D - Distinct fanout. Join fanout is allowed however none of the join
fanout rows are returned.
v U - Unique fanout. Join fanout is not allowed. Error situation if join
fanout occurs.
QVFILES QVFILES Number of tables joined
QVBMSIZ QQI2 Bitmap size
QVBMCNT QVP151 Number of bitmaps created
QVBMIDS QVC3001 Internal bitmap IDs

Database monitor logical table 3022 - Summary Row for Bitmap Merge
|...+....1....+....2....+....3....+....4....+....5....+....6....+....7....+....8
A*
A* DB Monitor logical table 3022 - Summary Row for Bitmap Merge
A*
A* New row added for Visual Explain
A*
A R QQQ3022 PTABLE(*CURLIB/QAQQDBMN)
A QQRID
A QQTIME
A QQJFLD
A QQRDBN
A QQSYS
A QQJOB
A QQUSER
A QQJNUM
A QQTHRD RENAME(QQI9) +
COLHDG(’Thread’ +
’Identifier’)
A QQUCNT
A QQUDEF
A QQQDTN
A QQQDTL
A QQMATN
A QQMATL
A QDQDTN RENAME(QVP15A) +
COLHDG(’Decomposed’ +
’Subselect’ +
’Number’)
A QDQDTT RENAME(QVP15B) +
COLHDG(’Number of’ +
’Decomposed’ +
’Subselects’)

198 DB2 UDB for iSeries Database Performance and Query Optimization V5R1
 A QDQDTR RENAME(QVP15C) +
COLHDG(’Decomposed’ +
’Reason’ +
’Code’)
A QDQDTS RENAME(QVP15D) +
COLHDG(’Starting’ +
’Decomposed’ +
’Subselect’)
A QVRCNT
A QVPARPF
A QVPARPL
A QVPARD
A QVPARU
A QVPARRC
A QQEPT
A QVCTIM
A QQREST
A QQAJN
A QQJNP
A QQJNDS RENAME(QQI6) +
COLHDG(’Data Space’ +
’Number’)
A QQJNMT RENAME(QQC21) +
COLHDG(’Join’ ’Method’)
A QQJNTY RENAME(QQC22) +
COLHDG(’Join’ ’Type’)
A QQJNOP RENAME(QQC23) +
COLHDG(’Join’ ’Operator’)
A QVJFANO
A QVFILES
A QVBMSIZ RENAME(QQI2) +
COLHDG(’Bitmap’ +
’Size’)
A QVBMID RENAME(QVC101) +
COLHDG(’Internal’ +
’Bitmap’ ’ID’)
A QVBMIDMG RENAME(QVC3001) +
COLHDG(’Bitmaps’ +
’Merged’)
A K QQJFLD
A S QQRID CMP(EQ 3022)

Table 28. QQQ3022 - Summary Row for Bitmap Merge

Logical Column Physical Column
Name Name Description
QQRID QQRID Row identification
QQTIME QQTIME Time row was created
QQJFLD QQJFLD Join column (unique per job)
QQRDBN QQRDBN Relational database name
QQSYS QQSYS System name
QQJOB QQJOB Job name
QQUSER QQUSER Job user
QQJNUM QQJNUM Job number
QQTHRD QQI9 Thread identifier
QQUCNT QQUCNT Unique count (unique per query)
QQUDEF QQUDEF User defined column
QQQDTN QQQDTN Unique subselect number
QQQDTL QQQDTL Subselect nested level

Appendix A. Database monitor: DDS 199

Table 28. QQQ3022 - Summary Row for Bitmap Merge (continued)
Logical Column Physical Column
Name Name Description
QQMATN QQMATN Materialized view subselect number
QQMATL QQMATL Materialized view nested level
QDQDTN QVP15A Decomposed query subselect number, unique across all
decomposed subselects
QDQDTT QVP15B Total number of decomposed subselects
QDQDTR QVP15C Decomposed query subselect reason code
QDQDTS QVP15D Decomposed query subselect number for the first decomposed
subselect
QVRCNT QVRCNT Unique refresh counter
QVPARPF QVPARPF Parallel Prefetch (Y/N)
QVPARPL QVPARPL Parallel Preload (index used)
QVPARD QVPARD Parallel degree requested (index used)
QVPARU QVPARU Parallel degree used (index used)
QVPARRC QVPARRC Reason parallel processing was limited (index used)
QQEPT QQEPT Estimated processing time, in seconds
QVCTIM QVCTIM Estimated cumulative time, in seconds
QQREST QQREST Estimated rows selected
QQAJN QQAJN Estimated number of joined rows
QQJNP QQJNP Join position - when available
QQJNDS QQI6 dataspace number/Original table position
QQJNMT QQC21 Join method - when available
v NL - Nested loop
v MF - Nested loop with selection
v HJ - Hash join
QQJNTY QQC22 Join type - when available
v IN - Inner join
v PO - Left partial outer join
v EX - Exception join
QQJNOP QQC23 Join operator - when available
v EQ - Equal
v NE - Not equal
v GT - Greater than
v GE - Greater than or equal
v LT - Less than
v LE - Less than or equal
v CP - Cartesian product
QVJFANO QVJFANO Join fan out. Possible values are:
v N - Normal join situation where fanout is allowed and each
matching row of the join fanout is returned.
v D - Distinct fanout. Join fanout is allowed however none of the join
fanout rows are returned.
v U - Unique fanout. Join fanout is not allowed. Error situation if join
fanout occurs.

200 DB2 UDB for iSeries Database Performance and Query Optimization V5R1
Table 28. QQQ3022 - Summary Row for Bitmap Merge (continued)
Logical Column Physical Column
Name Name Description
QVFILES QVFILES Number of tables joined
QVBMSIZ QQI2 Bitmap size
QVBMID QVC101 Internal bitmap ID
QVBMIDMG QVC3001 IDs of bitmaps merged together

Database monitor logical table 3023 - Summary for Temp Hash Table
Created
|...+....1....+....2....+....3....+....4....+....5....+....6....+....7....+....8
A*
A* DB Monitor logical table 3023 - Summary for Temp Hash Table Created
A*
A* New row added for Visual Explain
A*
A R QQQ3023 PTABLE(*CURLIB/QAQQDBMN)
A QQRID
A QQTIME
A QQJFLD
A QQRDBN
A QQSYS
A QQJOB
A QQUSER
A QQJNUM
A QQTHRD RENAME(QQI9) +
COLHDG(’Thread’ +
’Identifier’)
A QQUCNT
A QQUDEF
A QQQDTN
A QQQDTL
A QQMATN
A QQMATL
A QDQDTN RENAME(QVP15A) +
COLHDG(’Decomposed’ +
’Subselect’ +
’Number’)
A QDQDTT RENAME(QVP15B) +
COLHDG(’Number of’ +
’Decomposed’ +
’Subselects’)
A QDQDTR RENAME(QVP15C) +
COLHDG(’Decomposed’ +
’Reason’ +
’Code’)
A QDQDTS RENAME(QVP15D) +
COLHDG(’Starting’ +
’Decomposed’ +
’Subselect’)
A QVRCNT
A QVPARPF
A QVPARPL
A QVPARD
A QVPARU
A QVPARRC
A QQEPT
A QVCTIM
A QQREST
A QQAJN
A QQJNP

Appendix A. Database monitor: DDS 201

 A QQJNDS RENAME(QQI8) +
COLHDG(’Data Space’ +
’Number’)
A QQJNMT RENAME(QQC21) +
COLHDG(’Join’ ’Method’)
A QQJNTY RENAME(QQC22) +
COLHDG(’Join’ ’Type’)
A QQJNOP RENAME(QQC23) +
COLHDG(’Join’ ’Operator’)
A QVJFANO
A QVFILES
A QVHTRC RENAME(QVC1F) +
COLHDG(’Hash’ +
’Table’ +
’Reason Code’)
A QVHTENT RENAME(QQI2) +
COLHDG(’Hash’ +
’Table’ +
’Entries’)
A QVHTSIZ RENAME(QQI3) +
COLHDG(’Hash’ +
’Table’ +
’Size’)
A QVHTRSIZ RENAME(QQI4) +
COLHDG(’Hash’ +
’Table’ +
’Row’ ’Size’)
A QVHTKSIZ RENAME(QQI5) +
COLHDG(’Hash’ +
’Key’ +
’Size’)
A QVHTESIZ RENAME(QQI6) +
COLHDG(’Hash’ +
’Element’ +
’Size’)
A QVHTPSIZ RENAME(QQI7) +
COLHDG(’Pool’ +
’Size’)
A QVHTPID RENAME(QQI8) +
COLHDG(’Pool’ +
’ID’)
A QVHTNAM RENAME(QVC101) +
COLHDG(’Hash’ +
’Table’ +
’Name’)
A QVHTLIB RENAME(QVC102) +
COLHDG(’Hash’ +
’Table’ +
’Library’)
A QVHTCOL RENAME(QVC3001) +
COLHDG(’Hash’ +
’Table’ +
’Columns’)
A K QQJFLD
A S QQRID CMP(EQ 3023)

Table 29. QQQ3023 - Summary for Temp Hash Table Created

Logical Column Physical Column
Name Name Description
QQRID QQRID Row identification
QQTIME QQTIME Time row was created
QQJFLD QQJFLD Join column (unique per job)
QQRDBN QQRDBN Relational database name

202 DB2 UDB for iSeries Database Performance and Query Optimization V5R1
Table 29. QQQ3023 - Summary for Temp Hash Table Created (continued)
Logical Column Physical Column
Name Name Description
QQSYS QQSYS System name
QQJOB QQJOB Job name
QQUSER QQUSER Job user
QQJNUM QQJNUM Job number
QQTHRD QQI9 Thread identifier
QQUCNT QQUCNT Unique count (unique per query)
QQUDEF QQUDEF User defined column
QQQDTN QQQDTN Unique subselect number
QQQDTL QQQDTL Subselect nested level
QQMATN QQMATN Materialized view subselect number
QQMATL QQMATL Materialized view nested level
QDQDTN QVP15A Decomposed query subselect number, unique across all
decomposed subselects
QDQDTT QVP15B Total number of decomposed subselects
QDQDTR QVP15C Decomposed query subselect reason code
QDQDTS QVP15D Decomposed query subselect number for the first decomposed
subselect
QVRCNT QVRCNT Unique refresh counter
QVPARPF QVPARPF Parallel Prefetch (Y/N)
QVPARPL QVPARPL Parallel Preload (index used)
QVPARD QVPARD Parallel degree requested (index used)
QVPARU QVPARU Parallel degree used (index used)
QVPARRC QVPARRC Reason parallel processing was limited (index used)
QQEPT QQEPT Estimated processing time, in seconds
QVCTIM QVCTIM Estimated cumulative time, in seconds
QQREST QQREST Estimated rows selected
QQAJN QQAJN Estimated number of joined rows
QQJNP QQJNP Join position - when available
QQJNDS QQI6 dataspace number/Original table position
QQJNMT QQC21 Join method - when available
v NL - Nested loop
v MF - Nested loop with selection
v HJ - Hash join
QQJNTY QQC22 Join type - when available
v IN - Inner join
v PO - Left partial outer join
v EX - Exception join

Appendix A. Database monitor: DDS 203

Table 29. QQQ3023 - Summary for Temp Hash Table Created (continued)
Logical Column Physical Column
Name Name Description
QQJNOP QQC23 Join operator - when available
v EQ - Equal
v NE - Not equal
v GT - Greater than
v GE - Greater than or equal
v LT - Less than
v LE - Less than or equal
v CP - Cartesian product
QVJFANO QVJFANO Join fan out. Possible values are:
v N - Normal join situation where fanout is allowed and each
matching row of the join fanout is returned.
v D - Distinct fanout. Join fanout is allowed however none of the join
fanout rows are returned.
v U - Unique fanout. Join fanout is not allowed. Error situation if join
fanout occurs.
QVFILES QVFILES Number of tables joined
QVHTRC QVC1F Hash table reason code
v J - Created for hash join
v G - Created for hash grouping
QVHTENT QQI2 Hash table entries
QVHTSIZ QQI3 Hash table size
QVHTRSIZ QQI4 Hash table row size
QVHTKSIZ QQI5 Hash table key size
QVHTESIZ QQIA Hash table element size
QVHTPSIZ QQI7 Hash table pool size
QVHTPID QQI8 Hash table pool ID
QVHTNAM QVC101 Hash table internal name
QVHTLIB QVC102 Hash table library
QVHTCOL QVC3001 Columns used to create hash table

Database monitor logical table 3025 - Summary Row for Distinct

Processing
|...+....1....+....2....+....3....+....4....+....5....+....6....+....7....+....8
A*
A* DB Monitor logical table 3025 - Summary Row for Distinct Processing
A*
A* New row added for Visual Explain
A*
A R QQQ3025 PTABLE(*CURLIB/QAQQDBMN)
A QQRID
A QQTIME
A QQJFLD
A QQRDBN
A QQSYS
A QQJOB
A QQUSER
A QQJNUM

204 DB2 UDB for iSeries Database Performance and Query Optimization V5R1
 A QQTHRD RENAME(QQI9) +
COLHDG(’Thread’ +
’Identifier’)
A QQUCNT
A QQUDEF
A QQQDTN
A QQQDTL
A QQMATN
A QQMATL
A QDQDTN RENAME(QVP15A) +
COLHDG(’Decomposed’ +
’Subselect’ +
’Number’)
A QDQDTT RENAME(QVP15B) +
COLHDG(’Number of’ +
’Decomposed’ +
’Subselects’)
A QDQDTR RENAME(QVP15C) +
COLHDG(’Decomposed’ +
’Reason’ +
’Code’)
A QDQDTS RENAME(QVP15D) +
COLHDG(’Starting’ +
’Decomposed’ +
’Subselect’)
A QVRCNT
A QVPARPF
A QVPARPL
A QVPARD
A QVPARU
A QVPARRC
A QQEPT
A QVCTIM
A QQREST
A K QQJFLD
A S QQRID CMP(EQ 3025)

Table 30. QQQ3025 - Summary Row for Distinct Processing

Logical Column Physical Column
Name Name Description
QQRID QQRID Row identification
QQTIME QQTIME Time row was created
QQJFLD QQJFLD Join column (unique per job)
QQRDBN QQRDBN Relational database name
QQSYS QQSYS System name
QQJOB QQJOB Job name
QQUSER QQUSER Job user
QQJNUM QQJNUM Job number
QQTHRD QQI9 Thread identifier
QQUCNT QQUCNT Unique count (unique per query)
QQUDEF QQUDEF User defined column
QQQDTN QQQDTN Unique subselect number
QQQDTL QQQDTL Subselect nested level
QQMATN QQMATN Materialized view subselect number
QQMATL QQMATL Materialized view nested level

Appendix A. Database monitor: DDS 205

Table 30. QQQ3025 - Summary Row for Distinct Processing (continued)
Logical Column Physical Column
Name Name Description
QDQDTN QVP15A Decomposed query subselect number, unique across all
decomposed subselects
QDQDTT QVP15B Total number of decomposed subselects
QDQDTR QVP15C Decomposed query subselect reason code
QDQDTS QVP15D Decomposed query subselect number for the first decomposed
subselect
QVRCNT QVRCNT Unique refresh counter
QVPARPF QVPARPF Parallel Prefetch (Y/N)
QVPARPL QVPARPL Parallel Preload (index used)
QVPARD QVPARD Parallel degree requested (index used)
QVPARU QVPARU Parallel degree used (index used)
QVPARRC QVPARRC Reason parallel processing was limited (index used)
QQEPT QQEPT Estimated processing time, in seconds
QVCTIM QVCTIM Estimated cumulative time, in seconds
QQREST QQREST Estimated rows selected

Database monitor logical table 3027 - Summary Row for Subquery

Merge
|...+....1....+....2....+....3....+....4....+....5....+....6....+....7....+....8
A*
A* DB Monitor logical table 3027 - Summary Row for Subquery Merge
A*
A* New row added for Visual Explain
A*
A R QQQ3027 PTABLE(*CURLIB/QAQQDBMN)
A QQRID
A QQTIME
A QQJFLD
A QQRDBN
A QQSYS
A QQJOB
A QQUSER
A QQJNUM
A QQTHRD RENAME(QQI9) +
COLHDG(’Thread’ +
’Identifier’)
A QQUCNT
A QQUDEF
A QQQDTN
A QQQDTL
A QQMATN
A QQMATL
A QDQDTN RENAME(QVP15A) +
COLHDG(’Decomposed’ +
’Subselect’ +
’Number’)
A QDQDTT RENAME(QVP15B) +
COLHDG(’Number of’ +
’Decomposed’ +
’Subselects’)
A QDQDTR RENAME(QVP15C) +
COLHDG(’Decomposed’ +

206 DB2 UDB for iSeries Database Performance and Query Optimization V5R1
 ’Reason’ +
’Code’)
A QDQDTS RENAME(QVP15D) +
COLHDG(’Starting’ +
’Decomposed’ +
’Subselect’)
A QVRCNT
A QVPARPF
A QVPARPL
A QVPARD
A QVPARU
A QVPARRC
A QQEPT
A QVCTIM
A QQREST
A QQAJN
A QQJNP
A QQJNDS RENAME(QQI6) +
COLHDG(’Data Space’ +
’Number’)
A QQJNMT RENAME(QQC21) +
COLHDG(’Join’ ’Method’)
A QQJNTY RENAME(QQC22) +
COLHDG(’Join’ ’Type’)
A QQJNOP RENAME(QQC23) +
COLHDG(’Join’ ’Operator’)
A QVJFANO
A QVFILES
A QVIQDTN RENAME(QVP151) +
COLHDG(’Subselect’ +
’Number’ +
’Inner’ +
’Subselect’)
A QVIQDTL RENAME(QVP152) +
COLHDG(’Subselect’ +
’Level’ +
’Inner’ +
’Subselect’)
A QVIMATN RENAME(QVP153) +
COLHDG(’View’ +
’Number’ +
’Inner’ +
’Subselect’)
A QVIMATL RENAME(QVP154) +
COLHDG(’View’ +
’Level’ +
’Inner’ +
’Subselect’)
A QVSUBOP RENAME(QQC101) +
COLHDG(’Subquery’ +
’Operator’)
A QVCORRI RENAME(QQC11) +
COLHDG(’Correlated’ +
’Columns’ +
’Exist’)
A QVCORRC RENAME(QVC3001) +
COLHDG(’Correlated’ +
’Columns’)
A K QQJFLD
A S QQRID CMP(EQ 3027)

Table 31. QQQ3027 - Summary Row for Subquery Merge

Logical Column Physical Column
Name Name Description
QQRID QQRID Row identification

Appendix A. Database monitor: DDS 207

Table 31. QQQ3027 - Summary Row for Subquery Merge (continued)
Logical Column Physical Column
Name Name Description
QQTIME QQTIME Time row was created
QQJFLD QQJFLD Join column (unique per job)
QQRDBN QQRDBN Relational database name
QQSYS QQSYS System name
QQJOB QQJOB Job name
QQUSER QQUSER Job user
QQJNUM QQJNUM Job number
QQTHRD QQI9 Thread identifier
QQUCNT QQUCNT Unique count (unique per query)
QQUDEF QQUDEF User defined column
QQQDTN QQQDTN Subselect number for outer subquery
QQQDTL QQQDTL Subselect level for outer subquery
QQMATN QQMATN Materialized view subselect number for outer subquery
QQMATL QQMATL Materialized view subselect level for outer subquery
QDQDTN QVP15A Decomposed query subselect number, unique across all
decomposed subselects
QDQDTT QVP15B Total number of decomposed subselects
QDQDTR QVP15C Decomposed query subselect reason code
QDQDTS QVP15D Decomposed query subselect number for the first decomposed
subselect
QVRCNT QVRCNT Unique refresh counter
QVPARPF QVPARPF Parallel Prefetch (Y/N)
QVPARPL QVPARPL Parallel Preload (index used)
QVPARD QVPARD Parallel degree requested (index used)
QVPARU QVPARU Parallel degree used (index used)
QVPARRC QVPARRC Reason parallel processing was limited (index used)
QQEPT QQEPT Estimated processing time, in seconds
QVCTIM QVCTIM Estimated cumulative time, in seconds
QQREST QQREST Estimated rows selected
QQAJN QQAJN Estimated number of joined rows
QQJNP QQJNP Join position - when available
QQJNDS QQI6 dataspace number/Original table position
QQJNMT QQC21 Join method - when available
v NL - Nested loop
v MF - Nested loop with selection
v HJ - Hash join
QQJNTY QQC22 Join type - when available
v IN - Inner join
v PO - Left partial outer join
v EX - Exception join

208 DB2 UDB for iSeries Database Performance and Query Optimization V5R1
Table 31. QQQ3027 - Summary Row for Subquery Merge (continued)
Logical Column Physical Column
Name Name Description
QQJNOP QQC23 Join operator - when available
v EQ - Equal
v NE - Not equal
v GT - Greater than
v GE - Greater than or equal
v LT - Less than
v LE - Less than or equal
v CP - Cartesian product
QVJFANO QVJFANO Join fan out. Possible values are:
v N - Normal join situation where fanout is allowed and each
matching row of the join fanout is returned.
v D - Distinct fanout. Join fanout is allowed however none of the join
fanout rows are returned.
v U - Unique fanout. Join fanout is not allowed. Error situation if join
fanout occurs.
QVFILES QVFILES Number of tables joined
QVIQDTN QVP151 Subselect number for inner subquery
QVIQDTL QVP152 Subselect level for inner subquery
QVIMATN QVP153 Materialized view subselect number for inner subquery
QVIMATL QVP154 Materialized view subselect level for inner subquery
QVSUBOP QQC101 Subquery operator. Possible values are:
v EQ - Equal
v NE - Not Equal
v LT - Less Than or Equal
v LT - Less Than
v GE - Greater Than or Equal
v GT - Greater Than
v IN
v LIKE
v EXISTS
v NOT - Can precede IN, LIKE or EXISTS
QVCORRI QQC11 Correlated columns exist (Y/N)
QVCORRC QVC3001 List of correlated columns with corresponding QDT number

Database monitor logical table 3028 - Summary Row for Grouping

|...+....1....+....2....+....3....+....4....+....5....+....6....+....7....+....8
A*
A* DB Monitor logical table 3028 - Summary Row for Grouping
A*
A* New row added for Visual Explain
A*
A R QQQ3028 PTABLE(*CURLIB/QAQQDBMN)
A QQRID
A QQTIME
A QQJFLD
A QQRDBN
A QQSYS

Appendix A. Database monitor: DDS 209

 A QQJOB
A QQUSER
A QQJNUM
A QQTHRD RENAME(QQI9) +
COLHDG(’Thread’ +
’Identifier’)
A QQUCNT
A QQUDEF
A QQQDTN
A QQQDTL
A QQMATN
A QQMATL
A QDQDTN RENAME(QVP15A) +
COLHDG(’Decomposed’ +
’Subselect’ +
’Number’)
A QDQDTT RENAME(QVP15B) +
COLHDG(’Number of’ +
’Decomposed’ +
’Subselects’)
A QDQDTR RENAME(QVP15C) +
COLHDG(’Decomposed’ +
’Reason’ +
’Code’)
A QDQDTS RENAME(QVP15D) +
COLHDG(’Starting’ +
’Decomposed’ +
’Subselect’)
A QVRCNT
A QVPARPF
A QVPARPL
A QVPARD
A QVPARU
A QVPARRC
A QQEPT
A QVCTIM
A QQREST
A QQAJN
A QQJNP
A QQJNDS RENAME(QQI6) +
COLHDG(’Data Space’ +
’Number’)
A QQJNMT RENAME(QQC21) +
COLHDG(’Join’ ’Method’)
A QQJNTY RENAME(QQC22) +
COLHDG(’Join’ ’Type’)
A QQJNOP RENAME(QQC23) +
COLHDG(’Join’ ’Operator’)
A QVJFANO
A QVFILES
A QVGBYIM RENAME(QQC11) +
COLHDG(’Grouping’ +
’Implementation’)
A QVGBYIT RENAME(QQC15) +
COLHDG(’Index’ +
’Type’)
A QVGBYIX RENAME(QQC101) +
COLHDG(’Grouping’ +
’Index’)
A QVGBYIL RENAME(QQC102) +
COLHDG(’Grouping’ +
’Index’ +
’Library’)
A QVGBYIXL RENAME(QVINAM) +
COLHDG(’Grouping’ +
’Index’ +
’Long Name’)

210 DB2 UDB for iSeries Database Performance and Query Optimization V5R1
 A QVGBYILL RENAME(QVILIB) +
COLHDG(’Grouping’ +
’Library’ +
’Long Name’)
A QVGBYHV RENAME(QQC12) +
COLHDG(’Having’ +
’Selection’ +
’Exists’)
A QVGBYHW RENAME(QQC13) +
COLHDG(’Having to’ +
’Where’ +
’Conversion’)
A QVGBYN RENAME(QQI2) +
COLHDG(’Estimated’ +
’Number of’ +
’Groups’)
A QVGBYNA RENAME(QQI3) +
COLHDG(’Average’ +
’Rows per’ +
’Group’)
A QVGBYCOL RENAME(QVC3001) +
COLHDG(’Grouping’ +
’Columns’)
A QVGBYMIN RENAME(QVC3002) +
COLHDG(’MIN’ +
’Columns’)
A QVGBYMAX RENAME(QVC3003) +
COLHDG(’MAX’ +
’Columns’)
A QVGBYSUM RENAME(QVC3004) +
COLHDG(’SUM’ +
’Columns’)
A QVGBYCNT RENAME(QVC3005) +
COLHDG(’COUNT’ +
’Columns’)
A QVGBYAVG RENAME(QVC3006) +
COLHDG(’AVG’ +
’Columns’)
A QVGBYSTD RENAME(QVC3007) +
COLHDG(’STDDEV’ +
’Columns’)
A QVGBYVAR RENAME(QVC3008) +
COLHDG(’VAR’ +
’Columns’)
A K QQJFLD
A S QQRID CMP(EQ 3028)

Table 32. QQQ3028 - Summary Row for Grouping

Logical Column Physical Column
Name Name Description
QQRID QQRID Row identification
QQTIME QQTIME Time row was created
QQJFLD QQJFLD Join column (unique per job)
QQRDBN QQRDBN Relational database name
QQSYS QQSYS System name
QQJOB QQJOB Job name
QQUSER QQUSER Job user
QQJNUM QQJNUM Job number
QQTHRD QQI9 Thread identifier
QQUCNT QQUCNT Unique count (unique per query)

Appendix A. Database monitor: DDS 211

Table 32. QQQ3028 - Summary Row for Grouping (continued)
Logical Column Physical Column
Name Name Description
QQUDEF QQUDEF User defined column
QQQDTN QQQDTN Unique subselect number
QQQDTL QQQDTL Subselect nested level
QQMATN QQMATN Materialized view subselect number
QQMATL QQMATL Materialized view nested level
QDQDTN QVP15A Decomposed query subselect number, unique across all
decomposed subselects
QDQDTT QVP15B Total number of decomposed subselects
QDQDTR QVP15C Decomposed query subselect reason code
QDQDTS QVP15D Decomposed query subselect number for the first decomposed
subselect
QVRCNT QVRCNT Unique refresh counter
QVPARPF QVPARPF Parallel Prefetch (Y/N)
QVPARPL QVPARPL Parallel Preload (index used)
QVPARD QVPARD Parallel degree requested (index used)
QVPARU QVPARU Parallel degree used (index used)
QVPARRC QVPARRC Reason parallel processing was limited (index used)
QQEPT QQEPT Estimated processing time, in seconds
QVCTIM QVCTIM Estimated cumulative time, in seconds
QQREST QQREST Estimated rows selected
QQAJN QQAJN Estimated number of joined rows
QQJNP QQJNP Join position
QQJNDS QQI1 dataspace number/original table position
QQJNMT QQC21 Join method - when available
v NL - Nested loop
v MF - Nested loop with selection
v HJ - Hash join
QQJNTY QQC22 Join type - when available
v IN - Inner join
v PO - Left partial outer join
v EX - Exception join
QQJNOP QQC23 Join operator - when available
v EQ - Equal
v NE - Not equal
v GT - Greater than
v GE - Greater than or equal
v LT - Less than
v LE - Less than or equal
v CP - Cartesian product

212 DB2 UDB for iSeries Database Performance and Query Optimization V5R1
Table 32. QQQ3028 - Summary Row for Grouping (continued)
Logical Column Physical Column
Name Name Description
QVJFANO QVJFANO Join fan out. Possible values are:
v N - Normal join situation where fanout is allowed and each
matching row of the join fanout is returned.
v D - Distinct fanout. Join fanout is allowed however none of the join
fanout rows are returned.
v U - Unique fanout. Join fanout is not allowed. Error situation if join
fanout occurs.
QVFILES QVFILES Number of tables joined
QVGBYI QQC11 Groupby implementation
v ’ ’ - No grouping
v I - Index
v H - Hash
QVGBYIT QQC15 Type of Index. Possible values are:
v B - Binary Radix Index
v C - Constraint (Binary Radix)
v E - Encoded Vector Index (EVI)
v X - Query created temporary index
QVGBYIX QQC101 Index, or constraint, used for grouping
QVGBYIL QQC102 Library of index used for grouping
QVGBYIXL QVINAM Long name of index, or constraint, used for grouping
QVGBYILL QVILIB Long name of index, or constraint, library used for grouping
QVGBYHV QQC12 Having selection exists (Y/N)
QVGBYHW QQC13 Having to Where conversion (Y/N)
QVGBYN QQI2 Estimated number of groups
QVGBYNA QQI3 Average number of rows in each group
QVGBYCOL QVC3001 Grouping columns
QVGBYMIN QVC3002 MIN columns
QVGBYMAX QVC3003 MAX columns
QVGBYSUM QVC3004 SUM columns
QVGBYCNT QVC3005 COUNT columns
QVGBYAVG QVC3006 AVG columns
QVGBYSTD QVC3007 STDDEV columns
QVGBYVAR QVC3008 VAR columns

Appendix A. Database monitor: DDS 213

214 DB2 UDB for iSeries Database Performance and Query Optimization V5R1
Appendix B. Memory Resident Database Monitor: DDS
This appendix contains the DDS that is used to create the memory resident database monitor physical and
logical files.

External table description (QAQQQRYI)

Table 33. QAQQQRYI - Summary Row for SQL Information
Column Name Description
QQKEY Join column (unique per query) used to link rows for a single query together
QQTIME Time row was created
QQJOB Job name
QQUSER Job user
QQJNUM Job number
QQTHID Thread Id
QQUDEF User defined column
QQPLIB Name of the library containing the program or package
QQCNAM Cursor name
QQPNAM Name of the package or name of the program that contains the current SQL
statement
QQSNAM Name of the statement for SQL statement, if applicable
QQCNT Statement usage count
QQAVGT Average runtime (ms)
QQMINT Minimum runtime (ms)
QQMAXT Maximum runtime (ms)
QQOPNT Open time for most expensive execution (ms)
QQFETT Fetch time for most expensive execution (ms)
QQCLST Close time for most expensive execution (ms)
QQOTHT Other time for most expensive execution (ms)
QQLTU Time statement last used
QQMETU Most expensive time used
QQAPRT Access plan rebuild time
QQFULO Number of full opens
QQPSUO Number of pseudo-opens
QQTOTR Total rows in table if non-join
QQRROW Number of result rows returned
QQRROW Statement function

S - Select - Update
I - Insert
D - Delete
L - Data definition language
O - Other

© Copyright IBM Corp. 2000, 2001 215

Table 33. QAQQQRYI - Summary Row for SQL Information (continued)
Column Name Description
QQSTOP Statement operation
v AL - Alter table
v CA - Call
v CC - Create collection
v CD - Create type
v CF - Create function
v CI - Create index
v CL - Close
v CM - Commit
v CN - Connect
v CO - Comment on
v CP - Create procedure
v CS - Create alias/synonym
v CT - Create table
v CV - Create view
v DE - Describe
v DI - Disconnect
v DL - Delete
v DM - Describe parameter marker
v DP - Declare procedure
v DR - Drop
v DT - Describe table
v EI - Execute immediate
v EX - Execute
v FE - Fetch
v FL - Free locator
v GR - Grant
v HC - Hard close
v IN - Insert
v JR - Server job reused
v LK - Lock
v LO - Label on
v MT - More text
v OP - Open
v PD - Prepare and describe
v PR - Prepare
v RE - Release
v RO - Rollback
v RT - Rename table
v RV - Revoke
v SC - Set connection
v SI - Select into
v SP - Set path
v SR - Set result set

216 DB2 UDB for iSeries Database Performance and Query Optimization V5R1
Table 33. QAQQQRYI - Summary Row for SQL Information (continued)
Column Name Description
QQSTOP (continued) v ST - Set transaction
v SV - Set variable
v UP - Update
v VI - Values into
QQODPI ODP implementation

R - Reusable ODP (ISV)

N - Non-reusable ODP (V2)
QQHVI Host variable implementation

I - Interface supplied values (ISV)

V - Host variables treated as constants (V2)
U - Table management row positioning (UP)
QQAPR Access plan rebuilt
A1 A table or member is not the same object as the one referenced when the
access plan was last built. Some reasons they could be different are:
v Object was deleted and recreated.
v Object was saved and restored.
v Library list was changed.
v Object was renamed.
v Object was moved.
v Object was overridden to a different object.
v This is the first run of this query after the object containing the query has
been restored.
A2 Access plan was built to use a reusable Open Data Path (ODP) and the
optimizer chose to use a non-reusable ODP for this call.
A3 Access plan was built to use a non-reusable Open Data Path (ODP) and the
optimizer chose to use a reusable ODP for this call.
A4 The number of rows in the table has changed by more than 10% since the
access plan was last built.
A5 A new index exists over one of the tables in the query.
A6 An index that was used for this access plan no longer exists or is no longer
valid.
A7 OS/400 Query requires the access plan to be rebuilt because of system
programming changes.
A8 The CCSID of the current job is different than the CCSID of the job that last
created the access plan.
A9 The value of one or more of the following is different for the current job than it
was for the job that last created this access plan:
v date format
v date separator
v time format
v time separator

Appendix B. Memory Resident Database Monitor: DDS 217

Table 33. QAQQQRYI - Summary Row for SQL Information (continued)
Column Name Description

AA The sort sequence table specified is different than the sort sequence table that
was used when this access plan was created.
AB Storage pool changed or DEGREE parameter of CHGQRYA command
changed.
AC The system feature DB2 multisystem has been installed or removed.
AD The value of the degree query attribute has changed.
AE A view is either being opened by a high level language or a view is being
materialized.
AF A user-defined type or user-defined function is not the same object as the one
referred to in the access plan.
B0 The options specified have changed as a result of the query options file
QAQQINI.
B1 The access plan was generated with a commitment control level that is
different in the current job.
B2 The access plan was generated with a static cursor answer set size that is
different than the previous access plan.
QQDACV Data conversion
N No.
0 Not applicable.
1 Lengths do not match.
2 Numeric types do not match.
3 C host variable is NUL-terminated.
4 Host variable or column is variable length and the other s not variable length.
5 CCSID conversion.
6 DRDA and NULL capable, variable length, contained in a partial row, derived
expression, or blocked fetch with not enough host variables.
7 Data, time, or timestamp column.
8 Too many host variables.
9 Target table of an insert is not an SQL table.
QQCTS Statement table scan usage count
QQCIU Statement index usage count
QQCIC Statement index creation count
QQCSO Statement sort usage count
QQCTF Statement temporary table count
QQCIA Statement index advised count
QQCAPR Statement access plan rebuild count
QQARSS Average result set size
QQC11 Reserved
QQC12 Reserved
QQC21 Reserved
QQC22 Reserved

218 DB2 UDB for iSeries Database Performance and Query Optimization V5R1
Table 33. QAQQQRYI - Summary Row for SQL Information (continued)
Column Name Description
QQI1 Reserved
QQI2 Reserved
QQC301 Reserved
QQC302 Reserved
QQC1000 Reserved

External table description (QAQQTEXT)

Table 34. QAQQTEXT - Summary Row for SQL Statement
Column Name Description
QQKEY Join column (unique per query) used to link rows for a single query together with
row identification
QQTIME Time row was created
QQSTTX Statement text
QQC11 Reserved
QQC12 Reserved
QQC21 Reserved
QQC22 Reserved
QQQI1 Reserved
QQI2 Reserved
QQC301 Reserved
QQC302 Reserved
QQ1000 Reserved

External table description (QAQQ3000)

Table 35. QAQQ3000 - Summary Row for Arrival Sequence
Column Name Description
QQKEY Join column (unique per query) used to link rows for a single query together with
row identification
QQTIME Time row was created
QQQDTN QDT number (unique per ODT)
QQQDTL QDT subquery nested level
QQMATN Materialized view QDT number
QQMATL Materialized view nested level
QQTLN Library
QQTFN Table
QQPTLN Physical library
QQPTFN Physical table
QQTOTR Total rows in table
QQREST Estimated number of rows selected

Appendix B. Memory Resident Database Monitor: DDS 219

Table 35. QAQQ3000 - Summary Row for Arrival Sequence (continued)
Column Name Description
QQAJN Estimated number of joined rows
QQEPT Estimated processing time, in seconds
QQJNP Join position - when available
QQJNDS Dataspace number
QQJNMT Join method - when available

NL - Nested loop
MF - Nested loop with selection
HJ - Hash join
QQJNTY Join type - when available

IN - Inner join
PO - Left partial outer join
EX - Exception join
QQJNOP Join operator - when available

EQ - Equal
NE - Not equal
GT - Greater than
GE - Greater than or equal
LT - Less than
LE - Less than or equal
CP - Cartesian product
QQDSS Dataspace selection

Y - Yes
N - No
QQIDXA Index advised

Y - Yes
N - No
QQRCOD Reason code

T1 - No indexes exist.
T2 - Indexes exist, but none could be used.
T3 - Optimizer chose table scan over available indexes.
QQLTLN Library-long
QQLTFN Table-long
QQLPTL Physical library-long
QQLPTF Table-long
QQIDXD Key columns for the index advised
QQC11 Reserved
QQC12 Reserved
QQC21 Reserved
QQC22 Reserved
QQI1 Number of advised key columns that use index scan-key positioning.
QQI2 Reserved

220 DB2 UDB for iSeries Database Performance and Query Optimization V5R1
Table 35. QAQQ3000 - Summary Row for Arrival Sequence (continued)
Column Name Description
QQC301 Reserved
QQC302 Reserved
QQ1000 Reserved

External table description (QAQQ3001)

Table 36. QQQ3001 - Summary Row for Using Existing Index
Column Name Description
QQKEY Join column (unique per query) used to link rows for a single query together
QQTIME Time row was created
QQQDTN QDT number (unique per QDT)
QQQDTL RQDT subquery nested levelelational database name
QQMATN Materialized view QDT number
QQMATL Materialized view nested level
QQTLN Library
QQTFN Table
QQPTLN Physical library
QQPTFN Physical table
QQILNM Index library
QQIFNM Index
QQTOTR Total rows in table
QQREST Estimated number of rows selected
QQFKEY Number of key positioning keys
QQKSEL Number of key selection keys
QQAJN Join position - when available
QQEPT Estimated processing time, in seconds
QQJNP Join position - when available
QQJNDS Dataspace number
QQJNMT Join method - when available

NL - Nested loop
MF - Nested loop with selection
HJ - Hash join
QQJNTY Join type - when available

IN - Inner join
PO - Left partial outer join
EX - Exception join

Appendix B. Memory Resident Database Monitor: DDS 221

Table 36. QQQ3001 - Summary Row for Using Existing Index (continued)
Column Name Description
QQJNOP Join operator - when available

EQ - Equal
NE - Not equal
GT - Greater than
GE - Greater than or equal
LT - Less than
LE - Less than or equal
CP - Cartesian product
QQIDXK Number of advised key columns that use index scan-key positioning
QQKP Index scan-key positioning

Y - Yes
N - No
QQKPN Number of key positioning columns
QQKS Index scan-key selection

Y - Yes
N - No
QQDSS Dataspace selection

Y - Yes
N - No
QQIDXA Index advised

Y - Yes
N - No
QQRCOD Reason code

I1 - Row selection
I2 - Ordering/Grouping
I3 - Row selection and
Ordering/Grouping
I4 - Nested loop join
I5 - Row selection using
bitmap processing
QQCST Constraint indicator

Y - Yes
N - No
QQCSTN Constraint name
QQLTLN Library-long
QQLTFN Table-long
QQLPTL Physical library-long
QQLPTF Table-long
QQLILN Index library – long
QQLIFN Index – long
QQIDXD Key columns for the index advised
QQC11 Reserved

222 DB2 UDB for iSeries Database Performance and Query Optimization V5R1
Table 36. QQQ3001 - Summary Row for Using Existing Index (continued)
Column Name Description
QQC12 Reserved
QQC21 Reserved
QQC22 Reserved
QQI1 Reserved
QQI2 Reserved
QQC301 Reserved
QQC302 Reserved
QQ1000 Reserved

External table description (QAQQ3002)

Table 37. QQQ3002 - Summary Row for Index Created
Column Name Description
QQKEY Join column (unique per query) used to link rows for a single query together
QQTIME Time row was created
QQQDTN QDT number (unique per QDT)
QQQDTL RQDT subquery nested levelelational database name
QQMATN Materialized view QDT number
QQMATL Materialized view nested level
QQTLN Library
QQTFN Table
QQPTLN Physical library
QQPTFN Physical table
QQILNM Index library
QQIFNM Index
QQNTNM NLSS table
QQNLNM NLSS library
QQTOTR Total rows in table
QQRIDX Number of entries in index created
QQREST Estimated number of rows selected
QQFKEY Number of index scan-key positioning keys
QQKSEL Number of index scan-key selection keys
QQAJN Estimated number of joined rows
QQJNP Join position - when available
QQJNDS Dataspace number
QQJNMT Join method - when available

NL - Nested loop
MF - Nested loop with selection
HJ - Hash join

Appendix B. Memory Resident Database Monitor: DDS 223

Table 37. QQQ3002 - Summary Row for Index Created (continued)
Column Name Description
QQJNTY Join type - when available

IN - Inner join
PO - Left partial outer join
EX - Exception join
QQJNOP Join operator - when available

EQ - Equal
NE - Not equal
GT - Greater than
GE - Greater than or equal
LT - Less than
LE - Less than or equal
CP - Cartesian product
QQIDXK Number of advised key columns that use index scan-key positioning
QQEPT Estimated processing time, in seconds
QQKP Index scan-key positioning

Y - Yes
N - No
QQKPN Number of index scan-key positioning columns
QQKS Index scan-key selection

Y - Yes
N - No
QQDSS Dataspace selection

Y - Yes
N - No
QQIDXA Index advised

Y - Yes
N - No
QQCST Constraint indicator

Y - Yes
N - No
QQCSTN Constraint name
QQRCOD Reason code

I1 - Row selection
I2 - Ordering/Grouping
I3 - Row selection and
Ordering/Grouping
I4 - Nested loop join
I5 - Row selection using
bitmap processing
QQTTIM Index create time
QQLTLN Library-long
QQLTFN Table-long

224 DB2 UDB for iSeries Database Performance and Query Optimization V5R1
Table 37. QQQ3002 - Summary Row for Index Created (continued)
Column Name Description
QQLPTL Physical library-long
QQLPTF Table-long
QQLILN Index library-long
QQLIFN Index-long
QQLNTN NLSS table-long
QQLNLN NLSS library-long
QQIDXD Key columns for the index advised
QQCRTK Key columns for index created
QQC11 Reserved
QQC12 Reserved
QQC21 Reserved
QQC22 Reserved
QQI1 Reserved
QQI2 Reserved
QQC301 Reserved
QQC302 Reserved
QQ1000 Reserved

External table description (QAQQ3003)

Table 38. QQQ3003 - Summary Row for Query Sort
Column Name Description
QQKEY Join column (unique per query) used to link rows for a single query together
QQTIME Time row was created
QQQDTN QDT number (unique per QDT)
QQQDTL RQDT subquery nested levelelational database name
QQMATN Materialized view QDT number
QQMATL Materialized view nested level
QQTTIM Sort time
QQRSS Number of rows selected or sorted
QQSIZ Size of sort space
QQPSIZ Pool size
QQPID Pool id
QQIBUF Internal sort buffer length
QQEBUF External sort buffer length

Appendix B. Memory Resident Database Monitor: DDS 225

 Table 38. QQQ3003 - Summary Row for Query Sort (continued)
Column Name Description
QQRCOD Reason code
F1 Query contains grouping columns (Group By) from more than one table, or
contains grouping columns from a secondary table of a join query that cannot
be reordered.
F2 Query contains ordering columns (Order By) from more than one table, or
contains ordering columns from a secondary table of a join query that cannot
be reordered.
F3 The grouping and ordering columns are not compatible.
F4 DISTINCT was specified for the query.
F5 UNION was specified for the query.
F6 Query had to be implemented using a sort. Key length of more than 2000
bytes or more than 120 columns specified for ordering.
F7 Query optimizer chose to use a sort rather than an index to order the results
of the query.
F8 Perform specified row selection to minimize I/O wait time.
FC The query contains grouping fields and there is a read trigger on at least one
of the physical files in the query.
QQC11 Reserved
QQC12 Reserved
QQC21 Reserved
QQC22 Reserved
QQI1 Reserved
QQI2 Reserved
QQC301 Reserved
QQC302 Reserved
QQ1000 Reserved

External table description (QAQQ3004)

| Table 39. QQQ3004 - Summary Row for Temporary Table
| Column Name Description
| QQKEY Join column (unique per query) used to link rows for a single query together
| QQTIME Time row was created
| QQQDTN QDT number (unique per QDT)
| QQQDTL RQDT subquery nested levelelational database name
| QQMATN Materialized view QDT number
| QQMATL Materialized view nested level
| QQTLN Library
| QQTFN Table
| QQTTIM Temporary table create time
| QQTMPR Number of rows in temporary

226 DB2 UDB for iSeries Database Performance and Query Optimization V5R1
| Table 39. QQQ3004 - Summary Row for Temporary Table (continued)
| Column Name Description
| QQRCOD Reason code
| F1 Query contains grouping columns (Group By) from more than one table, or
| contains grouping columns from a secondary table of a join query that cannot
| be reordered.
| F2 Query contains ordering columns (Order By) from more than one table, or
| contains ordering columns from a secondary table of a join query that cannot
| be reordered.
| F3 The grouping and ordering columns are not compatible.
| F4 DISTINCT was specified for the query.
| F5 UNION was specified for the query.
| F6 Query had to be implemented using a sort. Key length of more than 2000
| bytes or more than 120 columns specified for ordering.
| F7 Query optimizer chose to use a sort rather than an index to order the results
| of the query.
| F8 Perform specified row selection to minimize I/O wait time.
| F9 The query optimizer chose to use a hashing algorithm rather than an access
| path to perform the grouping for the query.
| FA The query contains a join condition that requires a temporary file.
| FB The query optimizer creates a run-time temporary file in order to implement
| certain correlated group by queries.
| FC The query contains grouping fields and there is a read trigger on at least one
| of the physical files in the query.
| H1 Table is a join logical file and its join type does not match the join type
| specified in the query.
| H2 Format specified for the logical table references more than one base table.
| H3 Table is a complex SQL view requiring a temporary results of the SQL view.
| H4 For an update-capable query, a subselect references a column in this table
| which matches one of the columns being updated.
| H5 For an update-capable query, a subselect references an SQL view which is
| based on the table being updated.
| H6 For a delete-capable query, a subselect references either the table from which
| rows are to be deleted, an SQL view, or an index based on the table from
| which rows are to be deleted.
| QQDFVL Default values may be present in temporary

| Y - Yes
| N - No
| QQLTLN Library-long
| QQLTFN Table-long
| QQC11 Reserved
| QQC12 Reserved
| QQC21 Reserved
| QQC22 Reserved
| QQI1 Reserved

Appendix B. Memory Resident Database Monitor: DDS 227

| Table 39. QQQ3004 - Summary Row for Temporary Table (continued)
| Column Name Description
| QQI2 Reserved
| QQC301 Reserved
| QQC302 Reserved
| QQ1000 Reserved

External table description (QAQQ3007)

Table 40. QQQ3007 - Summary Row for Optimizer Information
Column Name Description
QQKEY Join column (unique per query) used to link rows for a single query together
QQTIME Time row was created
QQQDTN QDT number (unique per QDT)
QQQDTL RQDT subquery nested levelelational database name
QQMATN Materialized view QDT number
QQMATL Materialized view nested level
QQTLN Library
QQTFN Table
QQPTLN Physical library
QQPTFN Table
QQTOUT Optimizer timed out

Y - Yes
N - No.
QQIRSN Reason code
QQLTLN Library-long
QQLTFN Table-long
QQPTL Physical library-long
QQPTF Table-long
QQIDXN Index names
QQC11 Reserved
QQC12 Reserved
QQC21 Reserved
QQC22 Reserved
QQI1 Reserved
QQI2 Reserved
QQC301 Reserved
QQC302 Reserved
QQ1000 Reserved

228 DB2 UDB for iSeries Database Performance and Query Optimization V5R1
External table description (QAQQ3008)
Table 41. QQQ3008 - Summary Row for Subquery Processing
Column Name Description
QQKEY Join column (unique per query) used to link rows for a single query together
QQTIME Time row was created
QQQDTN QDT number (unique per QDT)
QQQDTL RQDT subquery nested levelelational database name
QQMATN Materialized view QDT number
QQMATL Materialized view nested level
QQORGQ Materialized view QDT number
QQMRGQ Materialized view nested level
QQC11 Reserved
QQC12 Reserved
QQC21 Reserved
QQC22 Reserved
QQI1 Reserved
QQI2 Reserved
QQC301 Reserved
QQC302 Reserved
QQ1000 Reserved

External table description (QAQQ3010)

Table 42. QQQ3010 - Summary Row for Host Variable and ODP Implementation
Column Name Description
QQKEY Join column (unique per query) used to link rows for a single query together
QQTIME Time row was created
QQHVAR Host variable values
QQC11 Reserved
QQC12 Reserved
QQC21 Reserved
QQC22 Reserved
QQI1 Reserved
QQI2 Reserved
QQC301 Reserved
QQC302 Reserved

Appendix B. Memory Resident Database Monitor: DDS 229

230 DB2 UDB for iSeries Database Performance and Query Optimization V5R1
Index
A CLOSQLCSR parameter
using 117
access method
command
bitmap processing access 24
CHGQRYA 81
hashing access 23
CHGQRYA command 81
index only access 21
QAQQINI 82
index scan-key positioning 15
QAQQINI command 82
index scan-key selection 13
command (CL)
index-from-index 22
Change Query Attribute (CHGQRYA) command 11
parallel index scan-key positioning 19
Change Query Attributes (CHGQRYA) 60
parallel index scan-key selection access method 14
CHGQRYA (Change Query Attribute) command 11
parallel preload 22
CHGQRYA (Change Query Attributes) 60
parallel table prefetch 10
Delete Override (DLTOVR) 114
parallel table scan 11
Display Job (DSPJOB) 60
row selection method 3
Display Journal (DSPJRN) 116
sort access 28
DLTOVR (Delete Override) 114
summary table 5
DSPJOB (Display Job) 60
table scan 8
DSPJRN (Display Journal) 116
access path
Override Database File (OVRDBF) 121
index 3
OVRDBF (Override Database File) 121
sequential 3
Print SQL Information (PRTSQLINF) 60, 91
access plan
QAQQINI 84
validation 34
Start Database Monitor (STRDBMON) 60
access plan rebuilt
STRDBMON (Start Database Monitor) 60
summary row 176, 177, 178, 179, 180
Trace Job (TRCJOB) 116
advisor
TRCJOB (Trace Job) 116
query optimizer index 72
commands
ALLOCATE clause
End Database Monitor (ENDDBMON) 71
performance implications 127
Start Database Monitor (STRDBMON) 70
allow copy data (ALWCPYDTA) parameter 112
commitment control
ALWCPYDTA (allow copy data) parameter 112
displaying 60
ALWCPYDTA parameter
controlling parallel processing 92
effect on query optimimizer 32
copy of the data
using to improve performance 126
cost estimation
B query optimizer 32
bitmap created cursor
summary row 196, 197, 198 positions
bitmap merge retaining across program call 116
summary row 199, 200, 201 rules for retaining 116
bitmap processing access method 24 using to improve performance 116
blocking consideration
using, affect on performance 121
blocking, SQL
improving performance 121
D
data
paging
interactively displayed to improve
C performance 123
calls, number selecting from multiple tables
using affect on performance 50
FETCH statement 120 data path, open 66
cancelling a query 90 database monitor
Change Query Attribute (CHGQRYA) command 11 end 71
Change query attributes 81 examples 73, 76
Change Query Attributes (CHGQRYA) command 60 logical file DDS 136
changing physical file DDS 129
query options file 84 start 70
CHGQRYA (Change Query Attributes) command 60 database monitor performance rows 71

© Copyright IBM Corp. 2000, 2001 231

database query performance
monitoring 69
F
file
dataspace
query options 84
definition 4
filter factors, default
DDS
in query optimization 33
database monitor logical file 136
database monitor physical file 129
default filter factors 33
definitions G
bitmap processing method 24 generic query information
dataspace 4 summary row 189, 190, 191, 192
default filter factors 33 governor 89
dial 35 *DFT 90
hashing access method 23 *RQD 90
implementation cost 32 *SYSRPYL 90
index 3 CHGQRYA 89
index only access method 21 JOB 90
index scan-key positioning access method 15 QRYTIMLMT 89
index scan-key selection access method 13 time limit 90
index-from-index access method 22 grouping
isolatable 45 summary row 211, 212, 213
miniplan 34 grouping optimization 50
open data path 66
parallel index scan-key positioning access
method 19 H
parallel index scan-key selection access method 14 hash join 36
parallel table prefetch access method 10 hash table
parallel table scan method 11 summary row 202, 203, 204
primary table 35 hashing access method 23
secondary tables 35 host variable and ODP implementation
sort access method 28 summary row 185, 186
symmetrical multiprocessing 4
table scan 3
Delete Override (DLTOVR) command 114 I
deleted rows improving performance 112, 126
getting rid of using REUSEDLT(*YES) 8 blocking, using 121
getting rid of using RGZPFM 8 join queries 47
detail row paging interactively displayed data 123
rows retrieved 194, 195 PREPARE statement 123
Display Job (DSPJOB) command 60 retaining cursor positions across program call 116
Display Journal (DSPJRN) command 116 SELECT statements, using effectively 122
distinct processing selecting data from multiple tables 50
summary row 205, 206 SQL blocking 121
DSPJOB (Display Job) command 60 using
close SQL cursor (CLOSQLCSR) 116
FETCH FOR n ROWS 120
E INSERT n ROWS 121
End Database Monitor (ENDDBMON) command 71 parameter passing techniques 111
ENDDBMON (end database monitor) command 71 precompile options 125
examples index
database monitor 73, 76 access path 3
governor 91 columns used for keys 3
index 97 creating
performance analysis 73, 74, 75 from another index 22
reducing the number of open database temporary keyed
operation 114 from index 44
selecting data from multiple tables 50 from the table 44
using effectively, example 97
index advisor
query optimizer 72

232 DB2 UDB for iSeries Database Performance and Query Optimization V5R1
index created multiple
summary row 161, 162, 163, 164 table
index only access method 21 improving performance when selecting data
index scan-key positioning from 50
access method 15
index scan-key selection
access method 13 N
index-from-index nested loop join 35
access method 22 number of calls
indexes using a FETCH statement 120
using with sort sequence 100 number of open database operations
information messages improving performance by reducing 114
open data path 66, 67
performance 61, 66
INSERT n ROWS O
improving performance 121 ODP implementation and host variable
interactively displayed data, paging summary row 185, 186
affect on performance 123 open
closing 114
determining number 116
J effect on performance 114
JOB 90 reducing number 114
join open data path
hash 36 definition 66
optimization 35 information messages 66
join optimization OPNQRYF (Open Query File) command 59
performance tips 47 optimization 31
join order grouping 50
optimization 40 join 35
join position 64 join order 40
join secondary dials nested loop join 35
costing 41 OPTIMIZE FOR n ROWS clause
effect on query optimizer 32
optimizer
K operation 31
key range estimate 34 query index advisor 72
optimizer timed out
summary row 181, 182, 183
L options, precompile
limit, time 90 improving performance by using 125
live data output
using to improve performance 112 all queries that performed table scans 74
locks SQL queries that performed table scans 73
analyzing 60 Override Database File (OVRDBF) command 121
logical file DDS
database monitor 136
long object names P
performance 125 page fault 4
paging
interactively displayed data 123
M parallel index scan-key positioning access method 19
message parallel index scan-key selection access method 14
cause and user response 60 parallel preload
open data path information 66, 67 index-based 22
performance information 61, 66 table-based 22
running in debug mode 60 parallel processing
monitor (ENDDBMON) command, end database 71 controlling
monitoring in jobs (CHGQRYA command) 93
database query performance 69 system wide (QQRYDEGREE) value 92
parallel table prefetch
access method 10

Index 233
parallel table scan
access method 11
Q
QAQQINI 84
parameter passing
QDT 34
differences 112
QRYTIMLMT parameter
parameters, command
CHGQRYA (Change Query Attributes) command 60
ALWCPYDTA (allow copy data) 112, 126
query
CLOSQLCSR (close SQL cursor) 116
cancelling 90
path, open data 66
Query Definition Template (QDT) 34
performance 31
query optimizer 31
information messages 61, 66
cost estimation 32
monitoring 59
decision-making rules 31
monitoring query 69
default filter factors 33
open data path messages 66, 67
optimization goals 32
OPNQRYF 59
query optimizer index advisor 72
optimizing 59
query options
tools 59
file 84
using long object names 125
query options file
performance analysis
changing 84
example 1 73
Query options file 82
example 2 74
query performance
example 3 75
monitoring 69
performance considerations 91, 114
query sort
performance improvement
summary row 166, 167, 168
blocking, using 121
query time limit 90
paging interactively displayed data 123
PREPARE statement 123
reducing number of open database operation 114
retaining cursor positions across program call 116 R
SELECT statements, using effectively 122 reducing number of open database operations
selecting data from multiple tables 50 improving performance, example 114
SQL blocking 121 Reorganize Physical File Member (RGZPFM) command
using copy of the data 126 effect on variable-length columns 128
using INSERT n ROWS 121 getting rid of deleted rows 8
using live data 112 resource
using precompile options 125 optimization 31
performance rows retaining cursor positions
database monitor 71 across program call
physical file DDS improving performance 116
database monitor 129 all program calls
pre-fetching 8 rules 117
precompile options rows
improving performance, using 125 database monitor performance 71
precompiler rows retrieved
passing detail row 194, 195
host variables 111 ROWS, INSERT n
precompiler command improving performance 121
default 116 rule
precompiler parameter retaining cursor positions
ALWCPYDTA 112 program calls 117
CLOSQLCSR 117
predicate
transitive closure 45 S
Predictive Query Governor 89 SELECT statement
PREPARE statement using effectively to improve performance 122
improving performance 123 selecting
Print SQL Information (PRTSQLINF) 60, 91 data from multiple tables 50
problems setting query time limit 91
join query performance 46 sort access method 28
program calls sort sequence
rules for retaining cursor positions 117 using indexes 100

234 DB2 UDB for iSeries Database Performance and Query Optimization V5R1
SQL blocking table scan (continued)
improving performance 121 access method 8
SQL information summary row 149, 150, 151
summary row 140, 141, 142, 143, 144, 145, 146, table scans
147, 215, 216, 217, 218, 219 output for all queries 74
Start Database Monitor (STRDBMON) command 60, output for SQL queries 73
70 temporary index 44
statements temporary table
FETCH summary row 170, 171, 172
FOR n ROWS 120 tools
number of calls 120 performance 59
INSERT Trace Job (TRCJOB) command 116
n ROWS 121 transitive closure 45
PREPARE
improving performance 123
STRDBMON (Start Database Monitor) command 60, U
70 using
STRDBMON/ENDDBMON commands a copy of the data 112, 126
summary row 193 allow copy data (ALWCPYDTA) 112, 126
structure parameter passing 112 close SQL cursor (CLOSQLCSR) 112, 117
subquery merge FETCH statement 120
summary row 207, 208, 209 parameter passing techniques
subquery processing performance improvement 111
summary row 184, 185 using existing index
summary row summary row 154, 155, 156, 157, 219, 220, 221,
access plan rebuilt 176, 177, 178, 179, 180 222, 223, 224, 225, 226, 227, 228, 229
bitmap created 196, 197, 198 using JOB parameter 91
bitmap merge 199, 200, 201 using SQL
distinct processing 205, 206 application programs 31
generic query information 189, 190, 191, 192
grouping 211, 212, 213
hash table 202, 203, 204 V
host variable and ODP implementation 185, 186 variable-length data
index created 161, 162, 163, 164 tips 127
optimizer timed out 181, 182, 183
query sort 166, 167, 168
SQL information 140, 141, 142, 143, 144, 145, 146,
147
STRDBMON/ENDDBMON commands 193
subquery merge 207, 208, 209
subquery processing 184, 185
table locked 174, 175
table scan 149, 150, 151
temporary table 170, 171, 172
using existing index 154, 155, 156, 157
summary rows
SQL information 215, 216, 217, 218, 219
using existing index 219, 220, 221, 222, 223, 224,
225, 226, 227, 228, 229
symmetrical multiprocessing 4

T
table
data management methods 5
multiple
improving performance when selecting data
from 50
table locked
summary row 174, 175
table scan 3

Index 235
236 DB2 UDB for iSeries Database Performance and Query Optimization V5R1



Printed in U.S.A.