Performance and Tuning Series: Query

Processing and Abstract Plans

Adaptive Server® Enterprise

15.7
DOCUMENT ID: DC00743-01-1570-01

LAST REVISED: September 2011

Copyright © 2011 by Sybase, Inc. All rights reserved.

This publication pertains to Sybase software and to any subsequent release until otherwise indicated in new editions or technical notes.
Information in this document is subject to change without notice. The software described herein is furnished under a license agreement,
and it may be used or copied only in accordance with the terms of that agreement.

To order additional documents, U.S. and Canadian customers should call Customer Fulfillment at (800) 685-8225, fax (617) 229-9845.

Customers in other countries with a U.S. license agreement may contact Customer Fulfillment via the above fax number. All other
international customers should contact their Sybase subsidiary or local distributor. Upgrades are provided only at regularly scheduled
software release dates. No part of this publication may be reproduced, transmitted, or translated in any form or by any means, electronic,
mechanical, manual, optical, or otherwise, without the prior written permission of Sybase, Inc.

Sybase trademarks can be viewed at the Sybase trademarks page at http://www.sybase.com/detail?id=1011207. Sybase and the marks listed
are trademarks of Sybase, Inc. ® indicates registration in the United States of America.

SAP and other SAP products and services mentioned herein as well as their respective logos are trademarks or registered trademarks of
SAP AG in Germany and in several other countries all over the world.

Java and all Java-based marks are trademarks or registered trademarks of Sun Microsystems, Inc. in the U.S. and other countries.

Unicode and the Unicode Logo are registered trademarks of Unicode, Inc.

IBM and Tivoli are registered trademarks of International Business Machines Corporation in the United States, other countries, or both.

All other company and product names mentioned may be trademarks of the respective companies with which they are associated.

Use, duplication, or disclosure by the government is subject to the restrictions set forth in subparagraph (c)(1)(ii) of DFARS 52.227-7013
for the DOD and as set forth in FAR 52.227-19(a)-(d) for civilian agencies.

Sybase, Inc., One Sybase Drive, Dublin, CA 94568.

Contents

CHAPTER 1 Understanding Query Processing ................................................. 1

Query optimizer ................................................................................ 3
Factors analyzed in optimizing queries ..................................... 5
Transformations for query optimization ..................................... 6
Handling search arguments and useful indexes ..................... 10
Handling joins .......................................................................... 12
Optimization goals.......................................................................... 15
Limiting the time spent optimizing a query .............................. 16
Parallelism...................................................................................... 17
Optimization issues ........................................................................ 17
Lava query execution engine ......................................................... 20
Lava query plans ..................................................................... 21
How update operations are performed........................................... 27
Direct updates ......................................................................... 27
Deferred updates..................................................................... 30
Deferred index inserts ............................................................. 31
Restrictions on update modes through joins ........................... 33
Optimizing updates.................................................................. 34
Using sp_sysmon while tuning updates .................................. 36

CHAPTER 2 Using showplan............................................................................. 37

Displaying a query plan .................................................................. 37
Query plans in Adaptive Server Enterprise 15.0 and later ...... 38
Using set showplan with noexec .................................................... 39
Statement-level output ................................................................... 44
Query plan shape ........................................................................... 47
Query plan operators............................................................... 52
EMIT operator ......................................................................... 52
SCAN operator ........................................................................ 52
FROM cache message............................................................ 52
FROM or LIST ......................................................................... 53
FROM TABLE ......................................................................... 54
Union operators.............................................................................. 90
UNION ALL operator ............................................................... 90

Performance and Tuning Series: Query Processing and Abstract Plans iii
Contents

MERGE UNION operator ........................................................ 91

HASH UNION.......................................................................... 91
SCALAR AGGREGATE operator............................................ 93
RESTRICT operator ................................................................ 94
SORT operator ........................................................................ 94
STORE operator...................................................................... 95
SEQUENCER operator ........................................................... 97
REMOTE SCAN operator........................................................ 99
SCROLL operator.................................................................... 99
RID JOIN operator................................................................. 100
SQLFILTER operator ............................................................ 102
EXCHANGE operator............................................................ 104
INSTEAD-OF TRIGGER operators .............................................. 106
INSTEAD-OF TRIGGER operator......................................... 107
CURSOR SCAN operator ..................................................... 108
deferred_index and deferred_varcol messages .................... 110

CHAPTER 3 Displaying Query Optimization Strategies and Estimates....... 111

set commands for text format messages ..................................... 111
set commands for XML format messages.................................... 112
Using show_execio_xml to diagnose query plans................. 114
Showing cached plans in XML .............................................. 116
Diagnostic usage scenarios ......................................................... 119
Permissions for set commands .................................................... 122
Analyzing dynamic parameters .................................................... 122
Dynamic parameter example analysis .................................. 123

CHAPTER 4 Finding Slow Running Queries .................................................. 125

Saving diagnostics to a trace file.................................................. 125
Set options that save diagnostic information to a trace file ... 127
Which sessions are being traced? ........................................ 128
Rebinding a trace .................................................................. 129
Displaying SQL text...................................................................... 129
Retaining session settings ........................................................... 132

CHAPTER 5 Parallel Query Processing .......................................................... 133

Vertical, horizontal, and pipelined parallelism .............................. 133
Queries that benefit from parallel processing............................... 134
Enabling parallelism ..................................................................... 135
number of worker processes ................................................. 135
max parallel degree............................................................... 136
max resource granularity....................................................... 136

iv Adaptive Server Enterprise

 Contents

max repartition degree .......................................................... 137

max scan parallel degree ...................................................... 138
prod-consumer overlap factor ............................................... 138
min pages for parallel scan ................................................... 138
max query parallel degree ..................................................... 139
Controlling parallelism at the session level .................................. 139
set command examples ........................................................ 140
Controlling query parallelism ........................................................ 141
Query-level parallel clause examples.................................... 141
Using parallelism selectively ........................................................ 141
Using parallelism with large numbers of partitions....................... 143
When parallel query results differ................................................. 144
Queries that use set rowcount............................................... 145
Queries that set local variables ............................................. 145
Understanding parallel query plans.............................................. 146
Adaptive Server parallel query execution model.......................... 148
EXCHANGE operator............................................................ 148
Using parallelism in SQL operations ..................................... 154
Partition elimination ............................................................... 197
Partition skew ........................................................................ 198
Why queries do not run in parallel......................................... 199
Runtime adjustment .............................................................. 199
Recognizing and managing runtime adjustments ................. 200

CHAPTER 6 Eager and Lazy Aggregation ..................................................... 203

Overview ...................................................................................... 203
Eager aggregation ................................................................. 204
Aggregation and query processing .............................................. 205
Examples ..................................................................................... 208
Using eager aggregation .............................................................. 215
Enabling eager aggregation .................................................. 215
Checking for eager aggregation ............................................ 215
Forcing eager aggregation with abstract plans ..................... 218

CHAPTER 7 Controlling Optimization ............................................................ 221

Special optimizing techniques...................................................... 221
Viewing current optimizer settings ............................................... 222
Setting the optimization level................................................. 225
Optimizer Diagnostic Utility .......................................................... 229
Configuring Adaptive Server to run sp_opt_querystats......... 230
Running sp_opt_querystats................................................... 230
Specifying query processor choices............................................. 231
Specifying table order in joins ...................................................... 232

Performance and Tuning Series: Query Processing and Abstract Plans v

Contents

Specifying the number of tables considered by the query processor..

233
Specifying query index ................................................................. 234
Specifying I/O size in a query....................................................... 236
Index type and large I/O size................................................. 237
When prefetch specification cannot be followed ................... 238
setting prefetch...................................................................... 239
Specifying cache strategy ............................................................ 239
In select, delete, and update statements............................... 240
Controlling large I/O and cache strategies ................................... 241
Getting information on cache strategies................................ 241
Asynchronous log service ............................................................ 241
Understanding the user log cache (ULC) architecture .......... 243
When to use ALS .................................................................. 243
Using the ALS ....................................................................... 244
Enabling and disabling merge joins ............................................. 244
Enabling and disabling hash joins ................................................ 245
Enabling and disabling join transitive closure .............................. 245
Controlling literal parameterization............................................... 246
Suggesting a degree of parallelism for a query............................ 248
Query level parallel clause examples.................................... 250
Optimization goals........................................................................ 250
Setting optimization goals ..................................................... 251
Optimization criteria ..................................................................... 252
Limiting optimization time ............................................................. 255
Controlling parallel optimization ................................................... 256
number of worker processes ................................................. 257
Specifying the number of worker processes available for parallel
processing ...................................................................... 257
max resource granularity....................................................... 257
max repartition degree .......................................................... 258
Concurrency optimization for small tables ................................... 258
Changing the locking scheme ............................................... 259

CHAPTER 8 Optimization for Cursors ............................................................ 261

Definition ...................................................................................... 261
Set-oriented versus row-oriented programming .................... 262
Example ................................................................................ 263
Resources required at each stage ............................................... 264
Memory use and execute cursors ......................................... 266
Cursor modes............................................................................... 266
Index use and requirements for cursors....................................... 267
Allpages-locked tables .......................................................... 267
Data-only-locked tables......................................................... 268

vi Adaptive Server Enterprise

 Contents

Comparing performance with and without cursors....................... 269

Sample stored procedure without a cursor............................ 269
Sample stored procedure with a cursor................................. 270
Cursor versus noncursor performance comparison .............. 271
Locking with read-only cursors..................................................... 272
Isolation levels and cursors.......................................................... 273
Partitioned heap tables and cursors............................................. 274
Optimizing tips for cursors............................................................ 274
Optimizing for cursor selects using a cursor ......................... 275
Using union instead of or clauses or in lists .......................... 275
Declaring the cursor’s intent .................................................. 275
Specifying column names in the for update clause ............... 276
Using set cursor rows ............................................................ 277
Keeping cursors open across commits and rollbacks ........... 277
Opening multiple cursors on a single connection .................. 278

CHAPTER 9 Query Processing Metrics.......................................................... 279

Overview ...................................................................................... 279
Executing QP metrics................................................................... 280
Accessing metrics ........................................................................ 280
sysquerymetrics view ............................................................ 280
Using metrics ............................................................................... 282
Examples............................................................................... 283
Clearing metrics ........................................................................... 284
Restricting query metrics capture................................................. 285
Understanding the UID in sysquerymetrics.................................. 286

CHAPTER 10 Using Statistics to Improve Performance................................. 287

Statistics maintained in Adaptive Server...................................... 287
Importance of statistics ................................................................ 288
Nonbinary character set histogram interpolation ................... 289
Updating statistics ........................................................................ 290
Adding statistics for unindexed columns ............................... 290
Limitations for updating statistics on proxy tables and views 291
update statistics commands .................................................. 291
Using sampling for update statistics...................................... 293
Automatically updating statistics .................................................. 294
datachange function .............................................................. 295
Configuring automatic update statistics ....................................... 297
Using Job Scheduler to update statistics .............................. 297
Examples of updating statistics with datachange .................. 299
Column statistics and statistics maintenance............................... 300
Creating and updating column statistics ...................................... 302

Performance and Tuning Series: Query Processing and Abstract Plans vii
 When additional statistics may be useful............................... 303
Adding statistics for a column with update statistics.............. 305
Adding statistics for minor columns with update index statistics ..
305
Adding statistics for all columns with update all statistics...... 306
Choosing step numbers for histograms........................................ 306
Choosing a step number........................................................ 307
Scan types, sort requirements, and locking.................................. 307
Sorts for unindexed or nonleading columns .......................... 308
Locking, scans, and sorts during update index statistics....... 308
Locking, scans and sorts during update all statistics............. 309
Using the with consumers clause .......................................... 309
Reducing the impact of update statistics on concurrent processes
309
Using the delete statistics command ............................................ 310
When row counts may be inaccurate ........................................... 311

CHAPTER 11 Introduction to Abstract Plans................................................... 313

Overview....................................................................................... 313
Managing abstract plans .............................................................. 314
Relationship between query text and query plans........................ 315
Limits of options for influencing query plans.......................... 315
Full versus partial plans................................................................ 316
Creating a partial plan............................................................ 317
Abstract plan groups..................................................................... 318
How abstract plans are associated with queries .......................... 318
Abstract plans in cached statements ..................................... 319

CHAPTER 12 Creating and Using Abstract Plans ........................................... 321

Using set commands to capture and associate plans .................. 321
Enabling plan capture mode with set plan dump ................... 322
Associating queries with stored plans.................................... 323
Using replace mode during plan capture ............................... 323
Using dump, load, and replace modes simultaneously ......... 324
Compile-time changes for some set parameters ................... 326
set plan exists check option.......................................................... 327
Using other set options with abstract plans .................................. 328
Using show_abstract_plan to view plans............................... 328
Using showplan ..................................................................... 329
Using noexec ......................................................................... 329
Using fmtonly ......................................................................... 330
Using forceplan...................................................................... 330
Server-wide abstract plan capture and association modes .......... 330

viii Adaptive Server Enterprise

 Creating plans using SQL............................................................. 331
Using create plan................................................................... 331
Using the plan clause ............................................................ 333

CHAPTER 13 Abstract Query Plan Guide......................................................... 335

Overview....................................................................................... 335
Abstract plan language .......................................................... 336
Identifying tables.................................................................... 339
Identifying indexes ................................................................. 341
Specifying join order .............................................................. 341
Specifying the join type.......................................................... 345
Specifying partial plans and hints .......................................... 346
Creating abstract plans for subqueries .................................. 349
Abstract plans for materialized processing of views .............. 356
Abstract plans for queries containing aggregates.................. 357
Abstract plans for queries containing unions ......................... 358
Using abstract plans when queries need ordering................. 359
Specifying the reformatting strategy ...................................... 360
Specifying the OR strategy .................................................... 361
When the store operator is not specified ............................... 361
Abstract plans for parallel processing.................................... 361
Tips on writing abstract plans ....................................................... 363
Using abstract plans at the query level......................................... 363
Operator name alignment for abstract plan and optimizer criteria
365
Extending the optimizer criteria set syntax ............................ 366
Comparing plans before and after ................................................ 366
Effects of enabling server-wide capture mode....................... 367
Time and space to copy plans ............................................... 368
Abstract plans for stored procedures............................................ 368
Procedures and plan ownership ............................................ 369
Procedures with variable execution paths and optimization .. 369
Ad hoc queries and abstract plans ............................................... 370

CHAPTER 14 Managing Abstract Plans with System Procedures................. 373

Managing an abstract plan group ................................................. 373
Creating a group .................................................................... 373
Dropping a group ................................................................... 374
Getting information about a group ......................................... 374
Renaming a group ................................................................. 376
Finding abstract plans .................................................................. 377
Managing individual abstract plans .............................................. 377
Viewing a plan ....................................................................... 378

Performance and Tuning Series: Query Processing and Abstract Plans ix

 Copying a plan to another group ........................................... 378
Dropping an individual abstract plan...................................... 379
Comparing two abstract plans ............................................... 379
Changing an existing plan ..................................................... 380
Managing all plans in a group....................................................... 381
Copying all plans in a group .................................................. 381
Comparing all plans in a group .............................................. 382
Dropping all abstract plans in a group ................................... 384
Importing and exporting groups of plans ...................................... 385
Exporting plans to a user table .............................................. 385
Importing plans from a user table .......................................... 385

Index............................................................................................................................................ 387

x Adaptive Server Enterprise

CH A PTE R 1 Understanding Query Processing

This chapter provides an overview of the query processor in Adaptive

Server® Enterprise.
Topic Page
Query optimizer 3
Optimization goals 15
Parallelism 17
Optimization issues 17
Lava query execution engine 20

The query processor is designed to process queries you specify. The

processor yields highly efficient query plans that execute using minimal
resources, and ensure that results are consistent and correct.
To process a query efficiently, the query processor uses:
• The specified query
• Statistics about the tables, indexes, and columns named in the query
• Configurable variables
The query processor has to execute several steps, using several modules,
to successfully process a query:

Performance and Tuning Series: Query Processing and Abstract Plans 1

 Figure 1-1: : Query processor modules

1: Parser
2:Normalization
3: Preprocessor

4: Optimizer

5: Code generator

6: Procedural execution engine 6: Query execution

engine
1 The parser converts the text of the SQL statement to an internal
representation called a query tree.
2 This query tree is normalized. This involves determining column and table
names, transforming the query tree into conjugate normal form (CNF), and
resolving datatypes. At this point, you can determine if the statement may
benefit from using the statement cache.
3 The preprocessor transforms the query tree for some types of SQL
statements, such as SQL statements with subqueries and views, to a more
efficient query tree.
4 The optimizer analyzes the possible combinations of operations (join
ordering, access and join methods, parallelism) to execute the SQL
statement, and selects an efficient one based on the cost estimates of the
alternatives.
5 The code generator converts the query plan generated by the optimizer
into a format more suitable for the query execution engine.
6 The procedural engine executes command statements such as create table,
execute procedure, and declare cursor directly. For data manipulation
language (DML) statements, such as select, insert, delete, and update, the
engine sets up the execution environment for all query plans and calls the
query execution engine.
7 The query execution engine executes the ordered steps specified in the
query plan provided by the code generator.

2 Adaptive Server Enterprise

 CHAPTER 1 Understanding Query Processing

Query optimizer
The query optimizer provides speed and efficiency for online transaction
processing (OLTP) and operational decision-support systems (DSS)
environments. You can choose an optimization strategy that best suits your
query environment.
The query optimizer is self-tuning, and requires fewer interventions than
versions of Adaptive Server Enterprise ealier than 15.0. It relies infrequently
on worktables for materialization between steps of operations; however, the
query optimizer may use more worktables when it determines that hash and
merge operations are more effective.
Some of the key features in the release 15.0 query optimizer include support
for:
• New optimization techniques and query execution operator supports that
enhance query performance, such as:
• On-the-fly grouping and ordering operator support using in-memory
sorting and hashing for queries with group by and order by clauses
• hash and merge join operator support for efficient join operations

• index union and index intersection strategies for queries with predicates
on different indexes
Table 1-1 on page 4 is a list of optimization techniques and operator
support provided in Adaptive Server Enterprise. Many of these techniques
map directly to the operators supported in the query execution. See “Lava
query execution engine” on page 20.
• Improved index selection, especially for joins with or clauses, and joins
with and search arguments (SARGs) with mismatched but compatible
datatypes
• Improved costing that employs join histograms to prevent inaccuracies that
might otherwise arise due to data skews in joining columns
• New cost-based pruning and timeout mechanisms in join ordering and plan
strategies for large, multiway joins, and for star and snowflake schema
joins
• New optimization techniques to support data and index partitioning
(building blocks for parallelism) that are especially beneficial for very
large data sets
• Improved query optimization techniques for vertical and horizontal
parallelism. See Chapter 5, “Parallel Query Processing.”

Performance and Tuning Series: Query Processing and Abstract Plans 3

Query optimizer

• Improved problem diagnosis and resolution through:

• Searchable XML format trace outputs
• Detailed diagnostic output from new set commands. See Chapter 3,
“Displaying Query Optimization Strategies and Estimates.”
Table 1-1: Optimization operator support
operator Description
hash join Determines whether the query optimizer may use the hash join algorithm. hash join
may consume more runtime resources, but is valuable when the joining columns do not
have useful indexes or when a relatively large number of rows satisfy the join
condition, compared to the product of the number of rows in the joined tables.
hash union distinct Determines whether the query optimizer may use the hash union distinct algorithm,
which is inefficient if most rows are distinct.
merge join Determines whether the query optimizer may use the merge join algorithm, which
relies on ordered input. merge join is most valuable when input is ordered on the merge
key, for example, from an index scan. merge join is less valuable if sort operators are
required to order input.
merge union all Determines whether the query optimizer may use the merge algorithm for union all.
merge union all maintains the ordering of the result rows from the union input. merge
union all is particularly valuable if the input is ordered and a parent operator (such as
merge join) benefits from that ordering. Otherwise, merge union all may require sort
operators that reduce efficiency.
merge union distinct Determines whether the query optimizer may use the merge algorithm for union.
merge union distinct is similar to merge union all, except that duplicate rows are not
retained. merge union distinct requires ordered input and provides ordered output.
nested-loop-join The nested-loop-join algorithm is the most common type of join method and is most
useful in simple OLTP queries that do not require ordering.
append union all Determines whether the query optimizer may use the append algorithm for union all.
distinct hashing Determines whether the query optimizer may use a hashing algorithm to eliminate
duplicates, which is very efficient when there are few distinct values compared to the
number of rows.
distinct sorted Determines whether the query optimizer may use a single-pass algorithm to eliminate
duplicates. distinct sorted relies on an ordered input stream, and may increase the
number of sort operators if its input is not ordered.
group-sorted Determines whether the query optimizer may use an on-the-fly grouping algorithm.
group-sorted relies on an input stream sorted on the grouping columns, and it preserves
this ordering in its output.
distinct sorting Determines whether the query optimizer may use the sorting algorithm to eliminate
duplicates. distinct sorting is useful when the input is not ordered (for example, if there
is no index) and the output ordering generated by the sorting algorithm could benefit;
for example, in a merge join.

4 Adaptive Server Enterprise

 CHAPTER 1 Understanding Query Processing

operator Description
group hashing Determines whether the query optimizer may use a group hashing algorithm to process
aggregates.

Technique Description
multi table store ind Determines whether the query optimizer may use reformatting on the result of a
multiple table join. Using multi table store ind may increase the use of worktables.
opportunistic distinct view Determines whether the query optimizer may use a more flexible algorithm when
enforcing distinctness.
index intersection Determines whether the query optimizer may use the intersection of multiple index
scans as part of the query plan in the search space.

Factors analyzed in optimizing queries

Query plans consist of retrieval tactics and an ordered set of execution steps,
which retrieve the data needed by the query. In developing query plans, the
query optimizer examines:
• The size of each table in the query, both in rows and data pages, and the
number of OAM and allocation pages to be read.
• The indexes that exist on the tables and columns used in the query, the type
of index, and the height, number of leaf pages, and cluster ratios for each
index.
• The index coverage of the query; that is, whether the query can be satisfied
by retrieving data from the index leaf pages without accessing the data
pages. Adaptive Server can use indexes that cover queries, even if no
where clauses are included in the query.

• The density and distribution of keys in the indexes.

• The size of the available data cache or caches, the size of I/O supported by
the caches, and the cache strategy to be used.
• The cost of physical and logical reads; that is, reads of physical I/O pages
from the disk, and of logical I/O reads from main memory.
• join clauses, with the best join order and join type, considering the costs and
number of scans required for each join and the usefulness of indexes in
limiting the I/O.

Performance and Tuning Series: Query Processing and Abstract Plans 5

Query optimizer

• Whether building a worktable (an internal, temporary table) with an index

on the join columns is faster than repeated table scans if there are no useful
indexes for the inner table in a join.
• Whether the query contains a max or min aggregate that can use an index
to find the value without scanning the table.
• Whether data or index pages must be used repeatedly, to satisfy a query
such as a join, or whether a fetch-and-discard strategy can be employed
because the pages need to be scanned only once.
For each plan, the query optimizer determines the total cost by computing the
costs of logical and physical I/Os, and CPU processing. If there are proxy
tables, additional network related costs are evaluated as well. The query
optimizer then selects the cheapest plan.
The query processor for Adaptive Server versions 15.0.2 and later defers the
optimization of statements in a stored procedure until it executes the statement.
This benefits the query processor because the values for local variables are
available for optimization for their respective statements.
Earlier versions of Adaptive Server used default guesses for selectivity
estimates on predicates using local variables.

Transformations for query optimization

After a query is parsed and preprocessed, but before the query optimizer begins
its plan analysis, the query is transformed to increase the number of clauses that
can be optimized. The transformation changes made by the optimizer are
transparent unless the output of such query tuning tools as showplan, dbcc(200),
statistics io, or the set commands is examined. If you run queries that benefit
from the addition of optimized search arguments, the added clauses are visible.
In showplan output, it appears as “Keys are” messages for tables for which you
specify no search argument or join.

Search arguments converted to equivalent arguments

The optimizer looks for query clauses to convert to the form used for search
arguments. These are listed in Table 1-2.

6 Adaptive Server Enterprise

 CHAPTER 1 Understanding Query Processing

Table 1-2: Search argument equivalents

Clause Conversion
between Converted to >= and <= clauses. For example, between 10 and 20
is converted to >= 10 and <= 20.
like If the first character in the pattern is a constant, like clauses may
be converted to greater than or less than queries. For example, like
"sm%" becomes >= "sm" and < "sn".
If the first character is a wildcard, a clause such as like "%x" cannot
use an index for access, but histogram values can be used to
estimate the number of matching rows.
in(values_list) Converted to a list of or queries, that is, int_col in (1, 2, 3) becomes
int_col = 1 or int_col = 2 or int_col = 3. The maximum number of
elements in an in-list is 1025

Search argument transitive closure applied where applicable

The optimizer applies transitive closure to search arguments. For example, the
following query joins titles and titleauthor on title_id and includes a search
argument on titles.title_id:
select au_lname, title
from titles t, titleauthor ta, authors a
where t.title_id = ta.title_id
and a.au_id = ta.au_id
and t.title_id = “T81002"
This query is optimized as if it also included the search argument on
titleauthor.title_id:

select au_lname, title

from titles t, titleauthor ta, authors a
where t.title_id = ta.title_id
and a.au_id = ta.au_id
and t.title_id = “T81002”
and ta.title_id = “T81002"
With this additional clause, the query optimizer can use index statistics on
titles.title_id to estimate the number of matching rows in the titleauthor table.
The more accurate cost estimates improve index and join order selection.

equi-join predicate transitive closure applied where applicable

The optimizer applies transitive closure to join columns for a normal equi-join.
The following query specifies the equi-join of t1.c11 and t2.c21, and the equi-join
of t2.c21 and t3.c31:

Performance and Tuning Series: Query Processing and Abstract Plans 7

Query optimizer

select *
from t1, t2, t3
where t1.c11 = t2.c21
and t2.c21 = t3.c31
and t3.c31 = 1
Without join transitive closure, the only join orders considered are (t1, t2, t3),
(t2, t1, t3), (t2, t3, t1),and (t3, t2, t1). By adding the join on t1.c11 = t3.31, the
query processor expands the list of join orders to include: (t1, t3, t2) and (t3, t1,
t2). Search argument transitive closure applies the condition specified by t3.c31
= 1 to the join columns of t1 and t2.

Similarly, equi-join transitive closure is also applied to equi-joins with or

predicates as follows:
select *
from R,S
where R.a = S.a
and (R.a = 5 OR S.b = 6)
The query optimizer infers that this would be equivalent to:
select *
from R,S
where R.a = S.a
and (S.a = 5 or S.b = 6)
The or predicate could be evaluated on the scan of S and possibly be used for
an or optimization, thereby effectively using the indexes of S.
Another example of join transitive closure is its application to compex SARGs,
so that a query such as:
select *
from R,S
where R.a = S.a and (R.a + S.b = 6)
is transformed and inferred as:
select *
from R,S
where R.a = S.a
and (S.a + S.b = 6)
The complex predicate could be evaluated on the scan of S, resulting in
significant performance improvements due to early result set filtering.
Transitive closure is used only for normal equi-joins, as shown. join transitive
closure is not performed for:
• Non-equi-joins; for example, t1.c1 > t2.c2

8 Adaptive Server Enterprise

 CHAPTER 1 Understanding Query Processing

• Outer joins; for example, t1.c11 *= t2.c2, or left join or right join
• Joins across subquery boundaries
• Joins used to check referential integrity or the with check option on views

Note As of Adaptive Server Enterprise 15.0, the sp_configure option to turn on

or off join transitive closure and sort merge join is discontinued. Whenever
applicable, join transitive closure is always applied in Adaptive Server
Enterprise 15.0 and later.

Predicate transformation and factoring to provide additional optimization paths

Predicate transformation and factoring increases the number of choices
available to the query processor. It adds optimizable clauses to a query by
extracting clauses from blocks of predicates linked with or into clauses linked
by and. The additional optimized clauses mean there are more access paths
available for query execution. The original or predicates are retained to ensure
query correctness.
During predicate transformation:
1 Simple predicates (joins, search arguments, and in lists) that are an exact
match in each or clause are extracted. In the query in step 3, below, this
clause matches exactly in each block, so it is extracted:
t.pub_id = p.pub_id
between clauses are converted to greater-than-or-equal and less-than-or-
equal clauses before predicate transformation. The sample query uses
between 15 in both query blocks (though the end ranges are different). The
equivalent clause is extracted by step 1:
price >=15
2 Search arguments on the same table are extracted; all terms that reference
the same table are treated as a single predicate during expansion. Both type
and price are columns in the titles table, so the extracted clauses are:
(type = "travel" and price >=15 and price <= 30)
or
(type = "business" and price >= 15 and price <= 50)
3 in lists and or clauses are extracted. If there are multiple in lists for a table
within a blocks, only the first is extracted. The extracted lists for the
sample query are:

Performance and Tuning Series: Query Processing and Abstract Plans 9

Query optimizer

p.pub_id in (“P220”, “P583”, “P780”)

or
p.pub_id in (“P651", “P066", “P629”)
Since these steps can overlap and extract the same clause, duplicates are
eliminated.
Each generated term is examined to determine whether it can be used as
an optimized search argument or a join clause. Only those terms that are
useful in query optimization are retained.
The additional clauses are added to the query clauses specified by the user.
For example, all clauses optimized in this query are enclosed in the or
clauses:
select p.pub_id, price
from publishers p, titles t
where (
t.pub_id = p.pub_id
and type = “travel"
and price between 15 and 30
and p.pub_id in (“P220", “P583", “P780")
)
or (
t.pub_id = p.pub_id
and type = “business"
and price between 15 and 50
and p.pub_id in (“P651", “P066", “P629")
)
Predicate transformation pulls clauses linked with and from blocks of clauses
linked with or, such as those shown above. It extracts only clauses that occur
in all parenthesized blocks. If the example above had a clause in one of the
blocks linked with or that did not appear in the other clause, that clause would
not be extracted.

Handling search arguments and useful indexes

It is important that you distinguish between where and having clause predicates
that can be used to optimize the query and those that are used later during query
processing to filter the returned rows.

10 Adaptive Server Enterprise

 CHAPTER 1 Understanding Query Processing

You can use search arguments to determine the access path to the data rows
when a column in the where clause matches an index key. The index can be
used to locate and retrieve the matching data rows. Once the row has been
located in the data cache or has been read into the data cache from disk, any
remaining clauses are applied.
For example, if the authors table has on an index on au_lname and another on
city, either index can be used to locate the matching rows for this query:
select au_lname, city, state
from authors
where city = “Washington"
and au_lname = “Catmull"
The query optimizer uses statistics, including histograms, the number of rows
in the table, the index heights, and the cluster ratios for the index and data pages
to determine which index provides the cheapest access. The index that provides
the cheapest access to the data pages is chosen and used to execute the query,
and the other clause is applied to the data rows once they have been accessed.

Nonequality operators
The non-equality operators, < > and !=, are special cases. The query optimizer
checks whether it should cover nonclustered indexes if the column is indexed,
and uses a nonmatching index scan if an index covers the query. However, if
the index does not cover the query, the table is accessed through a row ID
lookup of the data pages during the index scan.

Examples of search argument optimization

Shown below are examples of clauses that can be fully optimized. If there are
statistics on these columns, they can be used to help estimate the number of
rows the query will return. If there are indexes on the columns, the indexes can
be used to access the data.
au_lname = “Bennett"
price >= $12.00
advance > $10000 and advance < $20000
au_lname like "Ben%" and price > $12.00
These search arguments cannot be optimized unless a functional index is built
on them:
advance * 2 = 5000 /*expression on column side
not permitted */
substring(au_lname,1,3) = "Ben" /* function on

Performance and Tuning Series: Query Processing and Abstract Plans 11

Query optimizer

column name */
These two clauses can be optimized if written in this form:
advance = 5000/2
au_lname like "Ben%"
Consider this query, with the only index on au_lname:
select au_lname, au_fname, phone
from authors
where au_lname = “Gerland”
and city = "San Francisco"
The clause qualifies as a search argument:
au_lname = “Gerland"
• There is an index on au_lname
• There are no functions or other operations on the column name.
• The operator is a valid search argument operator.
This clause matches all the criteria above except the first; there is no index on
the city column. In this case, the index on au_lname is used for the query. All
data pages with a matching last name are brought into cache, and each
matching row is examined to see if the city matches the search criteria.

Handling joins
The query optimizer processes join predicates the same way it processes search
arguments, in that it uses statistics, number of rows in the table, index heights,
and the cluster ratios for the index and data pages to determine which index and
join method provides the cheapest access. In addition, the query optimizer also
uses join density estimates derived from join histograms that give accurate
estimates of qualifying joining rows and the rows to be scanned in the outer and
inner tables. The query optimizer also must decide on the optimal join ordering
that will yield the most efficient query plan. The next sections describe the key
techniques used in processing joins.

Join density and join histograms

The query optimizer uses a cost model for joins that uses table-normalized
histograms of the joining attributes. This technique gives an exact value for the
skewed values (that is, frequency count) and uses the range cell densities from
each histogram to estimate the cell counts of corresponding range cells.

12 Adaptive Server Enterprise

 CHAPTER 1 Understanding Query Processing

The join density is dynamically computed from the “join histogram,” which
considers the joining of histograms from both sides of the join operator. The
first histogram join occurs typically between two base tables when both
attributes have histograms. Every histogram join creates a new histogram on
the corresponding attribute of the parent join's projection.
The outcome of the join histogram technique is accurate join selectivity
estimates, even if data distributions of the joining columns are skewed,
resulting in superior join orders and performance.

Expression histogramming selectivity estimates

Versions of Adaptive Server earlier than 15.0.2 used default “guesses” for
selectivity estimates.
Adaptive Server versions 15.0.2 and later apply histogramming estimates to
single column predicates if the histogram exists on the column. This results in
more accurate row estimates, and improves the join order selection for query
plans.
In this example, if the expression is very selective, it may be better to place
table t1 at the beginning of the join order:
select * from t1,t2 where substring(t1.charcol, 1, 3)
= "LMC" and t1.a1 = t2.b

Joins with mixed datatypes

A basic requirement is the ability to build keys for index lookups whenever
possible, without regard to mixed datatypes of any of the join predicates versus
the index key. Consider the following query:
create table T1 (c1 int, c2 int)
create table T2 (c1 int, c2 float)
create index i1 on T1(c2)
create index i1 on T2(c2)

select * from T1, T2 where T1.c2=T2.c2

Assume that T1.c2 is of type int and has an index on it, and that T2.c2 is of type
float with an index.

Performance and Tuning Series: Query Processing and Abstract Plans 13

Query optimizer

As long as datatypes are implicitly convertible, the query optimizer can use
index scans to process the join. In other words, the query optimizer uses the
column value from the outer table to position the index scan on the inner table,
even when the lookup value from the outer table has a different datatype than
the respective index attribute of the inner table.

Joins with expressions and or predicates

See “Predicate transformation and factoring to provide additional
optimization paths” on page 9 for description of how the query optimizer
handles joins with expressions and or predicates.

join ordering
One of the key tasks of the query optimizer is to generate a query plan for join
queries so that the order of the relations in the joins processed during query
execution is optimal. This involves elaborate plan search strategies that can
consume significant time and memory. The query optimizer uses several
effective techniques to obtain the optimal join ordering. The key techniques
are:
• Use of a greedy strategy to obtain an initial good ordering that can be used
as an upper boundary to prune out other, subsequent join orderings. The
greedy strategy employs join row estimates and the nested-loop-join
method to arrive at the initial ordering.
• An exhaustive ordering strategy follows the greedy strategy. In this
strategy, a potentially better join ordering replaces the join ordering
obtained in the greedy strategy. This ordering may employ any join
method.
• Use of extensive cost-based and rule-based pruning techniques eliminates
undesirable join orders from consideration. The key aspect of the pruning
technique is that it always compares partial join orders (the prefix of a
potential join ordering) against the best complete join ordering to decide
whether to proceed with the given prefix. This significantly improves the
time required determine an optimal join order.

14 Adaptive Server Enterprise

 CHAPTER 1 Understanding Query Processing

• The query optimizer can recognize and process star or snowflake schema
joins and process their join ordering in the most efficient way. A typical
star schema join involves a large Fact table that has equi-join predicates that
join it with several Dimension tables. The Dimension tables have no join
predicates connecting each other; that is, there are no joins between the
Dimension tables themselves, but there are join predicates between the
Dimension tables and the Fact table. The query optimizer employs special
join ordering techniques during which the large Fact table is pushed to the
end of the join order and the Dimension tables are pulled up front, yielding
highly efficient query plans. The query optimizer does not, however, use
this technique if the star schema joins contain subqueries, outer joins, or or
predicates.

Optimization goals
Optimization goals are a convenient way to match query demands with the best
optimization techniques, thus ensuring optimal use of the optimizer’s time and
resources. The query optimizer allows you to configure three types of
optimization goals, which you can specify at three tiers: server level, session
level, and query level.
Set the optimization goal at the desired level. The server-level optimization
goal is overridden at the session level, which is overridden at the query level.
These optimization goals allow you to choose an optimization strategy that best
fits your query environment:
• allrows_mix – the default goal, and the most useful goal in a mixed-query
environment. allows_mix balances the needs of OLTP and DSS query
environments.
• allrows_dss – the most useful goal for operational DSS queries of medium
to high complexity. Currently, this goal is provided on an experimental
basis.
• allrows_oltp – the optimizer considers only nested-loop joins.

At the server level, use sp_configure. For example:

sp_configure "optimization goal", 0, "allrows_mix"
At the session level, use set plan optgoal. For example:
set plan optgoal allrows_dss

Performance and Tuning Series: Query Processing and Abstract Plans 15

Optimization goals

At the query level, use a select or other DML command. For example:
select * from A order by A.a plan
"(use optgoal allrows_dss)"
In general, you can set query-level optimization goals using select, update, and
delete statements. However, you cannot set query-level optimization goals in
pure insert statements, although you can set optimization goals in insert…select
statements.

Limiting the time spent optimizing a query

Long-running and complex queries can be time-consuming and costly to
optimize. The timeout mechanism helps to limit that time while supplying a
satisfactory query plan. The query optimizer provides a mechanism by which
the optimizer can limit the time taken by long-running and complex queries;
timing out allows the query processor to stop optimizing when it is reasonable
to do so.
The optimizer triggers timeout during optimization when both these
circumstances are met:
• At least one complete plan has been retained as the best plan, and
• The user configured timeout percentage limit has been exceeded.
You can limit the amount of time Adaptive Server spends optimizing a query
at every level, using the optimization timeout limit parameter, which you can set
to anyvalue between 0 and 1000. optimization timeout limit represents the
percentage of estimated query execution time that Adaptive Server must spend
to optimize the query. For example, specifying a value of 10 tells Adaptive
Server to spend 10% of the estimated query execution time in optimizing the
query. Similarly, a value of 1000 tells Adaptive Server to spend 1000% of the
estimated query execution time, or 10 times the estimated query execution
time, in optimizing the query.
See Chapter 5, “Setting Configuration Parameters,” in the System
Administartion Guide for more information about optimization timeout limit.
A large timeout value may be useful for optimization of stored procedures with
complex queries. Generally, longer optimization time of the stored procedures
yields better plans; the longer optimization time can be amortized over several
executions of the stored procedure.

16 Adaptive Server Enterprise

 CHAPTER 1 Understanding Query Processing

A small timeout value may be used when you want a faster compilation time
from complex ad hoc queries that normally take a long time to compile.
However, for most queries, the default timeout value of 10 should suffice.
Use sp_configure to set the optimization timeout limit configuration parameter
at the server level. For example, to limit optimization time to 10% of total
query processing time, enter:
sp_configure “optimization timeout limit", 10
Use set to set timeout at the session level:
set plan opttimeoutlimit <n>
where n is any integer between 0 and 4000.
Use select to limit optimization time at the query level:
select * from <table> plan "(use opttimeoutlimit <n>)"
where n is any integer between 0 and 1000.

Parallelism
Adaptive Server supports horizontal and vertical parallelism for query
execution. Vertical parallelism is the ability to run multiple operators at the
same time by employing different system resources such as CPUs, disks, and
so on. Horizontal parallelism is the ability to run multiple instances of an
operator on the specified portion of the data.
See Chapter 5, “Parallel Query Processing,” for a more detailed discussion of
parallel query optimization in Adaptive Server.

Optimization issues
Although the query optimizer can efficiently optimize most queries, these
issues may effect the optimizer’s efficiency:
• If statistics have not been updated recently, the actual data distribution
may not match the values used to optimize queries.

Performance and Tuning Series: Query Processing and Abstract Plans 17

Optimization issues

• The rows referenced by a specified transaction may not fit the pattern
reflected by the index statistics.
• An index may access a large portion of the table.
• where clauses (SARGS) are written in a form that cannot be optimized.

• No appropriate index exists for a critical query.

• A stored procedure was compiled before significant changes to the
underlying tables were performed.
• No statistics exists for the SARG or joining columns.
These situations highlight the need to follow some best practices that allow the
query optimizer to perform at its full potential:
Creating search Follow these guidelines when you write search arguments for your queries:
arguments
• Avoid functions, arithmetic operations, and other expressions on the
column side of search clauses. When possible, move functions and other
operations to the expression side of the clause.
• Use as many search arguments as you can, to give the query processor as
much as possible to work with.
• If a query has more than 400 predicates for a table, put the most potentially
useful clauses near the beginning of the query, since only the first 102
SARGs on each table are used during optimization. (All of the search
conditions are used to qualify the rows.)
• Queries using > (greater than) may perform better if you can rewrite them
to use >= (greater than or equal to). For example, this query, with an index
on int_col, uses the index to find the first value where int_col equals 3, and
then scans forward to find the first value that is greater than 3. If there are
many rows where int_col equals 3, the server must scan many pages to find
the first row where int_col is greater than 3:
select * from table1 where int_col > 3
It is more efficient to write the query this way:
select * from table1 where int_col >= 4
This optimization is more difficult with character strings and floating-
point data.
• Check the showplan output to see which keys and indexes are used.

18 Adaptive Server Enterprise

 CHAPTER 1 Understanding Query Processing

• If an index is not being used when you expect it to be, use output from the
set commands in Table 3-1 on page 112 to see whether the query
processor is considering the index.
Use of SQL derived Queries expressed as a single SQL statement exploit the query processor better
tables than queries expressed in two or more SQL statements. SQL-derived tables
enable you to express, in a single step, what might otherwise require several
SQL statements and temporary tables, especially where intermediate aggregate
results must be stored. For example:
select dt_1.* from
(select sum(total_sales)
from titles_west group by total_sales)
dt_1(sales_sum),
(select sum(total_sales)
from titles_east group by total_sales)
dt_2(sales_sum)
where dt_1.sales_sum = dt_2.sales_sum
Here, aggregate results are obtained from the SQL derived tables dt_1 and dt_2,
and a join is computed between the two SQL derived tables. Everything is
accomplished in a single SQL statement.
For more information, see Chapter 9, “SQL Derived Tables,” in the Transact-
SQL User's Guide.
Tuning according to To understand query and system behavior, know the sizes of your tables and
object sizes indexes. At several stages of tuning work, you need size data to:
• Understand statistics i/o reports for a specific query plan.
• Understand the query processor's choice of query plan. The Adaptive
Server cost-based query processor estimates the physical and logical I/O
required for each possible access method and selects the cheapest method.
• Determine object placement, based on the sizes of database objects and on
the expected I/O patterns on the objects.
To improve performance, distribute database objects across physical
devices, so that reads and writes to disk are evenly distributed.
Object placement is described in the Performance and Tuning Series:
Physical Database Tuning.
• Understand changes in performance. If objects grow, their performance
characteristics can change. For example, consider a table that is heavily
used and is usually 100 percent cached. If the table grows too large for its
cache, queries that access the table can suffer poor performance. This is
particularly true of joins that require multiple scans.

Performance and Tuning Series: Query Processing and Abstract Plans 19

Lava query execution engine

• Do capacity planning. Whether you are designing a new system or

planning for the growth of an existing system, you must know the space
requirements to plan for physical disks and memory needs.
• Understand output from sp_sysmon reports on physical I/O.
See the System Administration Guide: Volume 2 for more information on
sizing.

Lava query execution engine

In Adaptive Server, all query plans are submitted to the procedural execution
engine for execution. The Procedural Execution Engine drives execution of the
query plan by:
• Executing simple SQL statements such as set, while, and goto directly.
• Calling out to the utility modules of the query plan to execute create table
and create index and other utility commands.
• Setting up the context for and driving the execution of stored procedures
and triggers.
• Setting up the execution context and calling the query execution engine to
execute query plans for select, insert, delete, and update statements.
• Setting up the cursor execution context for cursor open, fetch, and close
statements, and calling the query execution engine to execute these
statements.
• Doing transaction processing and post execution cleanup.
To support the demands of today’s applications, the query execution engine for
Adaptive Server 15.0 and later has been completely rewritten. With a new
query execution engine and query optimizer in place, the procedural execution
engine in Adaptive Server 15.0 and later passes all query plans generated by
the new query optimizer to the Lava query execution engine.

20 Adaptive Server Enterprise

 CHAPTER 1 Understanding Query Processing

The Lava query execution engine executes Lava query plans. All query plans
chosen by the optimizer are compiled into Lava query plans. However, SQL
statements that are not optimized, such as set or create, are compiled into query
plans like those in versions of Adaptive Server earlier than 15.0, and are not
executed by the Lava query execution engine. Non-Lava query plans are either
executed by the procedural execution engine or by utility modules called by the
procedural engine. Adaptive Server version 15.0 and later has two distinct
kinds of query plans and this is clearly seen in the showplan output (see
Chapter 3, “Displaying Query Optimization Strategies and Estimates”).

Lava query plans

A Lava query plan is built as an upside down tree of Lava operators: The top
Lava operator can have one or more child operators, which in turn can have one
or more child operators, and so on, thus building a bottom-up tree of operators.
The exact shape of the tree and the operators in it are chosen by the optimizer.
An example of a Lava query plan for this query is shown in Figure 1-2:
Select o.id from sysobjects o, syscolumns c
where o.id <= 1 and o.id < 2

Performance and Tuning Series: Query Processing and Abstract Plans 21

Lava query execution engine

Figure 1-2: Lava query plan

Emit

NestedLoopJoin

IndexScan IndexScan
sysobjects(o) syscolumns(c)

The Lava query plan for this query consists of four Lava operators. The top
operator is an Emit (also called Root) operator that dispatches the results of
query execution either by sending the rows to the client or by assigning values
to local variables.
The only child operator of the Emit is a NestedLoopJoin (NL Join)that uses the
nested loop join algorithm to join the rows coming from its two child operators,
(1) the Scan of sysobjects and (2) the scan of syscolumns.
Since the optimizer optimizes all select, insert, delete and update statements,
these are always compiled into Lava query plans and executed by the Lava
query engine.
Some SQL statements are compiled into hybrid query plans. Such plans have
multiple steps, some of which are executed by the Utility modules, and a final
step that is a Lava query plan. An example is the select into statement; select
into is compiled into a two-step query plan:

• create table creates the target table of the statement.

• A Lava query plan to inserts the rows into the target table. To execute this
query plan, the Procedural Execution Engine calls the create table utility
to execute the first step to create the table.
Then the procedural engine calls the Lava query execution engine to execute
the Lava query plan to select and insert the rows into the target table.

22 Adaptive Server Enterprise

 CHAPTER 1 Understanding Query Processing

The two other SQL statements that generate hybrid query plans are alter table
(but only when data copying is required) and reorg rebuild.
A Lava query plan is also generated and executed to support bcp. The support
for bcp in Adaptive Server has always been in the bcp utility. Now, in version
15.0 and later, the bcp utility generates a Lava query plan and calls the Lava
query execution engine to execute the plan.
See Chapter 2, “Using showplan” for more examples of Lava query plans.

Lava operators
Lava query plans are built of Lava operators. Each Lava operator is a self-
contained software object that implements one of the basic physical operations
that the optimizer uses to build query plans. Each Lava operator has five
methods that can be called by its parent operator. These five methods
correspond to the five phases of query execution:
• Acquire

• Open

• Next

• Close,

• Release

Because the Lava operators all provide the same methods (that is, the same
API), they can be interchanged like building blocks in a Lava query plan. For
example, you can replace the NLJoin operator in Figure 1-2 with a MergeJoin
operator or a HashJoin operator without impacting any of the other three
operators in the query plan.
The Lava operators that can be chosen by the optimizer to build Lava query
plans are listed in Table 1-3:
Table 1-3: Lava operators
Operator Description
BulkOp Executes the part of bcp processing that is done in the Lava query engine. Only found
in query plans that are created by the bcp utility, not those created by the optimizer.
CacheScanOp Reads rows from an in-memory table.
DelTextOp Deletes text page chains as part of alter table drop column processing.
DeleteOp Deletes rows from a local table.
Deletes rows from a proxy table when the entire SQL statement cannot be shipped to
the remote server. See also RemoteScanOp.

Performance and Tuning Series: Query Processing and Abstract Plans 23

Lava query execution engine

Operator Description
EmitOp (RootOp) Routes query execution result rows. Can send results to the client or assign result values
to local variables or fetch into variables. An EmitOp is always the top operator in a Lava
query plan.
EmitExchangeOp Routes result rows from a subplan that is executed in parallel to the ExchangeOp in the
parent plan fragment. EmitExchangeOp always appears directly under an ExchangeOp.
See Chapter 5, “Parallel Query Processing.”
GroupSortedOp Performs vector aggregation (group by) when the input rows are already sorted on the
(Aggregation) group-by columns. See also HashVectorAggOp.
GroupSorted (Distinct) Eliminates duplicate rows. Requires the input rows to be sorted on all columns. See also
HashDistinctOp and SortOp (Distinct).
HashVectorAggOp Performs vector aggregation (group by). Uses a Hash algorithm to group the input rows,
so no requirements on ordering of the input rows. See also GroupSortedOp
(Aggregation).
HashDistinctOp Eliminates duplicate rows using a hashing algorithm to find duplicate rows. See also
GroupSortedOp (Distinct) and SortOp (Distinct).
HashJoinOp Performs a join of two input row streams using the HashJoin algorithm.
HashUnionOp Performs a union operation of two or more input row streams using a hashing algorithm
to find and eliminate duplicate rows. See also MergeUnionOp and UnionAllOp.
InsScrollOp Implements extra processing needed to support insensitive scrollable cursors. See also
SemiInsScrollOp.
InsertOp Inserts rows to a local table.
Inserts rows to a proxy table when the entire SQL statement cannot be shipped to the
remote server. See also RemoteScanOp.
MergeJoinOp Performs a join of two streams of rows that are sorted on the joining columns using the
mergejoin algorithm.
MergeUnionOp Performs a union or union all operation on two or more sorted input streams. Guarantees
that the output stream retains the ordering of the input streams. See also HashUnionOp
and UnionAllOp.
NestedLoopJoinOp Performs a join of two input streams using the NestedLoopJoin algorithm.
NaryNestedLoopJoinOp Performs a join of three or more input streams using an enhanced NestedLoopJoin
algorithm. This operator replaces a left-deep tree of NestedLoopJoin operators and can
lead to significant performance improvements when rows of some of the input streams
can be skipped.
OrScanOp Inserts the in or or values into an in-memory table, sorts the values, and removes the
duplicates.Then returns the values, one at a time. Only used for SQL statements with in
clauses or multiple or clauses on the same column.
PtnScanOp Reads rows from a local table (partitioned or not) using either a table scan or an index
scan to access the rows.
RIDJoinOp Receives one or more row identifiers (RIDs) from its left-child operator and calls on its
right-child operator (PtnScanOp) to find the corresponding rows. Used only on SQL
statements with or clauses on different columns of the same table.

24 Adaptive Server Enterprise

 CHAPTER 1 Understanding Query Processing

Operator Description
RIFilterOp (Direct) Drives the execution of a subplan to enforce referential integrity constraints that can be
checked on a row-by-row basis.
Appears only in insert, delete, or update queries on tables with referential integrity
constraints.
RIFilterOp (Deferred) Drives the execution of a sub-plan to enforce referential integrity constraints that can
only be checked after all rows that are affected by the query have been processed.
RemoteScanOp Accesses proxy tables. The RemoteScanOp can:
• Read rows from a single proxy table for further processing in a Lava query plan on
the local host.
• Pass complete SQL statements to a remote host for execution: insert, delete, update,
and select statements. In this case, the Lava query plan consists of an EmitOp with a
RemoteScanOp as its only child operator.
• Pass an arbitrarily complex query plan fragment to a remote host for execution and
read in the result rows (function shipping).
RestrictOp Evaluates expressions.
SQFilterOp Drives the execution of a subplan to execute one or more subqueries.
ScalarAggOp Performs scalar aggregation, such as aggregates without group by.
SemiInsScrollOp Performs extra processing to support semi-insensitive scrollable cursors. See also
InsScrollOp.
SequencerOp Enforces sequential execution of different sub-plans in the query plan.
SortOp Sorts its input rows based upon specified keys.
SortOp (Distinct) Sorts its input and removes duplicate rows. See also HashDisitnctOp and
GroupSortedOp (Distinct).
StoreOp Creates and coordinates the filling of a worktable, and creates a clustered index on the
worktable if required. StoreOp can only have an InsertOp as a child; the InsertOp
populates the worktable.
UnionAllOp Performs a union all operation on two or more input streams. See also HashUnionOp and
MergeUnionOp.
UpdateOp Changes the value of columns in rows of a local table or of a proxy table when the entire
update statement cannot be sent to the remote server. See also RemoteScanOp.
ExchangeOp Enables and coordinates parallel execution of Lava query plans. The ExchangeOp can
be inserted between almost any two Lava operators in a query plan to divide the plan
into sub-plans that can be executed in parallel. See Chapter 5, “Parallel Query
Processing.”

Lava query execution

Execution of a Lava query plan involves five phases:

Performance and Tuning Series: Query Processing and Abstract Plans 25

Lava query execution engine

1 Acquire – acquires resources needed for execution, such as memory

buffers and creating worktables.
2 Open – prepares to return result rows.
3 Next – generates the next result row.
4 Close – cleans up; for example, notifies the access layer that scanning is
complete or truncates worktables
5 Release – releases resources acquired during Acquire, such as memory
buffers, drops worktables.
Each Lava operator has a method with the same name as the phase, which is
invoked for each of these phases.
Figure 1-2 on page 22 demonstrates query plan execution:
• Acquire phase
The acquire method of the Emit operator is invoked. The Emit operator
calls Acquire of its child, the NLJoin operator, which in turn calls Acquire
on its left-child operator (the index scan of sysobjects) and then on its right
child operator (the index scan of syscolumns).
• Open phase
The Open method of the Emit operator is invoked. The Emit operator calls
Open on the NLJoin operator, which calls Open only on its left-child
operator.
• Next phase
The Next method of the Emit operator is invoked. Emit calls Next on the
NLJoin operator, which calls Next on its left child, the Index Scan of
sysobjects. The index scan operator reads the first row from sysobjects and
returns it to the NLJoin operator. The NLJoin operator then calls the Open
method of its right child operator, the Index Scan of syscolumns. Then the
NLJoin operator calls the Next method of the Index Scan of syscolumns to
get a row that matches the joining key of the row from sysobjects. When a
matching row has been found, it is returned to the Emit operator, which
sends it back to the client. Repeated invocations of the Next method of the
Emit operator generate more result rows.

• Close phase
After all rows have been returned, the Close method of the Emit operator
is invoked, which in turn calls Close of the NLJoin operator, which in turn
calls Close on both of its child operators.

26 Adaptive Server Enterprise

 CHAPTER 1 Understanding Query Processing

• Release phase
The Release method of the Emit operator is invoked and the calls to the
Release method of the other operators is propagated down the query plan.
After successfully completing the Release phase of execution, the Lava query
engine returns control to the Procedural Execution Engine for final statement
processing.

How update operations are performed

Adaptive Server handles updates in different ways, depending on the changes
being made to the data and the indexes used to locate the rows. The two major
types of updates are deferred updates and direct updates. Adaptive Server
performs direct updates whenever possible.

Direct updates
Adaptive Server performs direct updates in a single pass:
• It locates the affected index and data rows.
• It writes the log records for the changes to the transaction log.
• It makes the changes to the data pages and any affected index pages.
There are three techniques for performing direct updates:
• In-place updates
• Cheap direct updates
• Expensive direct updates
Direct updates require less overhead than deferred updates and are generally
faster, as they limit the number of log scans, reduce logging, save traversal of
index B-trees (reducing lock contention), and save I/O because Adaptive
Server does not have to refetch pages to perform modifications based on log
records.

In-place updates
Adaptive Server performs in-place updates whenever possible.

Performance and Tuning Series: Query Processing and Abstract Plans 27

How update operations are performed

When Adaptive Server performs an in-place update, subsequent rows on the

page are not moved; row IDs remain the same and the pointers in the row offset
table are not changed.
For an in-place update, the following requirements must be met:
• The row being changed cannot change its length.
• The column being updated cannot be the key, or part of the key, of a
clustered index on an allpages-locked table. Because the rows in a
clustered index on an allpages-locked table are stored in key order, a
change to the key almost always means that the row location is changed.
• One or more indexes must be unique or must allow duplicates.
• The update statement satisfies the conditions listed in “Restrictions on
update modes through joins” on page 33.
• The affected columns are not used for referential integrity.
• There cannot be a trigger on the column.
• The table cannot be replicated (via Replication Server).
An in-place update is the fastest type of update because it makes a single
change to the data page. It changes all affected index entries by deleting the old
index rows and inserting the new index row. In-place updates affect only
indexes whose keys are changed by the update, since the page and row
locations are not changed.

Cheap direct updates

If Adaptive Server cannot perform an update in place, it tries to perform a
cheap direct update—changing the row and rewriting it at the same offset on
the page. Subsequent rows on the page are moved up or down so that data
remains contiguous on the page, but row IDs remain the same. The pointers in
the row offset table change to reflect the new locations.
A cheap direct update must meet these requirements:
• The length of the data in the row is changed, but the row still fits on the
same data page, or the row length is not changed, but there is a trigger on
the table or the table is replicated.
• The column being updated cannot be the key, or part of the key, of a
clustered index. Because Adaptive Server stores the rows of a clustered
index in key order, a change to the key almost always means that the row
location is changed.

28 Adaptive Server Enterprise

 CHAPTER 1 Understanding Query Processing

• One or more indexes must be unique or must allow duplicates.

• The update statement satisfies the conditions listed in “Restrictions on
update modes through joins” on page 33.
• The affected columns are not used for referential integrity.
Cheap direct updates are almost as fast as in-place updates. They require the
same amount of I/O, but slightly more processing. Two changes are made to
the data page (the row and the offset table). Any changed index keys are
updated by deleting old values and inserting new values. Cheap direct updates
affect only indexes whose keys are changed by the update, since the page and
row ID are not changed.

Expensive direct updates

If the data does not fit on the same page, Adaptive Server performs an
expensive direct update, if possible. An expensive direct update deletes the
data row, including all index entries, and then inserts the modified row and
index entries.
Adaptive Server uses a table scan or an index to find the row in its original
location and then deletes the row. If the table has a clustered index, Adaptive
Server uses the index to determine the new location for the row; otherwise,
Adaptive Server inserts the new row at the end of the heap.
An expensive direct update must meet these requirements:
• The length of a data row is changed so that the row no longer fits on the
same data page, and the row is moved to a different page, or the update
affects key columns for the clustered index.
• The index used to find the row is not changed by the update.
• The update statement satisfies the conditions listed in “Restrictions on
update modes through joins” on page 33.
• The affected columns are not used for referential integrity.
An expensive direct update is the slowest type of direct update. The delete is
performed on one data page, and the insert is performed on a different data
page. All index entries must be updated, since the row location is changed.

Performance and Tuning Series: Query Processing and Abstract Plans 29

How update operations are performed

Deferred updates
Adaptive Server uses deferred updates when direct update conditions are not
met. A deferred update is the slowest type of update.
In a deferred update, Adaptive Server:
• Locates the affected data rows, writing the log records for deferred delete
and insert of the data pages as rows are located.
• Reads the log records for the transaction and performs the deletes on the
data pages and any affected index rows.
• Reads the log records a second time, and performs all inserts on the data
pages, and inserts any affected index rows.

When deferred updates are required

Deferred updates are always required for:
• Updates that use self-joins
• Updates to columns used for self-referential integrity
• Updates to a table referenced in a correlated subquery
Deferred updates are also required when:
• An update moves a row to a new page while the table is being accessed by
a table scan or a clustered index.
• Duplicate rows are not allowed in the table, and there is no unique index
to prevent them.
• The index used to find the data row is not unique, and the row is moved
because the update changes the clustered index key or because the new
row does not fit on the page.
Deferred updates incur more overhead than direct updates because they require
Adaptive Server to reread the transaction log to make the final changes to the
data and indexes. This involves additional traversal of the index trees.
For example, if there is a clustered index on title, this query performs a deferred
update:
update titles set title = "Portable C Software" where
title = "Designing Portable Software"

30 Adaptive Server Enterprise

 CHAPTER 1 Understanding Query Processing

Deferred index inserts

Adaptive Server performs deferred index updates when the update affects the
index used to access the table or when the update affects columns in a unique
index. In this type of update, Adaptive Server:
• Deletes the index entries in direct mode
• Updates the data page in direct mode, writing the deferred insert records
for the index
• Reads the log records for the transaction and inserts the new values in the
index in deferred mode
Deferred index insert mode must be used when the update changes the index
used to find the row or when the update affects a unique index. A query must
update a single, qualifying row only once—deferred index update mode
ensures that a row is found only once during the index scan and that the query
does not prematurely violate a uniqueness constraint.
The update in Figure 1-3 on page 32 changes only the last name, but the index
row is moved from one page to the next. To perform the update, Adaptive
Server:
1 Reads index page 1133, deletes the index row for “Greene” from that page,
and logs a deferred index scan record.
2 Changes “Green” to “Hubbard” on the data page in direct mode and
continues the index scan to see if more rows need to be updated.
3 Inserts the new index row for “Hubbard” on page 1127.
Figure 1-3 shows the index and data pages prior to the deferred update
operation, and the sequence in which the deferred update changes the data and
index pages.

Performance and Tuning Series: Query Processing and Abstract Plans 31

How update operations are performed

Figure 1-3: Deferred index update

update employee
set lname = "Hubbard"
where lname = "Green"

Pointer
Page 1242

Key
Before update: 10 O’Leary
11 Ringer
Page 1132

Pointer
Row ID
12 White
Row ID

Pointer

Bennet 1421,1 13 Jenkins

Key
Key

Chan 1129,3
Dull 1409,1 Page 1307
Page 1001 Page 1007 Edwards 1018,5 14 Hunter
Bennet 1421,1 Bennet 1421,1 15 Smith
1007 1132 16 Ringer
Greane 1307,4 Page 1133
Karsen 1411,3 17 Greane
1133 Greane 1307,4
1009
Green 1421,2
Page 1421
Greene 1409,2
Page 1009 18 Bennet
Karsen 1411,3 19 Green
1315 Page 1127 20
Hunter 1307,1 Yokomoto
Jenkins 1242,4 Page 1409
21 Dull
22 Greene
23 White

Root page Intermediate Leaf pages Data pages

Step 1: Write log

Gree

Update steps Page 1133

records, then Greane 1307,4
n

delete index row. Greene 1409,2

Page 1421
Step 2: Change 18 Bennet
19
data page.
Hubbard
20

Step 3: Read log, Page 1127

insert index row. Hubbard 1421,2
Hunter 1307,1
Jenkins 1242,4

32 Adaptive Server Enterprise

 CHAPTER 1 Understanding Query Processing

Assume a similar update to the titles table:

update titles
set title = "Computer Phobic’s Manual",
advance = advance * 2
where title like "Computer Phob%"
This query shows a potential problem. If a scan of the nonclustered index on
the title column found “Computer Phobia Manual,” changed the title, and
multiplied the advance by 2, and then found the new index row “Computer
Phobic’s Manual” and multiplied the advance by 2, the advance wold be very
skewed against the reality.
A deferred index delete may be faster than an expensive direct update, or it may
be substantially slower, depending on the number of log records that need to be
scanned and whether the log pages are still in cache.
During deferred update of a data row, there can be a significant time interval
between the delete of the index row and the insert of the new index row. During
this interval, there is no index row corresponding to the data row. If a process
scans the index during this interval at isolation level 0, it does not return the old
or new value of the data row.

Restrictions on update modes through joins

Updates and deletes that involve joins can be performed in direct,
deferred_varcol, or deferred_index mode when the table being updated is the
outermost table in the join order, or when it is preceded in the join order by
tables where only a single row qualifies.

Joins and subqueries in update and delete statements

The use of the from clause to perform joins in update and delete statements is a
Transact-SQL extension to ANSI SQL. Subqueries in ANSI SQL form can be
used in place of joins for some updates and deletes.
This example uses the from syntax to perform a join:
update t1 set t1.c1 = t1.c1 + 50
from t1, t2
where t1.c1 = t2.c1
and t2.c2 = 1
The following example shows the equivalent update using a subquery:
update t1 set c1 = c1 + 50

Performance and Tuning Series: Query Processing and Abstract Plans 33

How update operations are performed

where t1.c1 in (select t2.c1

from t2
where t2.c2 = 1)
The update mode that is used for the join query depends on whether the updated
table is the outermost query in the join order—if it is not the outermost table,
the update is performed in deferred mode. The update that uses a subquery is
always performed as a direct, deferred_varcol, or deferred_index update.
For a query that uses the from syntax and performs a deferred update due to the
join order, use showplan and statistics io to determine whether rewriting the
query using a subquery can improve performance. Not all queries using from
can be rewritten to use subqueries.

Deletes and updates in triggers versus referential integrity

Triggers that join user tables with the deleted or inserted tables are run in
deferred mode. If you are using triggers solely to implement referential
integrity, and not to cascade updates and deletes, then using declarative
referential integrity in place of triggers may avoid the penalty of deferred
updates in triggers.

Optimizing updates
showplan messages provide information about whether an update is performed
in direct mode or deferred mode. If a direct update is not possible, Adaptive
Server updates the data row in deferred mode. There are times when the
optimizer cannot know whether a direct update or a deferred update will be
performed, so two showplan messages are provided:
• The “deferred_varcol” message shows that the update may change the
length of the row because a variable-length column is being updated. If the
updated row fits on the page, the update is performed in direct mode; if the
update does not fit on the page, the update is performed in deferred mode.
• The “deferred_index” message indicates that the changes to the data pages
and the deletes to the index pages are performed in direct mode, but the
inserts to the index pages are performed in deferred mode.
These types of direct updates depend on information that is available only at
runtime, since the page actually has to be fetched and examined to determine
whether the row fits on the page.

34 Adaptive Server Enterprise

 CHAPTER 1 Understanding Query Processing

Designing for direct updates

When you design and code your applications, be aware of the differences that
can cause deferred updates. To help avoid deferred updates:
• Create at least one unique index on the table to encourage more direct
updates.
• Whenever possible, use nonkey columns in the where clause when
updating a different key.
• If you do not use null values in your columns, declare them as not null in
your create table statement.

Effects of update types and indexes on update modes

Table 1-4 on page 36 shows how indexes affect the update mode for three
different types of updates. In all cases, duplicate rows are not allowed. For the
indexed cases, the index is on title_id. The three types of updates are:
• Update of a variable-length key column:
update titles set title_id = value
where title_id = "T1234"
• Update of a fixed-length nonkey column:
update titles set pub_date = value
where title_id = "T1234"
• Update of a variable-length nonkey column:
update titles set notes = value
where title_id = "T1234"
Table 1-4 shows how a unique index can promote a more efficient update mode
than a nonunique index on the same key. Pay particular attention to the
differences between direct and deferred in the shaded areas of the table. For
example, with a unique clustered index, all of these updates can be performed
in direct mode, but they must be performed in deferred mode if the index is
nonunique.
For a table with a nonunique clustered index, a unique index on any other
column in the table provides improved update performance. In some cases, you
may want to add an IDENTITY column to a table to include the column as a
key in an index that would otherwise be nonunique.

Performance and Tuning Series: Query Processing and Abstract Plans 35

How update operations are performed

Table 1-4: Effects of indexing on update mode

Update to:
Variable-length Fixed-length Variable-length
Index key column column
No index N/A direct deferred_varcol
Clustered, unique direct direct direct
Clustered, not unique deferred deferred deferred
Clustered, not unique, with a unique index on another deferred direct deferred_varcol
column
Nonclustered, unique deferred_varcol direct direct
Nonclustered, not unique deferred_varcol direct deferred_varcol

If the key for an index is fixed length, the only difference in update modes from
those shown in the table occurs for nonclustered indexes. For a nonclustered,
nonunique index, the update mode is deferred_index for updates to the key. For
a nonclustered, unique index, the update mode is direct for updates to the key.
If the length of varchar or varbinary is close to the maximum length, use char or
binary instead. Each variable-length column adds row overhead and increases
the possibility of deferred updates.
Using max_rows_per_page to reduce the number of rows allowed on a page
increases direct updates, because an update that increases the length of a
variable-length column may still fit on the same page.
For more information on using max_rows_per_page, see the Performance and
Tuning Series: Physical Database Tuning.

Using sp_sysmon while tuning updates

You can use showplan to determine whether an update is deferred or direct, but
showplan does not give you detailed information about the type of deferred or
direct update. Output from the sp_sysmon or Adaptive Server Monitor supplies
detailed statistics about the types of updates performed during a sample
interval.
Run sp_sysmon as you tune updates, and look for reduced numbers of deferred
updates, reduced locking, and reduced I/O.
See the Performance and Tuning Series: Monitoring Adaptive Server with
sp_sysmon.

36 Adaptive Server Enterprise

CH A PTE R 2 Using showplan

This chapter describes the messages printed by the showplan utility, which
displays the query plan in a text-based format for each SQL statement in
a batch or stored procedure.
Topic Page
Displaying a query plan 37
Statement-level output 44
Query plan shape 47
Union operators 90
INSTEAD-OF TRIGGER operators 106

Displaying a query plan

To see query plans, use:
set showplan on
To stop displaying query plans, use:
set showplan off
You can use showplan in conjunction with other set commands.
To display query plans for a stored procedure, but not execute them, use
the set fmtonly command.
See Chapter 12, “Creating and Using Abstract Plans,” for information on
how options interact.

Note Do not use set noexec with stored procedures—compilation and

execution does not occur and you do not receive the necessary output.

Performance and Tuning Series: Query Processing and Abstract Plans 37

Displaying a query plan

Query plans in Adaptive Server Enterprise 15.0 and later

Adaptive Server traditionally classifies Transact-SQL statements into two
groups:
• Optimizable. For example, this query is optimizable because it has many
relations (tables):
select * from t1, t2, t3, t4
where t1.c1 = t2.c1 and . . .
order by t3.c4
The query processor requires the join order, type of join, search arguments,
and ordering to be optimized.
• Nonoptimizable. Utility commands like update statistics and dbcc are not
optimized.
In Adaptive Server version 15.0 and later the optimizer and most of the
execution engine was rewritten. The utility commands currently generate
nearly identical showplan output compared with earlier versions. However,
versions 15.0 and later generate new showplan output for optimizable
statements.
Some of the new features of the query plans that showplan must display
include:
• Plan elements – query plans can be composed from over thirty different
operators.
• Plan shape – query plans are upside-down trees of operators. In general,
more operators in a query plan result in more combinations of possible tree
shapes. Query plans in version 15.0 and later can be more complex than
those found in earlier versions. Nested indentation is provided to assist in
visualizing the tree shape of these query plans.
• Subplans that are executed in parallel.

Why do I get different query plans for the same query?

The query processor may return a different query plan depending on whether
you configure set plan optgoal for allrows_oltp, allrows_mix, or allrows_dss
(unless you force a plan with forceplan):
• allrows_oltp – the query processor uses the nested-loop join operator.

38 Adaptive Server Enterprise

 CHAPTER 2 Using showplan

• allrows_mix – the query processor allows both nested-loop joins and merge
joins. The query processor measures their relative costs to determine
which join it uses.
• allrows_dss – the query processor uses nested-loop, merge-, or hash-joins.
The query processor measures their relative costs to determine which join
it uses.

Using set showplan with noexec

You can use set noexec with set showplan on to view a query plan without
executing the query. For example, this query prints the query plan but also
executes the queries, which might be time consuming:
set showplan on
go
select * from really_big_table
select * from really_really_big_table
go
However, if you include set noexec, you can view the query plan without
running the query.
Stored procedures are compiled when they are first used, or if the resultant
compiled plan is already in use by another session, so set noexec can have
unexpected results, and Sybase® recommends that you use set fmtonly on
instead. If you include a stored procedures inside another stored procedure, the
second stored procedure is not run when you enable set noexec. For example,
if you create two stored procedures:
create procedure sp_B
as
begin
select * from authors
end
and
create procedure sp_A
as
begin
select * from titles
execute sp_B
end

Performance and Tuning Series: Query Processing and Abstract Plans 39

Using set showplan with noexec

Individually, their query plans look like this:

set showplan on
sp_B
QUERY PLAN FOR STATEMENT 1 (at line 1).

STEP 1
The type of query is EXECUTE.
QUERY PLAN FOR STATEMENT 1 (at line 4).

STEP 1

The type of query is SELECT.

1 operator(s) under root
|ROOT:EMIT Operator (VA = 1)
|
| |SCAN Operator (VA = 0)
| | FROM TABLE
| | titles
| | Table Scan.
| | Forward Scan.
| | Positioning at start of table.
| | Using I/O Size 2 Kbytes for data pages.
| | With LRU Buffer Replacement Strategy for data pages.
If you enable set noexec:
set noexec on
go
set showplan on
go
exec proc A
go
Adaptive Server produces no showplan output for procedure B because noexec
is enabled, so Adaptive Server is not actually executing or compiling procedure
B, and does not print any showplan output. If noexec was not enabled, Adaptive
Server would have compiled and printed plans for both A and B stored
procedures.
But if you use set fmtonly on:
use pubs2
go
create procedure sp_B
as
begin
select * from authors

40 Adaptive Server Enterprise

 CHAPTER 2 Using showplan

end
go
create procedure sp_A
as
begin
select * from titles
execute sp_B
end
go
set showplan on
go
set fmtonly on
go

QUERY PLAN FOR STATEMENT 1 (at line 1).

STEP 1
The type of query is SET OPTION ON.

sp_B
go

QUERY PLAN FOR STATEMENT 1 (at line 1).

STEP 1
The type of query is EXECUTE.

QUERY PLAN FOR STATEMENT 1 (at line 4).

STEP 1

The type of query is SELECT.

1 operator(s) under root
|ROOT:EMIT Operator (VA = 1)
|
| |SCAN Operator (VA = 0)
| | FROM TABLE
| | authors
| | Table Scan.
| | Forward Scan.
| | Positioning at start of table.
| | Using I/O Size 2 Kbytes for data pages.
| | With LRU Buffer Replacement Strategy for data pages.

Performance and Tuning Series: Query Processing and Abstract Plans 41

Using set showplan with noexec

au_id au_lname au_fname

phone address
city state country postalcode
---------- -------------------------------------- --------------------
------------ ----------------------------------------
------------------ ----- ---------- ----------

(0 rows affected)
(return status = 0)
sp_A
go

QUERY PLAN FOR STATEMENT 1 (at line 1).

STEP 1
The type of query is EXECUTE.

QUERY PLAN FOR STATEMENT 1 (at line 4).

STEP 1

The type of query is SELECT.

1 operator(s) under root
|ROOT:EMIT Operator (VA = 1)
|
| |SCAN Operator (VA = 0)
| | FROM TABLE
| | titles
| | Table Scan.
| | Forward Scan.
| | Positioning at start of table.
| | Using I/O Size 2 Kbytes for data pages.
| | With LRU Buffer Replacement Strategy for data pages.

QUERY PLAN FOR STATEMENT 2 (at line 5).

STEP 1
The type of query is EXECUTE.

title_id
title

42 Adaptive Server Enterprise

 CHAPTER 2 Using showplan

type pub_id price advance

total_sales
notes

pubdate contract
--------
--------------------------------------------------------------------
----------- -------- ------------------- -----------------------
-----------
---------------------------------------------------------------------
-----------------------------------------------------------------------------
-------------------------------------------------
-------------------- --------

(0 rows affected)

QUERY PLAN FOR STATEMENT 1 (at line 4).

STEP 1

The type of query is SELECT.

1 operator(s) under root
|ROOT:EMIT Operator (VA = 1)
|
| |SCAN Operator (VA = 0)
| | FROM TABLE
| | authors
| | Table Scan.
| | Forward Scan.
| | Positioning at start of table.
| | Using I/O Size 2 Kbytes for data pages.
| | With LRU Buffer Replacement Strategy for data pages.

au_id au_lname au_fname

phone address
city state country postalcode
---------- -------------------------------------- --------------------
------------ ----------------------------------------
------------------ ----- ---------- ----------
Both stored procedures are run and you see the resulting showplan output.

Performance and Tuning Series: Query Processing and Abstract Plans 43

Statement-level output

Statement-level output
The first section of showplan output for each query plan presents statement-
level information, including the statement and line number in the batch or
stored procedure of the query for which the query plan was generated:
QUERY PLAN FOR STATEMENT N (at line N).
This message may be followed by a series of messages that apply to the
statement’s entire query plan. If the query plan was generated using an abstract
plan about how the abstract plan was forced:
• If an explicit abstract plan was given by a plan clause in the SQL
statement, the message is:
Optimized using the Abstract Plan in the PLAN clause.
• If an abstract plan has been internally generated (that is, for alter table and
reorg commands that are executed in parallel), the message is:
Optimized using the forced options (internally
generated Abstract Plan).
• If a new statement is cached, the output includes:
STEP 1

The type of query is EXECUTE.

Executing a newly cached statement.
• If a cached statement is reused, the output includes:
STEP 1

The type of query is EXECUTE.

Executing a previously cached statement.
• If the query recompiles the statement, the output includes:
QUERY PLAN IS RECOMPILED DUE TO SCHEMACT.

THE RECOMPILED QUERY PLAN IS:

. . .

QUERY PLAN FOR STATEMENT 1 (at line 1)

. . .
• If an abstract plan has been retrieved from sysqueryplans because
automatic abstract plan usage is enabled, the message is:

44 Adaptive Server Enterprise

 CHAPTER 2 Using showplan

Optimized using an Abstract Plan (ID : N).

• If the query plan is a parallel query plan, the following message shows the
number of processes (coordinator plus worker) that are required to execute
the query plan:
Executed in parallel by coordinating process and N
worker processes.
• If the query plan was optimized using simulated statistics, this message
appears next:
Optimized using simulated statistics.
• The output includes VA= which indicates the virtual address for the
operator, and the order in which each operator is executed. The query
processor starts at VA=0. Generally, scan nodes (leaf nodes) are executed
first.

Note The VA= in the showplan output is available for Adaptive Server
version 15.0.2 ESD #2 and later. You will not see VA= in earlier versions
of Adaptive Server.

• Adaptive Server uses a scan descriptor for each database object that is
accessed during query execution. By default, each connection (or each
worker process for parallel query plans) has 28 scan descriptors. If the
query plan requires access to more than 28 database objects, auxiliary scan
descriptors are allocated from a global pool. If the query plan uses
auxiliary scan descriptors, this message is printed, showing the total
number required:
Auxiliary scan descriptors required: N
• This message shows the total number of operators appearing in the query
plan:
N operator(s) under root
• The next message shows the type of query for the query plan. For query
plans, the query type is select, insert, delete, or update:
The type of query is SELECT.
• A final statement-level message is printed at the end of showplan output if
Adaptive Server is configured to enable resource limits. The message
displays the optimizer’s total estimated cost of logical and physical I/O:
Total estimated I/O cost for statement N (at line M):
X.

Performance and Tuning Series: Query Processing and Abstract Plans 45

Statement-level output

The following query, with showplan output, shows some of these messages:
use pubs2
go
set showplan on
go
select stores.stor_name, sales.ord_num
from stores, sales, salesdetail
where salesdetail.stor_id = sales.stor_id
and stores.stor_id = sales.stor_id
plan " ( m_join ( i_scan salesdetailind salesdetail)
( m_join ( i_scan salesind sales ) ( sort ( t_scan stores ) ) ) )"

QUERY PLAN FOR STATEMENT 1 (at line 1).

Optimized using the Abstract Plan in the PLAN clause.

STEP 1
The type of query is SELECT.

6 operator(s) under root

ROOT:EMIT Operator (VA = 6)

|MERGE JOIN Operator (Join Type: Inner Join) (VA = 5)

| Using Worktable3 for internal storage.
| Key Count: 1
| Key Ordering: ASC
|
| |SCAN Operator (VA = 0)
| | FROM TABLE
| | salesdetail
| | Index : salesdetailind
| | Forward Scan.
| | Positioning at index start.
| | Index contains all needed columns. Base table will not be read.
| | Using I/O Size 2 Kbytes for index leaf pages.
| | With LRU Buffer Replacement Strategy for index leaf pages.
|
| |MERGE JOIN Operator (Join Type: Inner Join) (VA = 4)
| | Using Worktable2 for internal storage.
| | Key Count: 1
| | Key Ordering: ASC
| |
| | |SCAN Operator (VA = 1)
| | | FROM TABLE
| | | sales
| | | Table Scan.

46 Adaptive Server Enterprise

 CHAPTER 2 Using showplan

| | | Forward Scan.
| | | Positioning at start of table.
| | | Using I/O Size 2 Kbytes for data pages.
| | | With LRU Buffer Replacement Strategy for data pages.
| |
| | |SORT Operator (VA = 3)
| | | Using Worktable1 for internal storage.
| | |
| | | |SCAN Operator (VA = 2)
| | | | FROM TABLE
| | | | stores
| | | | Table Scan.
| | | | Forward Scan.
| | | | Positioning at start of table.
| | | | Using I/O Size 2 Kbytes for data pages.
| | | | With LRU Buffer Replacement Strategy for data pages.
After the statement level output, the query plan appears. The showplan output
of the query plan consists of two components:
• The names of the operators (some provide additional information) to show
which operations are being executed in the query plan.
• Vertical bars (the “|” symbol) with indentation to show the shape of the
query plan operator tree.

Query plan shape

The position of each operator in the tree determines its order of execution.
Execution starts down the left-most branch of the tree and proceeds to the right.
To illustrate execution, this section steps through the execution of the query
plan for the example in the previous section. Figure 2-1 shows a graphical
representation of the query plan.

Performance and Tuning Series: Query Processing and Abstract Plans 47

Query plan shape

Figure 2-1: Query plan

EMIT
VA=

MERGE JOIN (1)

Inner join
VA=

SCAN MERGE JOIN (2)

salesdetailind Inner join
VA= VA=

SCAN SORT
sales VA=
VA=

SCAN
stores
VA=

To generate a result row, the EMIT operator calls for a row from its child, the
MERGE JOIN operator (1), which calls for a row from its left child, the SCAN
operator for salesdetailind. When EMIT receives a row from its left child,
MERGE JOIN operator (1) calls for a row from its right child, MERGE JOIN
operator (2). MERGE JOIN operator (2) calls for a row from its left child, the
SCAN operator for sales.
When it receives a row from its left child, MERGE JOIN operator (2) calls for
a row from its right child, the SCAN operator. The SCAN operator is a data-
blocking operator. That is, it needs all of its input rows before it can sort them,
so the SORT operator keeps calling for rows from its child, the SCAN operator
for stores, until all rows have been returned. Then the SORT operator sorts
the rows and passes the first row to the MERGE JOIN operator (2).

48 Adaptive Server Enterprise

 CHAPTER 2 Using showplan

The MERGE JOIN operator (2) keeps calling for rows from either the left or
right child operators until it gets two rows that match on the joining keys. The
matching row is then passed up to MERGE JOIN operator (1). MERGE JOIN
operator (1) also calls for rows from its child operators until a match is found,
which is then passed up to the EMIT operator to be returned to the client. In
effect, the operators are processed using a left-deep postfix recursive strategy.
Figure 2-2 shows a graphical representation of an alternate query plan for the
same example query. This query plan contains all of the same operators, but the
shape of the tree is different.
Figure 2-2: Alternate query plan
EMIT
VA=

MergeJoinOp(1)
Inner join
VA=

MergeJoinOp(2) ScanOp
Inner join salesdetailind
VA= VA=

ScanOp SortOp
sales VA=
VA=

ScanOp
stores
VA=

The showplan output corresponding to the query plan in Figure 2-2 is:
QUERY PLAN FOR STATEMENT 1 (at line 1).

6 operator(s) under root

The type of query is SELECT.

Performance and Tuning Series: Query Processing and Abstract Plans 49

Query plan shape

ROOT:EMIT Operator

|MERGE JOIN Operator (Join Type: Inner Join)

| Using Worktable3 for internal storage.
| Key Count: 1
| Key Ordering: ASC
|
| |MERGE JOIN Operator (Join Type: Inner Join)
| | Using Worktable2 for internal storage.
| | Key Count: 1
| | Key Ordering: ASC
| |
| | |SCAN Operator
| | | FROM TABLE
| | | sales
| | | Table Scan.
| | | Forward Scan.
| | | Positioning at start of table.
| | | Using I/O Size 2 Kbytes for data pages.
| | | With LRU Buffer Replacement Strategy for data pages.
| |
| | |SORT Operator
| | | Using Worktable1 for internal storage.
| | |
| | | |SCAN Operator
| | | | FROM TABLE
| | | | stores
| | | | Table Scan.
| | | | Forward Scan.
| | | | Positioning at start of table.
| | | | Using I/O Size 2 Kbytes for data pages.
| | | | With LRU Buffer Replacement Strategy for data pages.
|
| |SCAN Operator
| | FROM TABLE
| | salesdetail
| | Index : salesdetailind
| | Forward Scan.
| | Positioning at index start.
| | Index contains all needed columns. Base table will not be read.
| | Using I/O Size 2 Kbytes for index leaf pages.
| | With LRU Buffer Replacement Strategy for index leaf pages.

50 Adaptive Server Enterprise

 CHAPTER 2 Using showplan

The showplan output conveys the shape of the query plan by using indentation
and the pipe (“|”) symbol to indicate which operators are under which and
which ones are on the same or different branches of the tree. There are two
rules to interpreting the tree shape:
• The pipe “|” symbols form a vertical line that starts at the operator’s name
and continue down past all of the operators that are under it on the same
branch.
• Child operators are indented to the left for each level of nesting.
Using these rules, the shape of the query plan in Figure 2-2 can be derived from
the previous showplan output with the following steps:
1 The ROOT or EMIT operator is at the top of the query plan tree.
2 MERGE JOIN operator (1) is the left child of the ROOT. The vertical line
that starts at MERGE JOIN operator (1) travels down the length of the entire
output, so all of the other operators are below MERGE JOIN operator (1)
and on the same branch.
3 The left child operator of the MERGE JOIN operator (1) is MERGE JOIN
operator (2).
4 The vertical line that starts at MERGE JOIN operator (2) travels down past
a SCAN, a SORT, and another SCAN operator before it ends. These operators
are all nested as a subbranch under MERGE JOIN operator (2).
5 The first SCAN under MERGE JOIN operator (2) is its left child, the SCAN
of the sales table.
6 The SORT operator is the right child of MERGE JOIN operator (2) and the
SCAN of the stores table is the only child of the SORT operator.
7 Below the output for the SCAN of the stores table, several vertical lines
end. This indicates that a branch of the tree has ended.
8 The next output is for the SCAN of the salesdetail table. It has the same
indentation as MERGE JOIN operator (2), indicating that it is on the same
level. In fact, this SCAN is the right child of MERGE JOIN operator (1).

Note Most operators are either unary or binary. That is, they have either a
single child operator or two child operators directly beneath. Operators that
have more than two child operators are called “nary”. Operators that have no
children are leaf operators in the tree and are termed “nullary.”

Performance and Tuning Series: Query Processing and Abstract Plans 51

Query plan shape

Another way to get a graphical representation of the query plan is to use the
command set statistics plancost on. See Adaptive Server Reference Manual:
Commands for more information. This command is used to compare the
estimated and actual costs in a query plan. It prints its output as a semigraphical
tree representing the query plan tree. It is a very useful tool for diagnosing
query performance problems.

Query plan operators

The query plan operators, and a description of each, are listed in Table 1-3 on
page 23. This section contains additional messages that give more detailed
information about each operator.

EMIT operator
The EMIT operator appears at the top of every query plan. EMIT is the root of
the query plan tree and always has exactly one child operator. The EMIT
operator routes the result rows of the query by sending them to the client (an
application or another Adaptive Server instance) or by assigning values from
the result row to local variables or fetch into variables.

SCAN operator
The SCAN operator reads rows into the query plan and makes them available
for further processing by the other operators in the query plan. The SCAN
operator is a leaf operator; that is, it never has any child operators. The SCAN
operator can read rows from multiple sources, so the showplan message
identifying it is always followed by a FROM message to identify what kind of
SCAN is being performed. The FROM messages are: FROM CACHE, FROM OR,
FROM LIST, and FROM TABLE.

FROM cache message

This message shows that a CACHE SCAN operator is reading a single-row in-
memory table.

52 Adaptive Server Enterprise

 CHAPTER 2 Using showplan

FROM or LIST
An OR list has as many as N rows; one for each distinct OR or IN value specified
in the query.
The first message shows that an OR scan is reading rows from an in-memory
table that contains values from an IN list or multiple or clauses on the same
column. The OR list appears only in query plans that use the special or strategy
for in lists. The second message shows the maximum number of rows (N) that
the in-memory table can have. Since OR list eliminates duplicate values when
filling the in-memory table, N may be less than the number of values appearing
in the SQL statement. As an example, the following query generates a query
plan with the special or strategy and an OR list:
select s.id from sysobjects s where s.id in (1, 0, 1, 2, 3)
go

QUERY PLAN FOR STATEMENT 1 (at line 1).

STEP 1
The type of query is SELECT.

4 operator(s) under root

ROOT:EMIT Operator (VA = 4)

|NESTED LOOP JOIN Operator (VA = 3) (Join Type: Inner Join)

|
| |SCAN Operator (VA = 2)
| | FROM OR List
| | OR List has up to 5 rows of OR/IN values.
|
| |RESTRICT Operator (VA = 2)(0)(0)(0)(8)(0)
| | |SCAN Operator (VA = 1)
| | | FROM TABLE
| | | sysobjects
| | | s
| | | Using Clustered Index.
| | | Index : csysobjects
| | | Forward Scan.
| | | Positioning by key.
| | | Index contains all needed columns. Base table will not be read.
| | | Keys are:
| | | id ASC
| | | Using I/O Size 2 Kbytes for index leaf pages.
| | | With LRU Buffer Replacement Strategy for index leaf pages.

Performance and Tuning Series: Query Processing and Abstract Plans 53

Query plan shape

This example has five values in the IN list, but only four are distinct, so the OR
list puts only the four distinct values in its in-memory table. In the example
query plan, the OR list is the left-child operator of the NESTED LOOP JOIN
operator and a SCAN operator is the right child of the NESTED LOOP JOIN
operator. When this plan executes, the NESTED LOOP JOIN operator calls the
or command to return a row from its in-memory table, then the NESTED LOOP
JOIN operator calls on the SCAN operator to find all matching rows (one at a
time), using the clustered index for lookup. This example query plan is much
more efficient than reading all of the rows of sysobjects and comparing the
value of sysobjects.id in each row to the five values in the IN list.

FROM TABLE
FROM TABLE shows that a PARTITION SCAN operator is reading a database
table. A second message gives the table name, and, if there is a correlation
name, it is printed on the next line. Under the FROM TABLE message in the
previous example output, sysobjects is the table name and s is the correlation
name. The previous example also shows additional messages under the FROM
TABLE message. These messages give more information about how the
PARTITION SCAN operator is directing the access layer of Adaptive Server to
get the rows from the table being scanned.
The messages below indicate whether the scan is a table scan or an index scan:
• Table Scan – the rows are fetched by reading the pages of the table.
• Using Clustered Index – a clustered index is used to fetch the rows
of the table.
• Index: indexname – an index is used to fetch the table rows. If this message
is not preceded by “using clustered index,” a nonclustered index is
used. indexname is the name of the index that will be used.
These messages indicate the direction of a table or index scan. The scan
direction depends on the ordering specified when the indexes were created and
the order specified for columns in the order by clause or other useful orderings
that can be exploited by operators further up in the query plan (for example, a
sorted ordering for a merge-join strategy).
Backward scans can be used when the order by clause contains the ascending
or descending qualifiers on index keys, exactly opposite of those in the create
index clause.

Forward scan

54 Adaptive Server Enterprise

 CHAPTER 2 Using showplan

Backward scan
The scan-direction messages are followed by positioning messages, which
describe how access to a table or to the leaf level of an index takes place:
• Positioning at start of table – a table scan that starts at the first
row of the table and goes forward.
• Positioning at end of table – a table scan that starts at the last row
of the table and goes backward.
• Positioning by key – the index is used to position the scan at the first
qualifying row.
• Positioning at index start/positioning at index end – these
messages are similar to the corresponding messages for table scans, except
that an index is being scanned instead of a table.
If the scan can be limited due to the nature of the query, the following messages
describe how:
• Scanning only the last page of the table – appears when the
scan uses an index and is searching for the maximum value for scalar
aggregation. If the index is on the column whose maximum is sought, and
the index values are in ascending order, the maximum value will be on the
last page.
• Scanning only up to the first qualifying row – appears when
the scan uses an index and is searching for the minimum value for scalar
aggregation.

Note If the index key is sorted in descending order, the above messages for
minimum and maximum aggregates are reversed.

In some cases, the index being scanned contains all of the columns of the table
that are needed in the query. In such a case, this message is printed:
Index contains all needed columns. Base table will not
be read.
If an index contains all the columns needed by the query, the optimizer may
choose an Index Scan over a Table Scan even though there are no useful
keys on the index columns. The amount of I/O required to read the index can
be significantly less than that required to read the base table. Index scans that
do not require base table pages to be read are called covered index scans.
If an index scan is using keys to position the scan, this message prints:

Performance and Tuning Series: Query Processing and Abstract Plans 55

Query plan shape

Keys are:
Key <ASC/DESC>
This message shows the names of the columns used as keys (each key on its
own output line) and shows the index ordering on that key: ASC for ascending
and DESC for descending.
After the messages that describe the type of access being used by the scan
operator, messages about the I/O sizes and buffer cache strategy are printed.

I/O size messages

The I/O messages are:
Using I/O size N Kbtyes for data pages.
Using I/O size N Kbtyes for index leaf pages.
These messages report the I/O sizes used in the query. Possible I/O sizes are 2,
4, 8, and 16 kilobytes.
If the table, index, or database used in the query uses a data cache with large
I/O pools, the optimizer can choose large I/O. It can choose to use one I/O size
for reading index leaf pages, and a different size for data pages. The choice
depends on the pool size available in the cache, the number of pages to be read,
the cache bindings for the objects, and the cluster ratio for the table or index
pages.
Either (or both) of these messages can appear in the showplan output for a SCAN
operator. For a table scan, only the first message is printed; for a covered index
scan, only the second message is printed. For an Index Scan that requires
base table access, both messages are printed.
After each I/O size message, a cache strategy message is printed:
With <LRU/MRU> Buffer Replacement Strategy for data
pages.
With <LRU/MRU> Buffer Replacement Strategy for index
leaf pages.
In an LRU replacement strategy, the most recently accessed pages are
positioned in the cache to be retained as long as possible. In an MRU
Replacement Strategy, the most recently accessed pages are positioned in the
cache for quick replacement.
Sample I/O and cache messages are shown in the following query:
use pubs2
go

56 Adaptive Server Enterprise

 CHAPTER 2 Using showplan

set showplan on
go
select au_fname, au_lname, au_id from authors
where au_lname = "Williams"
go

QUERY PLAN FOR STATEMENT 1 (at line 1).

1 operator(s) under root

The type of query is SELECT.

ROOT:EMIT Operator (VA = 1)

|SCAN Operator (VA = 0)

| FROM TABLE
| authors
| Index : aunmind
| Forward Scan.
| Positioning by key.
| Keys are:
| au_lname ASC
| Using I/O Size 2 Kbytes for index leaf pages.
| With LRU Buffer Replacement Strategy for index leaf pages.
| Using I/O Size 2 Kbytes for data pages.
| With LRU Buffer Replacement Strategy for data pages.
The SCAN operator of the authors table uses the index aunmind, but must also
read the base table pages to get all of the required columns from authors. In this
example, there are two I/O size messages, each followed by the corresponding
buffer replacement message.
There are two kinds of table SCAN operators that have their own messages—
the RID SCAN and the LOG SCAN.

RID scan
The Positioning by Row IDentifier (RID) scan is found only in query
plans that use the second or strategy that the optimizer can choose, the general
or strategy. The general or strategy may be used when multiple or clauses are
present on different columns. An example of a query for which the optimizer
can choose a general or strategy and its showplan output is:
use pubs2
go
set showplan on

Performance and Tuning Series: Query Processing and Abstract Plans 57

Query plan shape

go
select id from sysobjects where id = 4 or name = 'foo'

QUERY PLAN FOR STATEMENT 1 (at line 1).

6 operator(s) under root

The type of query is SELECT.

ROOT:EMIT Operator (VA = 6)

|RID JOIN Operator (VA = 5)

| Using Worktable2 for internal storage.
|
| |HASH UNION Operator has 2 children.
| | Using Worktable1 for internal storage.
| |
| | |SCAN Operator (VA = 0)
| | | FROM TABLE
| | | sysobjects
| | | Using Clustered Index.
| | | Index : csysobjects
| | | Forward Scan.
| | | Positioning by key.
| | | Index contains all needed columns. Base table will not be read.
| | | Keys are:
| | | id ASC
| | | Using I/O Size 2 Kbytes for index leaf pages.
| | | With LRU Buffer Replacement Strategy for index leaf pages.
| |
| | |SCAN Operator (VA = 1)
| | | FROM TABLE
| | | sysobjects
| | | Index : ncsysobjects
| | | Forward Scan.
| | | Positioning by key.
| | | Index contains all needed columns. Base table will not be read.
| | | Keys are:
| | | name ASC
| | | Using I/O Size 2 Kbytes for index leaf pages.
| | | With LRU Buffer Replacement Strategy for index leaf pages.
|
| |RESTRICT Operator (VA = 4)(0)(0)(0)(11)(0)
| |
| | |SCAN Operator(VA = 3)
| | | FROM TABLE

58 Adaptive Server Enterprise

 CHAPTER 2 Using showplan

| | | sysobjects
| | | Using Dynamic Index.
| | | Forward Scan.
| | | Positioning by Row IDentifier (RID).
| | | Using I/O Size 2 Kbytes for data pages.
| | | With LRU Buffer Replacement Strategy for data pages.
In this example, the where clause contains two disjunctions, each on a different
column (id and name). There are indexes on each of these columns (csysobjects
and ncsysobjects), so the optimizer chose a query plan that uses an index scan
to find all rows whose id column is 4 and another index scan to find all rows
whose name is “foo.”
Since it is possible that a single row has both an ID of 4 and a name of “foo,”
that row would appear twice in the result set. To eliminate these duplicate rows,
the index scans return only the row identifiers (RIDs) of the qualifying rows.
The two streams of RIDs are concatenated by the HASH UNION operator, which
also removes any duplicate RIDs.
]The stream of unique RIDs is passed to the RID JOIN operator. The rid join
operator creates a worktable and fills it with a single-column row with each
RID. The RID JOIN operator then passes its worktable of RIDs to the RID
SCAN operator. The RID SCAN operator passes the worktable to the access
layer, where it is treated as a keyless nonclustered index and the rows
corresponding to the RIDs are fetched and returned.
The last SCAN in the showplan output is the RID SCAN. As can be seen from
the example output, the RID SCAN output contains many of the messages
already discussed above, but it also contains two messages that are printed only
for the RID SCAN:
• Using Dynamic Index – indicates the SCAN is using the worktable with
RIDs that was built during execution by the RID JOIN operator as an
index to locate the matching rows.
• Positioning by Row Identifier (RID) – indicates the rows are
being located directly by the RID.

Log Scan
Log Scan appears only in triggers that access inserted or deleted tables. These
tables are dynamically built by scanning the transaction log when the trigger is
executed. Triggers are executed only after insert, delete, or update queries
modify a table with a trigger defined on it for the specific query type. The
following example is a delete query on the titles table, which has a delete trigger
called deltitle defined on it:

Performance and Tuning Series: Query Processing and Abstract Plans 59

Query plan shape

use pubs2
go
set showplan on
go
delete from titles where title_id = 'xxxx'

QUERY PLAN FOR STATEMENT 1 (at line 1).

STEP 1
The type of query is DELETE.

2 operator(s) under root

|ROOT:EMIT Operator (VA = 2)

|DELETE Operator (VA = 1)

| The update mode is direct.
|
| |SCAN Operator (VA = 0)
| | FROM TABLE
| | titles
| | Using Clustered Index.
| | Index : titleidind
| | Forward Scan.
| | Positioning by key.
| | Keys are:
| | title_id ASC
| | Using I/O Size 2 Kbytes for data pages.
| | With LRU Buffer Replacement Strategy for data pages.
|
| TO TABLE
| titles
| Using I/O Size 2 Kbytes for data pages.
The showplan output up to this point is for the actual delete query. The output
below is for the trigger, deltitle.
QUERY PLAN FOR STATEMENT 1 (at line 5).

STEP 1

The type of query is COND.

6 operator(s) under root

ROOT:EMIT Operator (VA = 6)

60 Adaptive Server Enterprise

 CHAPTER 2 Using showplan

|RESTRICT Operator (VA = 5)(0)(0)(0)(5)(0)

|
| |SCALAR AGGREGATE Operator (VA = 4)
| | Evaluate Ungrouped COUNT AGGREGATE.
| |
| | |MERGE JOIN Operator (Join Type: Inner Join) (VA = 3)
| | | Using Worktable2 for internal storage.
| | | Key Count: 1
| | | Key Ordering: ASC
| | |
| | | |SORT Operator (VA = 1)
| | | | Using Worktable1 for internal storage.
| | | |
| | | | |SCAN Operator (VA = 0)
| | | | | FROM TABLE
| | | | | titles
| | | | | Log Scan.
| | | | | Forward Scan.
| | | | | Positioning at start of table.
| | | | | Using I/O Size 2 Kbytes for data pages.
| | | | | With MRU Buffer Replacement Strategy for data pages.
| | |
| | | |SCAN Operator (VA = 2)
| | | | FROM TABLE
| | | | salesdetail
| | | | Index : titleidind
| | | | Forward Scan.
| | | | Positioning at index start.
| | | | Index contains all needed columns. Base table will not be
read.
| | | | Using I/O Size 2 Kbytes for index leaf pages.
| | | | With LRU Buffer Replacement Strategy for index leaf pages.

QUERY PLAN FOR STATEMENT 2 (at line 8).

STEP 1
The type of query is ROLLBACK TRANSACTION.

QUERY PLAN FOR STATEMENT 3 (at line 9).

STEP 1
The type of query is PRINT.

QUERY PLAN FOR STATEMENT 4 (at line 0).

STEP 1
The type of query is GOTO.

Performance and Tuning Series: Query Processing and Abstract Plans 61

Query plan shape

The procedure that defines the deltitle trigger consists of four SQL statements.
Use sp_helptext deltitle to display the text of deltitle. The first statement in
deltitle has been compiled into a query plan, the other three statements are
compiled into legacy query plans and are executed by the procedural execution
engine, not the query execution engine.
The showplan output for the SCAN operator for the titles table indicates that it is
doing a scan of the log by printing Log Scan.

DELETE, INSERT, and UPDATE operators

The DELETE, INSERT, and UPDATE operators usually have only one child
operator. However, they can have as many as two additional child operators to
enforce referential integrity constraints and to deallocate text data in the case
of alter table drop of a text column.
These operators modify data by inserting, deleting, or updating rows belonging
to a target table.
Child operators of DML operators can be SCAN operators, JOIN operators, or
any data streaming operator.
The data modification can be done using different update modes, as specified
by this message:
The Update Mode is <Update Mode>.
The table update mode may be direct, deferred, deferred for an
index, or deferred for a variable column. The update mode for a
worktable is always direct.
The target table for the data modification is displayed in this message:
TO TABLE
<Table Name>
Also displayed is the I/O size used for the data modification:
Using I/O Size <N> Kbytes for data pages.
The next example uses the DELETE operator:
use pubs2
go
set showplan on
go
delete from authors where postalcode = '90210'

QUERY PLAN FOR STATEMENT 1 (at line 1).

62 Adaptive Server Enterprise

 CHAPTER 2 Using showplan

STEP 1
The type of query is DELETE.

2 operator(s) under root

ROOT:EMIT Operator (VA = 2)

|DELETE Operator (VA = 1)

| The update mode is direct.
|
| |SCAN Operator (VA = 0)
| | FROM TABLE
| | authors
| | Table Scan.
| | Forward Scan.
| | Positioning at start of table.
| | Using I/O Size 4 Kbytes for data pages.
| | With LRU Buffer Replacement Strategy for data pages.
|
| TO TABLE
| authors
| Using I/O Size 4 Kbytes for data pages.

TEXT DELETE operator

Another type of query plan where DELETE, INSERT, and UPDATE operator can
have more than one child operator is the alter table drop textcol command,
where textcol is the name of a column whose datatype is text, image, or unitext.
This version of command used the TEXT DELETE operator in its query plan.
For example:
use tempdb
go
create table t1 (c1 int, c2 text, c3 text)
go
set showplan on
go
alter table t1 drop c2

QUERY PLAN FOR STATEMENT 1 (at line 1).

Optimized using the Abstract Plan in the PLAN clause.

STEP 1
The type of query is ALTER TABLE.

Performance and Tuning Series: Query Processing and Abstract Plans 63

Query plan shape

5 operator(s) under root

ROOT:EMIT Operator (VA = 5)

|INSERT Operator (VA = 52)

| The update mode is direct.
|
| |RESTRICT Operator (VA = 1)(0)(0)(3)(0)(0)
| |
| | |SCAN Operator (VA = 0)
| | | FROM TABLE
| | | t1
| | | Table Scan.
| | | Forward Scan.
| | | Positioning at start of table.
| | | Using I/O Size 2 Kbytes for data pages.
| | | With LRU Buffer Replacement Strategy for data pages.
|
| |TEXT DELETE Operator
| | The update mode is direct.
| |
| | |SCAN Operator (VA = 3)
| | | FROM TABLE
| | | t1
| | | Table Scan.
| | | Forward Scan.
| | | Positioning at start of table.
| | | Using I/O Size 2 Kbytes for data pages.
| | | With LRU Buffer Replacement Strategy for data pages.
|
| TO TABLE
| #syb__altab
| Using I/O Size 2 Kbytes for data pages.
One of the two text columns in t1 is dropped, using the alter table command.
The showplan output has the appearance of a select into query plan because alter
table internally generated a select into query plan.

The INSERT operator calls on its left child operator, the SCAN of t1, to read the
rows of t1, and builds new rows with only the c1 and c3 columns inserted into
#syb_altab. When all the new rows have been inserted into #syb_altab, the
INSERT operator calls on its right child, the TEXT DELETE operator, to delete
the text page chains for the c2 columns that have been dropped from t1.
Postprocessing replaces the original pages of t1 with those of #syb_altab to
complete the alter table command.

64 Adaptive Server Enterprise

 CHAPTER 2 Using showplan

The TEXT DELETE operator appears only in alter table commands that drop
some, but not all text columns of a table, and it always appears as the right child
of an INSERT operator.
The TEXT DELETE operator displays the update mode message, exactly like
the INSERT, UPDATE, and DELETE operators.

Query plans for referential integrity enforcement

When the INSERT, UPDATE, and DELETE operators are used on a table that has
one or more referential integrity constraints, the showplan output also shows
the DIRECT RI FILTER and DEFERRED RI FILTER child operators of the
DML operator. The type of referential integrity constraint determines whether
one or both of these operators are present.
The following example is for an insert into the titles table of the pubs3 database.
This table has a column called pub_id that references the pub_id column of the
publishers table. The referential integrity constraint on titles.pub_id requires
that every value that is inserted into titles.pub_id must have a corresponding
value in publishers.pub_id.
The query and its query plan are:
use pubs3
go
set showplan on
insert into titles values ("AB1234", "Abcdefg", "test", "9999", 9.95, 1000.00,
10, null, getdate(),1)

QUERY PLAN FOR STATEMENT 1 (at line 1).

STEP 1
The type of query is INSERT.

4 operator(s) under root

ROOT:EMIT Operator (VA = 3)

|INSERT Operator (VA = 2)

| The update mode is direct.
|
| |SCAN Operator (VA = 1)
| | FROM CACHE
|
| |DIRECT RI FILTER Operator has 1 children.

Performance and Tuning Series: Query Processing and Abstract Plans 65

Query plan shape

| |
| | |SCAN Operator (VA = 0)
| | | FROM TABLE
| | | publishers
| | | Index : publishers_6240022232
| | | Forward Scan.
| | | Positioning by key.
| | | Index contains all needed columns. Base table will not be
read.
| | | Keys are:
| | | pub_id ASC
| | | Using I/O Size 2 Kbytes for index leaf pages.
| | | With LRU Buffer Replacement Strategy for index leaf pages.
|
| TO TABLE
| titles
| Using I/O Size 2 Kbytes for data pages.
In the query plan, the INSERT operator’s left child operator is a CACHE SCAN,
which returns the row of values to be inserted into titles. The INSERT
operator’s right child is a DIRECT RI FILTER operator.
The DIRECT RI FILTER operator executes a scan of the publishers table to
find a row with a value of pub_id that matches the value of pub_id to be inserted
into titles. If a matching row is found, the DIRECT RI FILTER operator allows
the insert to proceed, but if a matching value of pub_id is not found in
publishers, the DIRECT RI FILTER operator aborts the command.

In this example, the DIRECT RI FILTER can check and enforce the referential
integrity constraint on titles for each row that is inserted, as it is inserted.
The next example shows a DIRECT RI FILTER operating in a different mode,
together with a DEFERRED RI FILTER operator:
use pubs3
go
set showplan on
go
update publishers set pub_id = '0001'

QUERY PLAN FOR STATEMENT 1 (at line 1).

STEP 1
The type of query is UPDATE.

13 operator(s) under root

ROOT:EMIT Operator (VA = 13)

66 Adaptive Server Enterprise

 CHAPTER 2 Using showplan

|UPDATE Operator (VA = 1)

| The update mode is deferred_index.
|
| |SCAN Operator (VA = 0)
| | FROM TABLE
| | publishers
| | Table Scan.
| | Forward Scan.
| | Positioning at start of table.
| | Using I/O Size 2 Kbytes for data pages.
| | With LRU Buffer Replacement Strategy for data pages.
|
| |DIRECT RI FILTER Operator (VA = 7) has 1 children.
| |
| | |INSERT Operator (VA = 6)
| | | The update mode is direct.
| | |
| | | |SQFILTER Operator (VA = 5) has 2 children.
| | | |
| | | | |SCAN Operator (VA = 2)
| | | | | FROM CACHE
| | | |
| | | | Run subquery 1 (at nesting level 0).
| | | |
| | | |QUERY PLAN FOR SUBQUERY 1 (at nesting level 0 and at
line 0).
| | | |
| | | | Non-correlated Subquery.
| | | | Subquery under an EXISTS predicate.
| | | |
| | | | |SCALAR AGGREGATE Operator (VA = 4)
| | | | | Evaluate Ungrouped ANY AGGREGATE.
| | | | | Scanning only up to the first qualifying row.
| | | | |
| | | | | |SCAN Operator (VA = 3)
| | | | | | FROM TABLE
| | | | | | titles
| | | | | | Table Scan.
| | | | | | Forward Scan.
| | | | | | Positioning at start of table.
| | | | | | Using I/O Size 2 Kbytes for data pages.
| | | | | | With LRU Buffer Replacement strategy for data
pages.
| | | |
| | | | END OF QUERY PLAN FOR SUBQUERY 1.

Performance and Tuning Series: Query Processing and Abstract Plans 67

Query plan shape

| | |
| | | TO TABLE
| | | Worktable1.
|
| |DEFERRED RI FILTER Operator has (VA = 12) 1 children.
| |
| | |SQFILTER Operator (VA = 11) has 2 children.
| | |
| | | |SCAN Operator (VA = 8)
| | | | FROM TABLE
| | | | Worktable1.
| | | | Table Scan.
| | | | Forward Scan.
| | | | Positioning at start of table.
| | | | Using I/O Size 2 Kbytes for data pages.
| | | | With LRU Buffer Replacement Strategy for data pages.
| | |
| | | Run subquery 1 (at nesting level 0).
| | |
| | | QUERY PLAN FOR SUBQUERY 1 (at nesting level 0 and at line 0).
| | |
| | | Non-correlated Subquery.
| | | Subquery under an EXISTS predicate.
| | |
| | | |SCALAR AGGREGATE Operator (VA = 10)
| | | | Evaluate Ungrouped ANY AGGREGATE.
| | | | Scanning only up to the first qualifying row.
| | | |
| | | | |SCAN Operator (VA = 9)
| | | | | FROM TABLE
| | | | | publishers
| | | | | Index : publishers_6240022232
| | | | | Forward Scan.
| | | | | Positioning by key.
| | | | | Index contains all needed columns. Base table will
not be read.
| | | | | Keys are:
| | | | | pub_id ASC
| | | | | Using I/O Size 2 Kbytes for index leaf pages.
| | | | | With LRU Buffer Replacement Strategy for index leaf
pages.
| | |
| | | END OF QUERY PLAN FOR SUBQUERY 1.|
| TO TABLE
| publishers
| Using I/O Size 2 Kbytes for data pages.

68 Adaptive Server Enterprise

 CHAPTER 2 Using showplan

The referential integrity constraint on titles requires that for every value of
titles.pub_id there must exist a value of publishers.pub_id. However, this
example query is changing the values of publisher.pub_id, so a check must be
made to maintain the referential integrity constraint.
The example query can change the value of publishers.pub_id for several rows
in publishers, so a check to make sure that all of the values of titles.pub_id still
exist in publisher.pub_id cannot be done until all rows of publishers have been
processed.
This example calls for deferred referential integrity checking: as each row of
publishers is read, the UPDATE operator calls upon the DIRECT RI FILTER
operator to search titles for a row with the same value of pub_id as the value that
is about to be changed. If a row is found, it indicates that this value of pub_id
must still exist in publishers to maintain the referential integrity constraint on
titles, so the value of pub_id is inserted into WorkTable1.

After all of the rows of publishers have been updated, the UPDATE operator calls
upon the DEFERRED RI FILTER operator to execute its subquery to verify that
all of the values in Worktable1 still exist in publishers. The left child operator of
the DEFERRED RI FILTER is a SCAN which reads the rows from Worktable1.
The right child is a SQFILTER operator that executes an existence subquery to
check for a matching value in publishers. If a matching value is not found, the
command is aborted.
The examples in this section used simple referential integrity constraints,
between only two tables. Adaptive Server allows up to 192 constraints per
table, so it can generate much more complex query plans. When multiple
constraints must be enforced, there is still only a single DIRECT RI FILTER
or DEFERRED RI FILTER operator in the query plan, but these operators can
have multiple subplans, one for each constraint that must be enforced.

JOIN operators
Adaptive Server provides four primary JOIN operator strategies: NESTED
LOOP JOIN, MERGE JOIN, HASH JOIN, and NARY NESTED LOOP JOIN,
which is a variant of NESTED LOOP JOIN. In versions earlier than 15.0,
NESTED LOOP JOIN was the primary JOIN strategy. MERGE JOIN was also
available, but was, by default, not enabled.
Each JOIN operator is described in further detail below, including a general
description of the each algorithm. These descriptions give a high-level
overview of the processing required for each JOIN strategy.

Performance and Tuning Series: Query Processing and Abstract Plans 69

Query plan shape

NESTED LOOP JOIN

NESTED LOOP JOIN, the simplest join strategy, is a binary operator with the
left child forming the outer data stream and the right child forming the inner
data stream.
For every row from the outer data stream, the inner data stream is opened.
Often, the right child is a scan operator. Opening the inner data stream
effectively positions the scan on the first row that qualifies all of the searchable
arguments.
The qualifying row is returned to the NESTED LOOP JOIN’s parent operator.
Subsequent calls to the join operator continue to return qualifying rows from
the inner stream.
After the last qualifying row from the inner stream is returned for the current
outer row, the inner stream is closed. A call is made to get the next qualifying
row from the outer stream. The values from this row provide the searchable
arguments used to open and position the scan on the inner stream. This process
continues until the NESTED LOOP JOIN’s left child returns End Of Scan.
-- Collect all of the title ids for books written by "Bloom".
select ta.title_id
from titleauthor ta, authors a
where a.au_id = ta.au_id
and au_lname = "Bloom"
go

QUERY PLAN FOR STATEMENT 1 (at line 2).

STEP 1
The type of query is SELECT.

3 operator(s) under root

ROOT:EMIT Operator (VA = 3)

|NESTED LOOP JOIN Operator (Join Type: Inner Join)

|
| |SCAN Operator (VA = 0)
| | FROM TABLE
| | authors
| | a
| | Index : aunmind
| | Forward Scan.
| | Positioning by key.
| | Keys are:

70 Adaptive Server Enterprise

 CHAPTER 2 Using showplan

| | au_lname ASC
| | Using I/O Size 2 Kbytes for index leaf pages.
| | With LRU Buffer Replacement Strategy for index leaf pages.
| | Using I/O Size 2 Kbytes for data pages.
| | With LRU Buffer Replacement Strategy for data pages.
|
| |SCAN Operator (VA = 1)
| | FROM TABLE
| | titleauthor
| | ta
| | Using Clustered Index.
| | Index : taind
| | Forward Scan.
| | Positioning by key.
| | Keys are:
| | au_id ASC
| | Using I/O Size 2 Kbytes for data pages.
| | With LRU Buffer Replacement Strategy for data pages.
The authors table is joined with the titleauthor table. A NESTED LOOP JOIN
strategy has been chosen. The NESTED LOOP JOIN operator’s type is “Inner
Join.” First, the authors table is opened and positioned on the first row (using
the aunmind index) containing an l_name value of “Bloom.” Then, the
titleauthor table is opened and positioned on the first row with an au_id equal to
the au_id value of the current authors’ row using the clustered index “taind.” If
there is no useful index for lookups on the inner stream, the optimizer may
generate a reformatting strategy.
Generally, a NESTED LOOP JOIN strategy is effective when there is a useful
index available for qualifying the join predicates on the inner stream.

MERGE JOIN
The MERGE JOIN operator is a binary operator. The left and right children are
the outer and inner data streams, respectively. Both data streams must be sorted
on the MERGE JOIN’s key values.
First, a row from the outer stream is fetched. This initializes the MERGE JOIN’s
join key values. Then, rows from the inner stream are fetched until a row with
key values that match or are greater than (less than if key column is
descending) is encountered. If the join key matches, the qualifying row is
passed on for additional processing, and a subsequent next call to the MERGE
JOIN operator continues fetching from the currently active stream.

Performance and Tuning Series: Query Processing and Abstract Plans 71

Query plan shape

If the new values are greater than the current comparison key, these values are
used as the new comparison join key while fetching rows from the other
stream. This process continues until one of the data streams is exhausted.
Generally, the MERGE JOIN strategy is effective when a scan of the data
streams requires that most of the rows must be processed, and that, if any of the
input streams are large, they are already sorted on the join keys.
select ta.title_id
from titleauthor ta, authors a
where a.au_id = ta.au_id
and au_lname = "Bloom"
go

QUERY PLAN FOR STATEMENT 1 (at line 2).

STEP 1
The type of query is EXECUTE.
Executing a newly cached statement.

QUERY PLAN FOR STATEMENT 1 (at line 1).

STEP 1
The type of query is SELECT.

3 operator(s) under root

ROOT:EMIT Operator (VA = 3)

|MERGE JOIN Operator (Join Type: Inner Join)

| Using Worktable2 for internal storage.
| Key Count: 1
| Key Ordering: ASC
|
| |SORT Operator
| | Using Worktable1 for internal storage.
| |
| | |SCAN Operator
| | | FROM TABLE
| | | authors
| | | a
| | | Index : aunmind
| | | Forward Scan.
| | | Positioning by key.
| | | Keys are:
| | | au_lname ASC
| | | Using I/O Size 2 Kbytes for index leaf pages.

72 Adaptive Server Enterprise

 CHAPTER 2 Using showplan

| | | With LRU Buffer Replacement Strategy for index leaf pages.

| | | Using I/O Size 2 Kbytes for data pages.
| | | With LRU Buffer Replacement Strategy for data pages.
|
| |SCAN Operator
| | FROM TABLE
| | titleauthor
| | ta
| | Index : auidind
| | Forward Scan.
| | Positioning at index start.
| | Using I/O Size 2 Kbytes for index leaf pages.
| | With LRU Buffer Replacement Strategy for index leaf pages.
| | Using I/O Size 2 Kbytes for data pages.
| | With LRU Buffer Replacement Strategy for data pages.
In this example, a sort operator is the left child, or outer stream. The data source
for the sort operator is the authors table. The sort operator is required because
the authors table has no index on au_id that would otherwise provide the
necessary sorted order. A scan of the titleauthor table is the right child/inner
stream. The scan uses the auidind index, which provides the necessary ordering
for the MERGE JOIN strategy.
A row is fetched from the outer stream (the authors table is the original source)
to establish an initial join key comparison value. Then rows are fetched from
the titleauthor table until a row with a join key equal to or greater than the
comparison key is found.
Inner stream rows with matching keys are stored in a cache in case they need
to be refetched. These rows are refetched when the outer stream contains
duplicate keys. When a titleauthor.au_id value that is greater than the current
join key comparison value is fetched, the MERGE JOIN operator starts fetching
from the outer stream until a join key value equal to or greater than the current
titleauthor.au_id value is found. The scan of the inner stream resumes at that
point.
The MERGE JOIN operator’s showplan output contains a message indicating the
worktable to be used for the inner stream’s backing store. The worktable is
written to if the inner rows with duplicate join keys no longer fits in cached
memory. The width of a cached row is limited to 64 kilobytes.

Performance and Tuning Series: Query Processing and Abstract Plans 73

Query plan shape

HASH JOIN
The HASH JOIN operator is a binary operator. The left child generates the build
input stream. The right child generates the probe input stream. The build set is
generated by completely draining the build input stream when the first row is
requested from the HASH JOIN operator. Every row is read from the input
stream and hashed into an appropriate bucket using the hash key.
If there is not enough memory to hold the entire build set, then a portion of it
spills to disk. This portion is referred to as a hash partition and should not be
confused with table partitions. A hash partition consists of a collection of hash
buckets. After the entire left child’s stream has been drained, the probe input is
read.
Each row from the probe set is hashed. A lookup is done in the corresponding
build bucket to check for rows with matching hash keys. This occurs if the
build set’s bucket is memory resident. If it has been spilled, the probe row is
written to the corresponding spilled probe partition. When a probe row’s key
matches a build row’s key, then the necessary projection of the two row’s
columns is passed up for additional processing.
Spilled partitions are processed in subsequent recursive passes of the HASH
JOIN algorithm. New hash seeds are used in each pass so that the data is
redistributed across different hash buckets. This recursive processing continues
until the last spilled partition is completely memory resident. When a hash
partition from the build set contains many duplicates, the HASH JOIN operator
reverts back to NESTED LOOP JOIN processing.
Generally, the HASH JOIN strategy is good in cases where most of the rows
from the source sets must be processed and there are no inherent useful
orderings on the join keys or there are no interesting orderings that can be
promoted to calling operators (for example, an order by clause on the join key).
HASH JOINs perform particularly well if one of the data sets is small enough
to be memory resident. In this case, no spilling occurs and no I/O is needed to
perform that HASH JOIN algorithm.
select ta.title_id
from titleauthor ta, authors a
where a.au_id = ta.au_id
and au_lname = "Bloom"

QUERY PLAN FOR STATEMENT 1 (at line 2).

3 operator(s) under root

The type of query is SELECT.

74 Adaptive Server Enterprise

 CHAPTER 2 Using showplan

ROOT:EMIT Operator

|HASH JOIN Operator (Join Type: Inner Join)

| Using Worktable1 for internal storage.
|
| |SCAN Operator
| | FROM TABLE
| | authors
| | a
| | Index : aunmind
| | Forward Scan.
| | Positioning by key.
| | Keys are:
| | au_lname ASC
| | Using I/O Size 2 Kbytes for index leaf pages.
| | With LRU Buffer Replacement Strategy for index leaf pages.
| | Using I/O Size 2 Kbytes for data pages.
| | With LRU Buffer Replacement Strategy for data pages.
|
| |SCAN Operator
| | FROM TABLE
| | titleauthor
| | ta
| | Index : auidind
| | Forward Scan.
| | Positioning at index start.
| | Using I/O Size 2 Kbytes for index leaf pages.
| | With LRU Buffer Replacement Strategy for index leaf pages.
| | Using I/O Size 2 Kbytes for data pages.
| | With LRU Buffer Replacement Strategy for data pages.
In this example, the source of the build input stream is an index scan of
author.aunmind.

Only rows with an au_lname value of “Bloom” are returned from this scan.
These rows are then hashed on their au_id value and placed into their
corresponding hash bucket. After the initial build phase is completed, the probe
stream is opened and scanned. Each row from the source index,
titleauthor.auidind, is hashed on the au_id column. The resulting hash value is
used to determine which bucket in the build set should be searched for
matching hash keys. Each row from the build set’s hash bucket is compared to
the probe row’s hash key for equality. If the row matches, the titleauthor.au_id
column is returned to the EMIT operator.

Performance and Tuning Series: Query Processing and Abstract Plans 75

Query plan shape

The HASH JOIN operator’s showplan output contains a message indicating the
worktable to be used for the spilled partition’s backing store. The input row
width is limited to 64 kilobytes.

NARY NESTED LOOP JOIN operator

The NARY NESTED LOOP JOIN strategy is never evaluated or chosen by the
optimizer. It is an operator that is constructed during code generation. If the
compiler finds series of two or more left-deep NESTED LOOP JOINs, it
attempts to transform them into a NARY NESTED LOOP JOIN operator. Two
additional requirements allow for transformation scan; each NESTED LOOP
JOIN operator has an “inner join” type and the right child of each NESTED
LOOP JOIN is a SCAN operator. A RESTRICT operator is permitted above the
SCAN operator.

NARY NESTED LOOP JOIN execution has a performance benefit over the
execution of a series of NESTED LOOP JOIN operators. The example below
demonstrates a fundamental difference between the two methods of execution.
With a series of NESTED LOOP JOIN, a scan may eliminate rows based on
searchable argument values initialized by an earlier scan. That scan may not be
the one that immediately preceded the failing scan. With a series of NESTED
LOOP JOINs, the previous scan would be completely drained although it has
no effect on the failing scan. This could result in a significant amount of
needless I/O. With NARY NESTED LOOP JOINs, the next row fetched comes
from the scan that produced the failing searchable argument value, which is far
more efficient.
select a.au_id, au_fname, au_lname
from titles t, titleauthor ta, authors a
where a.au_id = ta.au_id
and ta.title_id = t.title_id
and a.au_id = t.title_id
and au_lname = "Bloom"
QUERY PLAN FOR STATEMENT 1 (at line 1).

STEP 1
The type of query is SELECT.

4 operator(s) under root

|ROOT:EMIT Operator (VA = 4)

|
| |N-ARY NESTED LOOP JOIN Operator (VA = 3) has 3 children.
| |

76 Adaptive Server Enterprise

 CHAPTER 2 Using showplan

| | | SCAN Operator (VA = 0)

| | | FROM TABLE
| | | authors
| | | a
| | | Table Scan.
| | | Forward Scan.
| | | Positioning at start of table.
| | | Using I/O Size 2 Kbytes for data pages.
| | | With LRU Buffer Replacement Strategy for data pages.
|
| | |SCAN Operator (VA = 1)
| | | FROM TABLE
| | | titleauthor
| | | ta
| | | Table Scan.
| | | Forward Scan.
| | | Positioning at start of table.
| | | Using I/O Size 2 Kbytes for data pages.
| | | With LRU Buffer Replacement Strategy for data pages.
| |
| | | SCAN Operator (VA = 2)
| | | FROM TABLE
| | | titles
| | | t
| | | Index : titles_6720023942
| | | Forward Scan.
| | | Positioning by key.
| | | Index contains all needed columns. Base table will not be read.
| | | Keys are:
| | | title_id ASC
| | | Using I/O Size 2 Kbytes for index leaf pages.
| | | With LRU Buffer Replacement Strategy for index leaf pages.
Figure 2-3 depicts a series of NESTED LOOP JOINs.

Performance and Tuning Series: Query Processing and Abstract Plans 77

Query plan shape

Figure 2-3: Emit operator tree with Nested loop joins

Emit
(VA=6)

NestLoopJoin
InnerJoin
(VA = 5)

NestLoopJoin IndexScan
InnerJoin titleidind (t2)
(VA = 3) (VA = 4)

IndexScan Restrict
aunmid (a) (0) (0) (4) (0)
(VA = 0) (VA = 2)

IndexScan
auidind (ta)
(VA = 1)

All query processor operators are assigned a virtual address. The lines in
Figure 2-3 with VA = report the virtual address for a given operator.
The effective join order is authors, titleauthor, titles. A RESTRICT operator is the
parent operator of the scan on titleauthors. This plan is transformed into the
NARY NESTED LOOP JOIN plan below:

78 Adaptive Server Enterprise

 CHAPTER 2 Using showplan

Figure 2-4: NARY NESTED LOOP JOIN operator

EMIT
(VA=6)

NaryNLJoin
(VA = 4)
NaryNLJoin
(VA = 4)
IndexScan
(VA = 0)
aunmid (a) RESTRICT
(VA = 2) NaryNLJoin
(0) (0) (4) (0) (VA = 4)

IndexScan IndexScan
(VA = 1) (VA = 3)
auidind(ta) titleidind(t)

The transformation retains the original join order of authors, titleauthor, and
titles. In this example, the scan of titles has two searchable arguments on it—
ta.title_id = t.title_id and a.au_id = t.title_id. So, the scan of titles fails because of
the searchable argument value established by the scan of titleauthor, or it fails
because of the searchable argument value established by the scan of authors. If
no rows are returned from a scan of titles because of the searchable argument
value set by the scan of authors, there is no point in continuing the scan of
titleauthor. For every row fetched from titleauthor, the scan of titles fails. It is
only when a new row is fetched from authors that the scan of titles might
succeed. This is why NARY NESTED LOOP JOINs have been implemented;
they eliminate the useless draining of tables that have no impact on the rows
returned by successive scans.
In the example, the NARY NESTED LOOP JOIN operator closes the scan of
titleauthor, fetches a new row from authors, and repositions the scan of
titleauthor based on the au_id fetched from authors. Again, this can be a
significant performance improvement as it eliminates the needless draining of
the titleauthor table and the associated I/O that could occur.

Performance and Tuning Series: Query Processing and Abstract Plans 79

Query plan shape

semijoin
The semijoin is a variant of NESTED LOOP JOIN operator, and includes the
NESTED LOOP JOIN operator in its result set. When you make a semi-join
between two tables, Adaptive Server returns the rows from the first table that
contain one or more matches in the second table (a regular join returns the
matching rows from the first table only once). That is, instead of scanning a
table to return all matching values, an semijoin returns rows when it finds the
first matching value and then stops processing. Semijoins are also known as
“existence joins.”
For example, if you perform a semijoin on the titles and titleauthor tables:
select title
from titles
where title_id in (select title_id from titleauthor)
and title like "A Tutorial%"
QUERY PLAN FOR STATEMENT 1 (at line 1).

STEP 1
The type of query is SELECT.

4 operator(s) under root

|ROOT:EMIT Operator (VA = 4)

|
| |NESTED LOOP JOIN Operator (VA = 3) (Join Type: Left Semi Join)
| |
| | |RESTRICT Operator (VA = 1)(0)(0)(0)(6)(0)
| | |
| | | |SCAN Operator (VA = 0)
| | | | FROM TABLE
| | | | titles
| | | | Index : titleind
| | | | Forward Scan.
| | | | Positioning by key.
| | | | Keys are:
| | | | title ASC
| | | | Using I/O Size 2 Kbytes for index leaf pages.
| | | | With LRU Buffer Replacement Strategy for index leaf pages.
| | | | Using I/O Size 2 Kbytes for data pages.
| | | | With LRU Buffer Replacement Strategy for data pages.
| |
| | |SCAN Operator (VA = 2)
| | | FROM TABLE
| | | titleauthor
| | | Index : titleidind

80 Adaptive Server Enterprise

 CHAPTER 2 Using showplan

| | | Forward Scan.
| | | Positioning by key.
| | | Index contains all needed columns. Base table will not be read.
| | | Keys are:
| | | title_id ASC
| | | Using I/O Size 2 Kbytes for index leaf pages.
| | | With LRU Buffer Replacement Strategy for index leaf pages.

Distinct operators
There are three unary operators you can use to enforce distinctness: GROUP
SORTED Distinct, SORT Distinct, and HASH Distinct. Each has
advantages and disadvantages. The optimizer chooses an efficient distinct
operator with respect to its use within the entire query plan’s context.
See Table 1-3 on page 23 for a list and description of all query processor
operators.

GROUP SORTED Distinct operator

You can use the GROUP SORTED Distinct operator to apply distinctness.
GROUP SORTED Distinct requires that the input stream is already sorted on
the distinct columns. It reads a row from its child operator and initializes the
current distinct columns’ values to be filtered.
The row is returned to the parent operator. When the GROUP SORTED operator
is called again to fetch another row, it fetches another row from its child and
compares the values to the current cached values. If the value is a duplicate, the
row is discarded and the child is called again to fetch a new row.
This process continues until a new distinct row is found. The distinct columns’
values for this row are cached and are used later to eliminate nondistinct rows.
The current row is returned to the parent operator for further processing.
The GROUP SORTED Distinct operator returns a sorted stream. The fact that
it returns a sorted and distinct data stream are properties that the optimizer can
use to improve performance in additional upstream processing. The GROUP
SORTED Distinct operator is a nonblocking operator. It returns a distinct row
to its parent as soon as it is fetched. It does not require the entire input stream
to be processed before it can start returning rows. The following query collects
distinct last and first author’s names:
select distinct au_lname, au_fname
from authors
where au_lname = "Bloom"

Performance and Tuning Series: Query Processing and Abstract Plans 81

Query plan shape

QUERY PLAN FOR STATEMENT 1 (at line 2).

STEP 1
The type of query is SELECT.

2 operator(s) under root

ROOT:EMIT Operator (VA = 2)

|GROUP SORTED Operator (VA = 1)

|Distinct
|
| |SCAN Operator (VA = 0)
| | FROM TABLE
| | authors
| | Index : aunmind
| | Forward Scan.
| | Positioning by key.
| | Index contains all needed columns. Base table will not be read.
| | Keys are:
| | au_lname ASC
| | Using I/O Size 2 Kbytes for index leaf pages.
| | With LRU Buffer Replacement Strategy for index leaf pages.
The GROUP SORTED Distinct operator is chosen in this query plan to apply
the distinct property because the scan operator is returning rows in sorted order
for the distinct columns au_lname and au_fname. GROUP SORTED incurs no I/O
and minimal CPU overhead.
You can use the GROUP SORTED Distinct operator to implement vector
aggregation. See “Vector aggregation operators” on page 84. The showplan
output prints the line Distinct to indicate that this GROUP SORTED
Distinct operator is implementing the distinct property.

SORT Distinct operator

The SORT Distinct operator does not require that its input stream is already
sorted on the distinct key columns. It is a blocking operator that drains its child
operator’s stream and sorts the rows as they are read. A distinct row is returned
to the parent operator after all rows have been sorted. Rows are returned sorted
on the distinct key columns. An internal worktable is used as a backing store in
case the input set does not fit entirely in memory.
QUERY PLAN FOR STATEMENT 1 (at line 1)

82 Adaptive Server Enterprise

 CHAPTER 2 Using showplan

STEP 1
The type of query is SELECT.

2 operator(s) under root

ROOT:EMIT Operator (VA = 2)

|SORT Operator
| Using Worktable1 for internal storage.
|
| |SCAN Operator
| | FROM TABLE
| | authors
| | Table Scan.
| | Forward Scan.
| | Positioning at start of table.
| | Using I/O Size 2 Kbytes for data pages.
| | With LRU Buffer Replacement Strategy for data pages.
The scan of the authors table does not return rows sorted on the distinct key
columns. This requires that a SORT Distinct operator be used rather than a
GROUP SORTED Distinct operator. The SORT operator’s distinct key
columns are au_lname and au_fname. The showplan output indicates that
Worktable1 is used for disk storage in case the input set does not fit entirely in
memory.

HASH Distinct operator

The HASH Distinct operator does not require that its input set be sorted on
the distinct key columns. It is a nonblocking operator. Rows are read from the
child operator and are hashed on the distinct key columns. This determines the
row’s bucket position. The corresponding bucket is searched to see if the key
already exists. The row is discarded if it contains a duplicate key, and another
row is fetched from the child operator. The row is added to the bucket if no
duplicate distinct key already exists and the row is passed up to the parent
operator for further processing. Rows are not returned sorted on the distinct key
columns.
The HASH Distinct operator is generally used when the input set is not
already sorted on the distinct key columns or when the optimizer cannot use the
ordering coming out of the distinct processing later in the plan.
select distinct au_lname, au_fname
from authors
where city = "Oakland"
go

Performance and Tuning Series: Query Processing and Abstract Plans 83

Query plan shape

QUERY PLAN FOR STATEMENT 1 (at line 1).

STEP 1
The type of query is SELECT.

2 operator(s) under root

ROOT:EMIT Operator (VA = 2)

|HASH DISTINCT Operator (VA = 1)

| Using Worktable1 for internal storage.
|
| | SCAN Operator (VA = 0)
| | FROM TABLE
| | authors
| | Table Scan.
| | Forward Scan.
| | Positioning at start of table.
| | Using I/O Size 2 Kbytes for data pages.
| | With LRU Buffer Replacement Strategy for data pages.
In this example, the output of the authors table scan is not sorted. The optimizer
can choose either a SORT Distinct or HASH Distinct operator strategy.
The ordering provided by a SORT Distinct strategy is not useful anywhere
else in the plan, so the optimizer will probably choose a HASH Distinct
strategy. The optimizer’s decision is ultimately based on cost estimates. The
HASH Distinct is typically less expensive for unsorted input streams can
eliminate rows on the fly for resident partitions. The SORT Distinct operator
cannot eliminate any rows until the entire data set has been sorted.
The showplan output for the HASH Distinct operator reports that Worktable1
will be used. A worktable is needed in case the distinct row result set cannot fit
in memory. In that case, partially processed groups are written to disk.

Vector aggregation operators

There are three unary operators used for vector aggregation. They are the
GROUP SORTED COUNT AGGREGATE, the HASH VECTOR AGGREGATE, and the
GROUP INSERTING operators.
See Table 1-3 on page 23 for a list and description of all query processor
operators.

84 Adaptive Server Enterprise

 CHAPTER 2 Using showplan

GROUP SORTED COUNT AGGREGATE operator

The GROUP SORTED COUNT AGGREGATE nonblocking operator is a variant of
the GROUP SORTED Distinct operator described in “GROUP SORTED
Distinct operator” on page 81. The GROUP SORTED COUNT AGGREGATE
operator requires that input set to be sorted on the group by columns. The
algorithm is very similar to that of GROUP SORTED Distinct.
A row is read from the child operator. If the row is the start of a new vector, its
grouping columns are cached and the aggregation results are initialized.
If the row belongs to the current group being processed, the aggregate
functions are applied to the aggregate results. When the child operator returns
a row that starts a new group or End Of Scan, the current vector and its
aggregated values are returned to the parent operator.
The first row in the GROUP SORTED COUNT AGGREGATE operator is returned
after an entire group is processed, where the first row in the GROUP SORTED
Distinct operator is returned at the start of a new group. This example
collects a list of all cities with the number of authors that live in each city.
select city, total_authors = count(*)
from authors
group by city
plan
"(group_sorted
(sort (scan authors))
)"

QUERY PLAN FOR STATEMENT 1 (at line 1).

Optimized using the Abstract Plan in the PLAN clause.

STEP 1
The type of query is SELECT.

3 operator(s) under root

ROOT:EMIT Operator (VA = 3)

|GROUP SORTED Operator (VA = 2)

| Evaluate Grouped COUNT AGGREGATE.
|
| |SORT Operator (VA = 1)
| | Using Worktable1 for internal storage.
| |
| | | SCAN Operator (VA = 0)
| | | FROM TABLE

Performance and Tuning Series: Query Processing and Abstract Plans 85

Query plan shape

| | | authors
| | | Table Scan.
| | | Forward Scan.
| | | Positioning at start of table.
| | | Using I/O Size 2 Kbytes for data pages.
| | | With LRU Buffer Replacement Strategy for data pages.
In this query plan, the scan of authors does not return rows in grouping order.
A SORT operator is applied to order the stream based on the grouping column
city. At this point, a GROUP SORTED COUNT AGGREGATE operator can be
applied to evaluate the count aggregate.
The GROUP SORTED COUNT AGGREGATE operator showplan output reports the
aggregate functions being applied as:
| Evaluate Grouped COUNT AGGREGATE.

HASH VECTOR AGGREGATE operator

The HASH VECTOR AGGREGATE operator is a blocking operator. All rows from
the child operator must be processed before the first row from the HASH
VECTOR AGGREGATE operator can be returned to its parent operator. Other than
this, the algorithm is similar to the HASH Distinct operator’s algorithm.
Rows are fetched from the child operator. Each row is hashed on the query’s
grouping columns. The bucket that is hashed is searched to see if the vector
already exists.
If the group by values do not exist, the vector is added and the aggregate values
are initialized using this first row. If the group by values do exist, the current
row is aggregated to the existing values. This example collects a list of all cities
with the number of authors that live in each city.
select city, total_authors = count(*)
from authors
group by city

QUERY PLAN FOR STATEMENT 1 (at line 1).

STEP 1
The type of query is SELECT.

2 operator(s) under root

ROOT:EMIT Operator (VA = 2)

|HASH VECTOR AGGREGATE Operator (VA = 1)

| GROUP BY

86 Adaptive Server Enterprise

 CHAPTER 2 Using showplan

| Evaluate Grouped COUNT AGGREGATE.

| Using Worktable1 for internal storage.
| Key Count: 1
|
| |SCAN Operator (VA = 0)
| | FROM TABLE
| | authors
| | Table Scan.
| | Forward Scan.
| | Using I/O Size 2 Kbytes for data pages.
| | With LRU Buffer Replacement Strategy for data pages.
In this query plan, the HASH VECTOR AGGREGATE operator reads all of the
rows from its child operator, which is scanning the authors table. Each row is
checked to see if there is already an entry bucket entry for the current city value.
If there is not, a hash entry row is added with the new city grouping value and
the count result is initialized to 1. If there is already a hash entry for the new
row’s city value, the aggregation function is applied. In this case, the count
result is incriminated.
The showplan output prints a group by message specifically for the HASH
VECTOR AGGREGATE operator, then prints the grouped aggregation messages:
| Evaluate Grouped COUNT AGGREGATE.
The showplan output reports used to store spilled groups and unprocessed
rows:
| Using Worktable1 for internal storage.

GROUP INSERTING
GROUP INSERTING is a blocking operator. All rows from the child operator
must be processed before the first row can be returned from the GROUP
INSERTING.

GROUP INSERTING is limited to 31 or fewer columns in the group by clause.

The operator starts by creating a worktable with a clustered index of the
grouping columns. As each row is fetched from the child, a lookup into the
work table is done based on the grouping columns. If no row is found, then the
row is inserted. This effectively creates a new group and initializes its
aggregate values. If a row is found, the new aggregate values are updated based
on evaluating the new values. The GROUP INSERTING operator returns rows
ordered by the grouping columns.
select city, total_authors = count(*)
from authors
group by city

Performance and Tuning Series: Query Processing and Abstract Plans 87

Query plan shape

plan
'(group_inserting (i_scan auidind authors ))'
QUERY PLAN FOR STATEMENT 1 (at line 1).
Optimized using the Abstract Plan in the PLAN clause.

STEP 1
The type of query is SELECT.

2 operator(s) under root

|ROOT:EMIT Operator (VA = 2)
|
| |GROUP INSERTING Operator (VA = 1)
| | GROUP BY
| | Evaluate Grouped COUNT AGGREGATE
| | Using Worktable1 for internal storage.
| |
| | |SCAN Operator (VA = 0)
| | | FROM TABLE
| | | authors
| | | Table Scan.
| | | Forward Scan.
| | | Positioning at start of table.
| | | Using I/O Size 2 Kbytes for data pages.
| | | With LRU Buffer Replacement Strategy for data pages.
In this example, the group inserting operator starts by building a worktable
with a clustered index keyed on the city column. The group inserting operator
proceeds to drain the authors table. For each row, a lookup is done on the city
value. If there is no row in the aggregation worktable with the current city
value, then the row is inserted. This creates a new group for the current city
value with an initialized count value. If the row for the current city value is
found, then an evaluation is done to increment the COUNT AGGREGATE value.

compute by message
Processing is done in the EMIT operator, and requires that the EMIT operator’s
input stream be sorted according to any order by requirements in the query. The
processing is similar to what is done in the GROUP SORTED AGGREGATE
operator.

88 Adaptive Server Enterprise

 CHAPTER 2 Using showplan

Each row read from the child is checked to see if it starts a new group. If it does
not, aggregate functions are applied as appropriate to the query’s requested
groups. If a new group is started, the current group and its aggregated values
are returned to the user. A new group is then started and its aggregate values
are initialized from the new row’s values. This example collects an ordered list
of all cities and reports a count of the number of entries for each city after the
city list.
select city
from authors
order by city
compute count(city) by city

QUERY PLAN FOR STATEMENT 1 (at line 1).

STEP 1
The type of query is SELECT.

2 operator(s) under root

Emit with Compute semantics

ROOT:EMIT Operator (VA = 2)

|SORT Operator (VA = 1)

| Using Worktable1 for internal storage.
|
| |SCAN Operator (VA = 0)
| | FROM TABLE
| | authors
| | Table Scan.
| | Forward Scan.
| | Positioning at start of table.
| | Using I/O Size 2 Kbytes for data pages.
| | With LRU Buffer Replacement Strategy for data pages.
In this example, the EMIT operator’s input stream is sorted on the city attribute.
For each row, the compute by count value is incremented. When a new city
value is fetched, the current city’s values and associated count value is returned
to the user. The new city value becomes the new compute by grouping value
and its count is initialized to one.

Performance and Tuning Series: Query Processing and Abstract Plans 89

Union operators

Union operators

UNION ALL operator

The UNION ALL operator merges several compatible input streams without
performing any duplicate elimination. Every data row that enters the UNION
ALL operator is included in the operator’s output stream.

The UNION ALL operator is a nary operator that displays this message:
UNION ALL OPERATOR has N children.
N is the number of input streams into the operator.
This example demonstrates the use of UNION ALL:
select * from sysindexes where id < 100
union all
select * from sysindexes where id > 200

QUERY PLAN FOR STATEMENT 1 (at line 1).

STEP 1
The type of query is SELECT.

3 operator(s) under root

|ROOT:EMIT Operator (VA = 3)

|
| |UNION ALL Operator (VA = 2) has 2 children.
| |

| | |SCAN Operator (VA = 0)

| | | FROM TABLE
| | | sysindexes
| | | Using Clustered Index.
| | | Index : csysindexes
| | | Forward Scan.
| | | Positioning by key.
| | | Keys are:
| | | id ASC
| | | Using I/O Size 2 Kbytes for index leaf pages.
| | | With LRU Buffer Replacement Strategy for index leaf pages.
| | | Using I/O Size 2 Kbytes for data pages.
| | | With LRU Buffer Replacement Strategy for data pages.

90 Adaptive Server Enterprise

 CHAPTER 2 Using showplan

| |
| | |SCAN Operator (VA = 1)
| | | FROM TABLE
| | | sysindexes
| | | Using Clustered Index
| | | Index : csysindexes
| | | Forward scan.
| | | Positioning by key.
| | | Keys are:
| | | id ASC
| | | Using I/O Size 2 Kbytes for index leaf pages.
| | | With LRU Buffer Replacement Strategy for index leaf pages.
| | | Using I/O Size 2 Kbytes for data pages.
| | | With LRU Buffer Replacement Strategy for data pages.

The UNION ALL operator starts by fetching all rows from its leftmost child. In
this example, it returns all of the sysindexes rows with an ID less than 100. As
each child operator’s datastream is emptied, the UNION ALL operator moves
on to the child operator immediately to its right. This stream is opened and
emptied. This continues until the last (the Nth) child operator is emptied.

MERGE UNION operator

The MERGE UNION operator performs a UNION ALL operation on several
sorted compatible data streams and eliminates duplicates within these streams.
The MERGE UNION operator is a nary operator that displays this message:
MERGE UNION OPERATOR has <N> children.
<N> is the number of input streams into the operator.

HASH UNION
The HASH UNION operator uses Adaptive Server hashing algorithms to
simultaneously perform a UNION ALL operation on several data streams and
hash-based duplicate elimination.
The HASH UNION operator is a nary operator that displays this message:
HASH UNION OPERATOR has <N> children.
<N> is the number of input streams into the operator.

Performance and Tuning Series: Query Processing and Abstract Plans 91

Union operators

HASH UNION also displays the name of the worktable it uses, in this format:
HASH UNION OPERATOR Using Worktable <X> for internal
storage.
This worktable is used by the HASH UNION operator to temporarily store data
for the current iteration that cannot be processed in the memory currently
available.
This example demonstrates the use of HASH UNION:
select * from sysindexes
union
select * from sysindexes

QUERY PLAN FOR STATEMENT 1 (at line 1).

STEP 1
The type of query is SELECT.

3 operator(s) under root

| ROOT:EMIT Operator (VA = 3)

|
| |HASH UNION Operator has 2 children.
| | Using Worktable1 for internal storage.
| |
| | |SCAN Operator
| | | FROM TABLE
| | | sysindexes
| | | Table Scan.
| | | Forward Scan.
| | | Positioning at start of table.
| | | Using I/O Size 2 Kbytes for data pages.
| | | With LRU Buffer Replacement Strategy for data pages.
| |
| |SCAN Operator (VA = 1)
| | | FROM TABLE
| | | sysindexes
| | | Table Scan.
| | | Forward Scan.
| | | Positioning at start of table.
| | | Using I/O size 2 Kbytes for data pages.
| | | With LRU Buffer Replacement Strategy for data pages.

92 Adaptive Server Enterprise

 CHAPTER 2 Using showplan

SCALAR AGGREGATE operator

The SCALAR AGGREGATE operator keeps track of running information about
an input data stream, such as the number of rows in the stream, or the maximum
value of a given column in the stream.
The SCALAR AGGREGATE operator prints a list of up to 10 messages describing
the scalar aggregation operations it executes. The message has the following
format:
Evaluate Ungrouped <Type of Aggregate> Aggregate
<Type of Aggregate> can be any of the following: count, sum, average, min,
max, any, once-unique, count-unique, sum-unique, average-unique, or once.

The following query performs a SCALAR AGGREGATE (that is, unwrapped)

aggregation on the authors table in the pubs2 database:
select count(*) from authors

QUERY PLAN FOR STATEMENT 1 (at line 1).

STEP 1
The type of query is SELECT.

2 operator(s) under root

ROOT:EMIT Operator (VA = 2)

|SCALAR AGGREGATE Operator (VA = 1)

| Evaluate Ungrouped COUNT AGGREGATE.
|
| |SCAN Operator (VA =0)
| | FROM TABLE
| | authors
| | Index : aunmind
| | Forward Scan.
| | Positioning at index start.
| | Index contains all needed columns. Base table will not be read.
| | Using I/O Size 4 Kbytes for index leaf pages.
| | With LRU Buffer Replacement Strategy for index leaf pages.

The SCALAR AGGREGATE message indicates that the query to be executed is an

ungrouped count aggregation.

Performance and Tuning Series: Query Processing and Abstract Plans 93

Union operators

RESTRICT operator
The RESTRICT operator is a unary operator that evaluates expressions based
on column values. The RESTRICT operator is associated with multiple column
evaluations lists that can be processed before fetching a row from the child
operator, after fetching a row from the child operator, or to compute the value
of virtual columns after fetching a row from the child operator.

SORT operator
The SORT operator has only one child operator within the query plan. Its role
is to generate an output data stream from the input stream, using a specified
sorting key.
The SORT operator may execute a streaming sort when possible, but may also
have to store results temporarily into a worktable. The SORT operator displays
the worktable’s name in this format:
Using Worktable<N> for internal storage.
where <N> is a numeric identifier for the worktable within the showplan output.
Here is an example of a simple query plan using a SORT operator and a
worktable:
select au_id from authors order by postalcode

QUERY PLAN FOR STATEMENT 1 (at line 1).

STEP 1
The type of query is SELECT.

2 operator(s) under root

ROOT:EMIT Operator (VA = 2)

|SORT Operator (VA = 1)

| Using Worktable1 for internal storage.
|
| |SCAN Operator (VA = 0)
| | FROM TABLE
| | authors
| | Table Scan.
| | Forward Scan.
| | Positioning at start of table.
| | Using I/O Size 4 Kbytes for data pages.

94 Adaptive Server Enterprise

 CHAPTER 2 Using showplan

| | With LRU Buffer Replacement Strategy for data pages.

The SORT operator drains its child operator and sorts the rows. In this case, it
sorts each row fetched from the authors table using the postalcode attribute. If
all of the rows fit into memory, then no data is spilled to disk. But, if the input
data’s size exceeds the available buffer space, then sorted runs are spilled to
disk. These runs are recursively merged into larger sorted runs until there are
fewer runs than there are available buffers to read and merge the runs with.

STORE operator
The STORE operator is used to create a worktable, fill it, and possibly create an
index on it. As part of the execution of a query plan, the worktable is used by
other operators in the plan. A SEQUENCER operator guarantees that the plan
fragment corresponding to the worktable and potential index creation is
executed before other plan fragments that use the worktable. This is important
when a plan is executed in parallel, because execution processes operate
asynchronously.
Reformatting strategies use the STORE operator to create a worktable with a
clustered index on it.
If the STORE operator is used for a reformatting operation, it prints this
message:
Worktable <X> created, in <L> locking mode for
reformatting.
The locking mode <L> has to be one of “allpages,” “datapages,” or “datarows.”
The STORE operator also prints this message:
Creating clustered index.
If the STORE operator is not used for a reformatting operation, it prints this
message:
Worktable <X> created, in <L> locking mode.
The following example applies to the STORE operator, as well as to the
SEQUENCER operator.
select * from bigun a, bigun b where a.c4 = b.c4 and a.c2 < 10

QUERY PLAN FOR STATEMENT 1 (at line 1).

Optimized using the Abstract Plan in the PLAN clause.

STEP 1

Performance and Tuning Series: Query Processing and Abstract Plans 95

Union operators

The type of query is SELECT.

7 operator(s) under root

|ROOT:EMIT Operator (VA = 7)

|
| |SEQUENCER Operator (VA = 6) has 2 children.
| |
| | |STORE Operator (VA = 5)
| | | Worktable1 created, in allpages locking mode, for REFORMATTING.
| | | Creating clustered index.
| | |
| | | |INSERT Operator(VA = 4)
| | | | The update mode is direct.
| | | |
| | | | |SCAN Operator(VA = 0)
| | | | | FROM TABLE
| | | | | bigun
| | | | | b
| | | | | Table Scan.
| | | | | Forward Scan.
| | | | | Positioning at start of table.
| | | | | Using I/O Size 2 Kbytes for data pages.
| | | | | With LRU Buffer Replacement Strategy for data pages.
| | | |
| | | | TO TABLE (VA = 3)
| | | | Worktable1.
| |
| | |NESTED LOOP JOIN (Join Type: Inner Join)(VA = 7)
| | |
| | | |SCAN Operator (VA = 2)
| | | | FROM TABLE
| | | | bigun
| | | | a
| | | | Table Scan.
| | | | Forward Scan.
| | | | Positioning at start of table.
| | | | Using I/O Size 2 Kbytes for data pages.
| | | | With LRU Buffer Replacement Strategy for data pages.
| | |
| | | |SCAN Operator (VA = 1)
| | | | FROM TABLE
| | | | Worktable1.
| | | | Using Clustered Index.
| | | | Forward Scan.
| | | | Positioning key.

96 Adaptive Server Enterprise

 CHAPTER 2 Using showplan

| | | | Using I/O Size 2 Kbytes for data pages.

| | | | With LRU Buffer Replacement Strategy for data pages.
In the example plan shown above, the STORE operator is used in a reformatting
strategy. It is located directly below the SEQUENCER operator in the leftmost
child of the SEQUENCER operator.
The STORE operator creates Worktable1, which is filled by the INSERT operator
below it. The STORE operator then creates a clustered index on Worktable1. The
index is built on the join key b.c4.

SEQUENCER operator
The SEQUENCER operator is a nary operator used to sequentially execute each
the child plans below it. The SEQUENCER operator is used in reformatting plans,
and certain aggregate processing plans.
The SEQUENCER operator executes each of its child subplans, except for the
rightmost one. Once all the left child subplans are executed, the rightmost
subplan is executed.
The SEQUENCER operator displays this message:
SEQUENCER operator has N children.
select * from tab1 a, tab2 b where a.c4 = b.c4 and a.c2 < 10
QUERY PLAN FOR STATEMENT 1 (at line 1).
Optimized using the Abstract Plan in the PLAN clause.

STEP 1
The type of query is SELECT.

7 operator(s) under root

|ROOT:EMIT Operator (VA = 7)

|
| |SEQUENCER Operator (VA = 6) has 2 children.
| |
| | |STORE Operator (VA = 5)
| | | Worktable1 created, in allpages locking mode, for REFORMATTING.
| | | Creating clustered index.
| | |
| | | |INSERT Operator (VA = 4)
| | | | The update mode is direct.
| | | |
| | | | |SCAN Operator (VA = 0)

Performance and Tuning Series: Query Processing and Abstract Plans 97

Union operators

| | | | | FROM TABLE
| | | | | tab2
| | | | | b
| | | | | Table Scan.
| | | | | Forward Scan.
| | | | | Positioning at start of table.
| | | | | Using I/O Size 2 Kbytes for data pages.
| | | | | With LRU Buffer Replacement Strategy for data pages.
| | | |
| | | | TO TABLE
| | | | Worktable1.
| |
| | |NESTED LOOP JOIN Operator (Join Type: Inner Join) (VA = 3)
| | |
| | | |SCAN Operator (VA = 2)
| | | | FROM TABLE
| | | | tab1
| | | | a
| | | | Table Scan.
| | | | Forward Scan.
| | | | Positioning at start of table.
| | | | Using I/O Size 2 Kbytes for data pages.
| | | | With LRU Buffer Replacement Strategy for data pages.
| | |
| | | |SCAN Operator (VA = 1)
| | | | FROM TABLE
| | | | Worktable1.
| | | | Using Clustered Index.
| | | | Forward Scan.
| | | | Positioning by key.
| | | | Using I/O Size 2 Kbytes for data pages.
| | | | With LRU Buffer Replacement Strategy for data pages.
In this example, the SEQUENCER operator implements a reformatting strategy.
The leftmost branch of the SEQUENCER operator creates a clustered index on
Worktable1. This branch is executed and closed before the SEQUENCER
operator proceeds to the next child operator. The SEQUENCER operator arrives
at the rightmost child, opens, and begins to drain it, returning rows back to its
parent operator. The design intent of the SEQUENCER operator is for operators
in the rightmost branch to use the worktables created in the preceding outer
branches of the SEQUENCER operator. In this example, Worktable1 is used in a
nested-loop join strategy. The scan of Worktable1 is positioned by a key on its
clustered index for each row that comes from the outer scan of tab1.

98 Adaptive Server Enterprise

 CHAPTER 2 Using showplan

REMOTE SCAN operator

The REMOTE SCAN operator sends a SQL query to a remote server for
execution. It then processes the results returned by the remote server, if any.
REMOTE SCAN displays the formatted text of the SQL query it handles.

REMOTE SCAN has 0 or 1 child operators.

SCROLL operator
The SCROLL operator encapsulates the functionality of scrollable cursors in
Adaptive Server. Scrollable cursors may be insensitive, meaning that they
display a snapshot of their associated data, taken when the cursor is opened, or
semi-sensitive, meaning that the next rows to be fetched are retrieved from the
live data.
The SCROLL operator is a unary operator that displays this message:
SCROLL OPERATOR ( Sensitive Type: <T>)
The type may be insensitive or semi-sensitive.
This is an example of a plan featuring an insensitive scrollable cursor:
declare CI insensitive scroll cursor for
select au_lname, au_id from authors
go
set showplan on
go
open CI

QUERY PLAN FOR STATEMENT 1 (at line 1).

STEP 1
The type of query is OPEN CURSOR CI.

QUERY PLAN FOR STATEMENT 1 (at line 2).

STEP 1
The type of query is DECLARE CURSOR.

2 operator(s) under root

ROOT:EMIT Operator (VA = 2)

|SCROLL Operator (Sensitive Type: Insensitive) (VA = 1)

Performance and Tuning Series: Query Processing and Abstract Plans 99

Union operators

| Using Worktable1 for internal storage.

|
| |SCAN Operator (VA = 0)
| | FROM TABLE
| | authors
| | Table Scan.
| | Forward Scan.
| | Positioning at start of table.
| | Using I/O Size 4 Kbytes for data pages.
| | With LRU Buffer Replacement Strategy for data pages.
The SCROLL operator is the child operator of the root EMIT operator, and its
only child is the SCAN operator on the authors table. SCROLL message specifies
that the CI cursor is insensitive.
Scrollable cursor rows are initially cached in memory. Worktable1 is used as a
backing store for this cache when the amount of data processed exceeds the
cache’s physical memory limits.

RID JOIN operator

The RID JOIN operator is a binary operator that joins two data streams, based
on row IDs generated for the same source table. Each data row in a SQL table
is associated with a unique row ID (RID). Think of a rid-join as a special case
of a self-join query. The left child fills a worktable with the set of uniquely
qualifying RIDs. The RIDs are the result of applying a distinct filter to the
RIDs returned from two or more disparate index cases of the same source table.
The RID JOIN operator is used to implement the general or strategy. The
general-or strategy is often used when a query’s predicate contains a collection
of disjunctions that can be qualified by different indexes on the same table. In
this case, each index is scanned based on the predicates that can be qualified by
that index. For each index row that qualifies, a RID is returned.
The returned RIDs are processed for uniqueness so that the same row is not
returned twice, which might happen if two or more of the disjunctions qualify
the same row.
The RID JOIN operator inserts the unique RIDs into a worktable. The
worktable of unique RIDs is passed to the scan operator in the rid-join’s right
branch. The access methods can iteratively fetch the next RID to be processed
directly from the worktable, and look up the associated row. This row is then
returned to the RID JOIN parent operator.
The RID JOIN operator displays this message:

100 Adaptive Server Enterprise

 CHAPTER 2 Using showplan

Using Worktable <N> for internal storage.

This worktable is used to store the unique RIDs generated from the left child.
The following example demonstrates the showplan output for the RID JOIN
operator.
select * from tab1 a where a.c1 = 10 or a.c3 = 10

QUERY PLAN FOR STATEMENT 1 (at line 2).

STEP 1
The type of query is SELECT.

6 operator(s) under root.

|ROOT:EMIT Operator (VA = 6)

|
| |RID JOIN Operator (VA = 5)
| | Using Worktable2 for internal storage.
| |
| | |HASH UNION Operator (VA = 6) has 2 children.
| | | Key Count: 1
| | |
| | | |SCAN Operator (VA = 0)
| | | | FROM TABLE
| | | | tab1
| | | | a
| | | | Index:tab1idx
| | | | Forward Scan.
| | | | Positioning by key.
| | | | Index contains all needed columns. Base table will not be read.
| | | | Keys are:
| | | | c1 ASC
| | | | Using I/O Size 2 Kbytes for index leaf pages.
| | | | With LRU Buffer Replacement Strategy for index leaf pages.
| | |
| | | |SCAN Operator (VA = 4)
| | | | FROM TABLE
| | | | tab1
| | | | a
| | | | Index:tab1idx2
| | | | Forward Scan.
| | | | Positioning by key.
| | | | Index contains all needed columns. Base table will not be read.
| | | | Keys are:
| | | | c3 ASC

Performance and Tuning Series: Query Processing and Abstract Plans 101
Union operators

| | | | Using I/O Size 2 Kbytes for index leaf pages.

| | | | With LRU Buffer Replacement Strategy for index leaf pages.
| |
| | |RESTRICT Operator (VA = 3)
| | |
| | | |SCAN Operator (VA = 2)
| | | | FROM TABLE
| | | | tab1
| | | | a
| | | | Using Dynamic Index.
| | | | Forward Scan.
| | | | Positioning by Row IDentifier (RID).
| | | | Using I/O Size 2 Kbytes for data pages.
| | | | With LRU Buffer Replacement Strategy for data pages.
In this example, the index tab1idx is scanned to get all RIDs from tab1 that have
a c1 value of 10. Adaptive Server scans tab1idx2 to get all RIDs from tab1 that
have a c3 value of 10.
The HASH UNION operator is used to eliminate duplicate RIDs. There are
duplicate RIDs for any tab1 rows where both c1 and c3 rows have a value of 10.
The RID JOIN operator inserts all of the returned rows into Worktable2.
Worktable2 is passed to the scan of tab1 after it has been completely filled. The
access methods fetch the first RID, look up the associated row, and return it to
the RID JOIN operator. On subsequent calls to the tab1’s scan operator, the
access methods fetch the next RID to be processed and return its associated
row.

SQLFILTER operator
The SQLFILTER operator is a nary operator that executes subqueries. Its
leftmost child represents the outer query, and the other children represent query
plan fragments associated with one or more subqueries.
The leftmost child generates correlation values that are substituted into the
other child plans.
The SQLFILTER operator displays this message:
SQFILTER Operator has <N> children.
This example illustrates the use of SQLFILTER:
select pub_name from publishers
where pub_id =
(select distinct titles.pub_id from titles

102 Adaptive Server Enterprise

 CHAPTER 2 Using showplan

where publishers.pub_id = titles.pub_id

and price > $1000)
QUERY PLAN FOR STATEMENT 1 (at line 1).
4 operator(s) under root

STEP 1
The type of query is SELECT.

4 operator(s) under root

ROOT:EMIT Operator (VA = 4)

|SQFILTER Operator (VA = 3) has 2 children.

|
| |SCAN Operator (VA = 0)
| | FROM TABLE
| | publishers
| | Table Scan.
| | Forward Scan.
| | Positioning at start of table.
| | Using I/O Size 8 Kbytes for data pages.
| | With LRU Buffer Replacement Strategy for data pages.
|
| Run subquery 1 (at nesting level 1)
|
| QUERY PLAN FOR SUBQUERY 1 (at nesting level 1 and at line 3)
|
| Correlated Subquery
| Subquery under an EXPRESSION predicate.
|
| |SCALAR AGGREGATE Operator (VA = 2)
| | Evaluate Ungrouped ONCE-UNIQUE AGGREGATE
| |
| | |SCAN Operator (VA = 1)
| | | FROM TABLE
| | | titles
| | | Table Scan.
| | | Forward Scan.
| | | Postitioning at start of table.
| | | Using I/O Size 8 Kbytes for data pages.
| | | With LRU Buffer Replacement Strategy for data pages.
|
| END OF QUERY PLAN FOR SUBQUERY 1

Performance and Tuning Series: Query Processing and Abstract Plans 103
Union operators

The SQLFILTER operator has two children in this example. The leftmost child
is the query’s outer block. It is a simple scan of the publishers table. The right
child is used to evaluate the query’s subquery. SQLFILTER fetch rows from the
outer block. For every row from the outer block, SQLFILTER invokes the right
child to evaluate the subquery. If the subquery evaluates to TRUE, a row is
returned to the SQLFILTER’s parent operator.

EXCHANGE operator
The EXCHANGE operator is a unary operator that encapsulates parallel
processing of Adaptive Server SQL queries. EXCHANGE can be located almost
anywhere in a query plan and divides the query plan into plan fragments. A
plan fragment is a query plan tree that is rooted at an EMIT or EXCHANGE:EMIT
operator and has leaves that are SCAN or EXCHANGE operators. A serial plan is
a plan fragment that is executed by a single process.
An EXCHANGE operator’s child operator is always an EXCHANGE:EMIT
operator. EXCHANGE:EMIT is the root of a new plan fragment. An EXCHANGE
operator has an associated server process called the Beta process that acts as a
local execution coordinator for the EXCHANGE operator’s worker processes.
Worker processes execute the plan fragment as directed by the parent
EXCHANGE operator and its Beta process. The plan fragment is often executed
in a parallel fashion, using two or more processes. The EXCHANGE operator and
Beta process coordinate activities, including the exchange of data between the
fragment boundaries.
The topmost plan fragment, rooted at an EMIT operator rather than an
EXCHANGE:EMIT operator, is executed by the Alpha process. The Alpha
process is a consumer process associated with the user connection. The Alpha
process is the global coordinator of all of the query plan’s worker processes. It
is responsible for initially setting up all of the plan fragment’s worker processes
and eventually freeing them. It manages and coordinates all of the fragment’s
worker processes in the case of an exception.
The EXCHANGE operator displays this message:
Executed in parallel by N producer and P consumer processes.

104 Adaptive Server Enterprise

 CHAPTER 2 Using showplan

The number of producers refers to the number of worker processes that execute
the plan fragment located beneath the EXCHANGE operator. The number of
consumers refers to the number of worker processes that execute the plan
fragment that contains the EXCHANGE operator. The consumers process the data
passed to them by the producers. Data is exchanged between the producer and
consumer processes through a pipe set up in the EXCHANGE operator. The
producer’s EXCHANGE:EMIT operator writes rows into the pipe while
consumers read rows from this pipe. The pipe mechanism synchronizes
producer writes and consumer reads such that no data is lost.
This example illustrates a parallel query in the master database against the
system table sysmessages:
use master
go
set showplan on
go
select count(*) from sysmessages t1 plan '(t_scan t1) (prop t1 (parallel 4))

QUERY PLAN FOR STATEMENT 1 (at line 1).

Optimized using the forced options (internally generated Abstract Plan).
Executed in parallel by coordinating process and 4 worker processes.

4 operator(s) under root

The type of query is SELECT.

ROOT:EMIT Operator
|SCALAR AGGREGATE Operator
| Evaluate Ungrouped COUNT AGGREGATE.
|
| |EXCHANGE Operator
| |Executed in parallel by 4 Producer and 1 Consumer processes.

| |
| | |EXCHANGE:EMIT Operator
| | |
| | | |SCAN Operator
| | | | FROM TABLE
| | | | sysmessages
| | | | Table Scan.
| | | | Forward Scan.
| | | | Positioning at start of table.
| | | | Executed in parallel with a 4-way hash scan.
| | | | Using I/O Size 4 Kbytes for data pages.
| | | | With LRU Buffer Replacement Strategy for data pages.

Performance and Tuning Series: Query Processing and Abstract Plans 105
INSTEAD-OF TRIGGER operators

There are two plan fragments in this example. The first fragment in any plan,
parallel or not, is always rooted by an EMIT operator. The first fragment in this
example consists of the EMIT, SCALAR AGGREGATE, and EXCHANGE operators.
This first fragment is always executed by the single Alpha process. In this
example, it also acts as the Beta process responsible for managing the
EXCHANGE operator’s worker processes.
The second plan fragment is rooted at the EXCHANGE:EMIT operator. Its only
child operator is the SCAN operator. The SCAN operator is responsible for
scanning the sysmessages table. The scan is executed in parallel:
Executed in parallel with a 4-way hash scan
This indicates that each worker process is responsible for approximately a
quarter of the table. Pages are assigned to the worker processes based on
having the data page ID.
The EXCHANGE:EMIT operator writes data rows to the consumers by writing to
a pipe created by its parent EXCHANGE operator. In this example, the pipe is a
four-to-one demultiplexer, and include several pipe types that perform quite
different behaviors.

INSTEAD-OF TRIGGER operators

There are two operators associated with the instead-of triggers feature:
INSTEAD-OF TRIGGER and CURSOR SCAN. The instead-of trigger feature is
available as of Adaptive Server version 15.0.2. The instead-of trigger feature
uses pseudotables, which allow the user to apply specific actions for inserts,
deletes, and updates on views, when these actions would otherwise have been
ambiguous.

106 Adaptive Server Enterprise

 CHAPTER 2 Using showplan

INSTEAD-OF TRIGGER operator

The INSTEAD-OF TRIGGER operator appears only in query plans for insert,
update, or delete statements on a view that has an instead-of trigger created
upon it. Its function is to create and fill the inserted and deleted pseudotables
that are used in the trigger to examine the rows that would have been modified
by the original insert, update, or delete query. The only purpose of the query
plan that contains an INSTEAD-OF TRIGGER operator is to fill the inserted and
deleted tables—the actual operation of the original SQL statement is never
attempted on the view referenced in the statement. Rather, it is up to the trigger
to perform the updates to the view’s underlying tables based on the data
available in the inserted and deleted pseudo tables.
The following is an example of the INSTEAD-OF TRIGGER operator’s
showplan output:
create table t12 (c0 int primary key, c1 int null, c2 int null)
go
. . .
create view t12view as select c1,c2 from t12
go
create trigger v12updtrg on t12view
instead of update as
select * from deleted
go
update t12view set c1 = 3

QUERY PLAN FOR STATEMENT 1 (at line 1).

STEP 1
The type of query is SELECT.

2 operator(s) under root

|ROOT:EMIT Operator (VA = 1)

|
| |INSTEAD-OF TRIGGER Operator
| | Using Worktable1 for internal storage.
| | Using Worktable2 for internal storage.
| |
| | |SCAN Operator (VA = 0)
| | | FROM TABLE
| | | t12
| | | Table Scan.
| | | Forward Scan.
| | | Positioning at start of table.

Performance and Tuning Series: Query Processing and Abstract Plans 107
INSTEAD-OF TRIGGER operators

| | | Using I/O Size 2 Kbytes for data pages.

| | | With LRU Buffer Replacement Strategy for data pages.
In this example, the v12updtrig instead-of trigger is defined on the t12view. The
update to the t12view results in the creation of the INSTEAD-OF TRIGGER
operator. The INSTEAD-OF TRIGGER operator creates two worktables.
Worktable1 and Worktable2 are used to hold the inserted and deleted rows,
respectively. These worktables are unique in that they persist across
statements. Trigger execution results in the following showplan lines getting
printed.
QUERY PLAN FOR STATEMENT 1 (at line 3).

STEP 1
The type of query is SELECT.

1 operator(s) under root

|ROOT:EMIT Operator (VA = 1)

|
| |SCAN Operator (VA = 0)
| | FROM CACHE
The showplan statement output above is for the trigger’s statement, select *
from deleted. The rows to be deleted from the view were inserted into the
“deleted” cache when the initial update statement was executed. Then, the
trigger scans the table to report what rows would have been deleted from the
t12view view.

CURSOR SCAN operator

The CURSOR SCAN operator only appears in positioned delete or update (that
is, delete view-name where current of cursor_name) statements on a view that
has an instead-of trigger created upon it. As such, it appears only as a child
operator of the INSTEAD-OF TRIGGER operator. A positioned delete or update
accesses only the row on which the cursor is currently positioned. The CURSOR
SCAN operator reads the current row of the cursor directly from the EMIT
operator of the query plan for the fetch cursor statement. These values are
passed to the INSTEAD-OF TRIGGER operator to be inserted into the inserted
or deleted pseudo tables (this example uses the same table as the previous
example).
declare curs1 cursor for select * from t12view
go

108 Adaptive Server Enterprise

 CHAPTER 2 Using showplan

open curs1
go
fetch curs1

c1 c2
_________ ________
1 2

(1 row affected)
set showplan on
go
update t12view set c1 = 3
where current of curs1

QUERY PLAN FOR STATEMENT (at line 1).

STEP 1
The type of query is SELECT.

2 operator(s) under root

|ROOT:EMIT Operator (VA = 2)

|
| |INSTEAD-OF TRIGGER Operator (VA = 1)
| | Using Worktable1 for internal storage.
| | Using Worktable2 for internal storage.
| |
| | |CURSOR SCAN Operator (VA = 0)
| | | FROM EMIT OPERATOR
The showplan output in this example is identical to that from the previous
INSTEAD-OF TRIGGER operator example, with one exception. A CURSOR
SCAN operator appears as the child operator of the INSTEAD-OF TRIGGER
operator rather than a scan of the view’s underlying tables.
The CURSOR SCAN gets the values to be inserted into the pseudo tables by
accessing the result of the cursor fetch. This is conveyed by the FROM EMIT
OPERATOR message.
QUERY PLAN FOR STATEMENT 1 (at line 3).

STEP 1
The type of query is SELECT.

1 operator(s) under root

|ROOT:EMIT Operator (VA = 1)

Performance and Tuning Series: Query Processing and Abstract Plans 109
INSTEAD-OF TRIGGER operators

|
| |SCAN Operator (VA = 0)
| | FROM CACHE
The showplan statement above is for the trigger’s statement. It is identical to
the output in the INSTEAD-OF TRIGGER example.

deferred_index and deferred_varcol messages

The update mode is deferred_varcol.

The update mode is deferred_index.

These showplan messages indicate that Adaptive Server may process an update
command as a deferred index update.
Adaptive Server uses deferred_varcol mode when updating one or more
variable-length columns. This update may be done in deferred or direct mode,
depending on information that is available only at runtime.
Adaptive Server uses deferred_index mode when the index is unique or
may change as part of the update. In this mode, Adaptive Server deletes the
index entries in direct mode but inserts them in deferred mode.

110 Adaptive Server Enterprise

CH A PTE R 3 Displaying Query Optimization
Strategies and Estimates

This chapter describes the messages printed by the query optimization

options of the set command.
Topic Page
set commands for text format messages 111
set commands for XML format messages 112
Diagnostic usage scenarios 119
Permissions for set commands 122

set commands for text format messages

Either the query optimizer or the query execution layer can generate
diagnostic output. To generate diagnostic output in text format, use:
set option
{ {show | show_lop | show_managers | show_log_props |
show_parallel | show_histograms | show_abstract_plan |
show_search_engine | show_counters | show_best_plan |
show_code_gen | show_pio_costing | show_lio_costing |
show_pll_costing | show_elimination | show_missing_stats}
{normal | brief | long | on | off} }...

Note Each option specified must be followed by one of normal, brief, long,
on, or off. on and normal are equivalent. Each show option must include
one of these choices (normal, brief, and so on); specify more than one
option in a single set option command by separating each option or choice
pair with commas.

See “Diagnostic usage scenarios” on page 119 for examples of using the
set options.

Performance and Tuning Series: Query Processing and Abstract Plans 111
set commands for XML format messages

Table 3-1: Optimizer set commands for text format messages

Option Definition
show Shows a reasonable collection of details, where the collection depends on the choice of
{normal | brief | long | on | off}
show_lop Shows the logical operators used
show_managers Shows the data structure managers used during optimization
show_log_props Shows the logical properties evaluated
show_parallel Shows details of parallel query optimization
show_histograms Shows the processing of histograms associated with SARG/join columns
show_abstract_plan Shows the details of an abstract plan
show_search_engine Shows the details of the join-ordering algorithm
show_counters Shows the optimization counters
show_best_plan Shows the details of the best query plan selected by the optimizer
show_code_gen Shows details of code generation
show_pio_costing Shows estimates of physical input/output (reads/writes from/to the disk)
show_lio_costing Shows estimates of logical input/output (reads/writes from/to memory)
show_pll_costing Shows estimates relating to costing for parallel execution
show_elimination Shows partition elimination
show_missing_stats Shows details of useful statistics missing from SARG/join columns

set commands for XML format messages

You can regenerate diagnostics as an XML document. This makes it easier for
front-end tools to interpret a document. You can use the native XPath query
processor inside Adaptive Server to query this output if the XML option is
enabled.
Either the query optimizer or the query execution layer can generate
diagnostics output. To generate an XML document for the diagnostic output,
use this set plan command:
set plan for
{show_exec_xml, show_opt_xml, show_execio_xml,
show_lop_xml, show_managers_xml, show_log_props_xml,
show_parallel_xml, show_histograms_xml, show_final_plan_xml,
show_abstract_plan_xml, show_search_engine_xml,
show_counters_xml, show_best_plan_xml, show_pio_costing_xml,
show_lio_costing_xml, show_elimination_xml}
to {client | message} on

112 Adaptive Server Enterprise

 CHAPTER 3 Displaying Query Optimization Strategies and Estimates

Option Definition
show_exec_xml Gets the compiled plan output in XML, showing each of the query plan operators.
show_opt_xml Gets optimizer diagnostic output, which shows the different components such as
logical operators, output from the managers, some of the search engine diagnostics,
and the best query plan.
show_execio_xml Gets the plan output along with estimated and actual I/Os. show_execio_xml also
includes the query text.
show_lop_xml Gets the output logical operator tree in XML.
show_managers_xml Shows the output of the different component managers during the preparation phase
of the query optimizer.
show_log_props_xml Shows the logical properties for a given equivalence class (one or more groups of
relations in the query).
show_parallel_xml Shows the diagnostics related to the optimizer while generating parallel query plans.
show_histograms_xml Shows diagnostics related to histograms and the merging of histograms.
show_final_plan_xml Gets the plan output. Does not include the estimated and actual I/Os.
show_final_plan_xml includes the query text.
show_abstract_plan_xml Shows the generated abstract plan.
show_search_engine_xml Shows diagnostics related to the search engine.
show_counters_xml Shows plan object construction/destruction counters.
show_best_plan_xml Shows the best plan in XML.
show_pio_costing_xml Shows actual physical input/output costing in XML.
show_lio_costing_xml Shows actual logical input/output costing in XML.
show_elimination_xml Shows partition elimination in XML.
client When specified, output is sent to the client. By default, this is the error log. When trace
flag 3604 is active, however, output is sent to the client connection.
message When specified, output is sent to an internal message buffer.

To turn an option off, specify:

set plan for
{show_exec_xml, show_opt_xml, show_execio_xml, show_lop_xml,
show_managers_xml, show_log_props_xml, show_parallel_xml,
show_histograms_xml,show_final_plan_xml
show_abstract_plan_xml, show_search_engine_xml,
show_counters_xml, show_best_plan_xml, show_pio_costing_xml,
show_lio_costing_xml, show_elimination_xml} off
You need not specify the destination stream when turning the option off.
When message is specified, the client application must get the diagnostics
from the buffer using a built-in function called showplan_in_xml(query_num).

Performance and Tuning Series: Query Processing and Abstract Plans 113
set commands for XML format messages

query_num refers to the number of queries that are cached in the buffer.
Currently, a maximum of 20 queries are cached in the buffer. The cache stops
collecting query plans when it reaches 20 queries; it ignores the rest of the
query plans. However, the message buffer continues to collect query plans.
After 20 queries, you can display the message buffer only in its entirety by
using a value of 0.
Valid values for query_num are 1 – 20, -1, and 0 (zero). A value of -1 refers to
the last XML doc in the cache; a value of 0 refers to the entire message buffer.
The message buffer may overflow. If this occurs, there is no way to log all of
the XML document, which may result in a partial and invalid XML document.
When the message buffer is accessed using showplan_in_xml, the buffer is
emptied after execution.
You may want to use set textsize to set the maximum text size, as the XML
document is printed as a text column and the document is truncated if the
column is not large enough. For example, set the text size to 100000 bytes
using:
set textsize 100000
When set plan is issued with off, all XML tracing is turned off if all of the trace
options have been turned off. Otherwise, only specified options are turned off.
Other options previously turned on are still valid and tracing continues on the
specified destination stream. When you issue another set plan option, the
previous options are joined with the current options, but the destination stream
is switched unconditionally to a new one.

Using show_execio_xml to diagnose query plans

show_execio_xml includes diagnostic information that you may find can be
helpful for investigating problematic queries. Information from
show_execio_xml includes:

• The version level of the query plan. Each version of the plan is uniquely
identified. This is the first version of the plan:
<planVersion>1.0</planVersion>
• The statement number in a batch or stored procedure, along with the line
number of the statement in the original text. This is statement number 2,
but line number 6, in the query:
<statementNum>2</statementNum>

114 Adaptive Server Enterprise

 CHAPTER 3 Displaying Query Optimization Strategies and Estimates

<lineNum>6</lineNum>
• The abstract plan for the query. For example, this is the abstract plan for
the query select * from titles:
<abstractPlan>
<![CDATA[>
( i_scan titleidind titles ) ( prop titles ( parallel 1
) ( prefetch 8 ) ( lru ) )
]]>
</abstractPlan>
• The logical I/O, physical I/O, and CPU costs:
<costs>
<lio> 2 </lio>
<pio> 2 </pio>
<cpu> 18 </cpu>
</costs>
You can estimate the total costs with this formula (the 25, 2, and 0.1 are
constants):
25 X pio + 2 X lio + 0.1 X cpu
• The estimated execution resource usage, including the number of threads
and auxiliary scan descriptors used by the query plan.
• The number of plans the query engine viewed and the plans it determined
were valid, the total time the query spent in the query engine (in
milliseconds), the time the query engine took to determine the first legal
plan, and the amount of procedure cache used during the optimization
process.
<optimizerMetrics>
<optTimeMs>6</optTimeMs>
<optTimeToFirstPlanMs>3</optTimeToFirstPlanMs>
<plansEvaluated>1</plansEvaluated>
<plansValid>1</plansValid>
<procCacheBytes>140231</procCacheBytes>
</optimizerMetrics>
• The last time update statistics was run on the current table and whether the
query engine used an estimation constant for a given column that it could
have estimated better if statistics were available. This section includes
information about columns with missing statistics:
<optimizerStatistics>
<statInfo>
<objName>titles</objName>

Performance and Tuning Series: Query Processing and Abstract Plans 115
set commands for XML format messages

<columnStats>
<column>title_id</column>
<updateTime>Oct 5 2006 4:40:14:730PM</updateTime>
</columnStats>
<columnStats>
<column>title</column>
<updateTime>Oct 5 2006 4:40:14:730PM</updateTime>
</columnStats>
</statInfo>
</optimizerStatistics>
• An operator tree that includes table and index scans with information
about cache strategies and I/O sizes (inserts, updates, and deletes have the
same information for the target table). The operator tree also shows
whether updates are performed in “direct” or “deferred” mode. The
exchange operator includes information about the number of producer and
consumer processes the query used.
<TableScan>
<VA>0</VA>
<est>
<rowCnt>18</rowCnt>
<lio>2</lio>
<pio>2</pio>
<rowSz>218.5555</rowSz>
</est>
<varNo>0</varNo>
<objName>titles</objName>
<scanType>TableScan</scanType>
<partitionInfo>
<partitionCount>1</partitionCount>
</partitionInfo>
<scanOrder> ForwardScan </scanOrder>
<positioning> StartOfTable </positioning>
<dataIOSizeInKB>8</dataIOSizeInKB>
<dataBufReplStrategy> LRU </dataBufReplStrategy>
</TableScan>

Showing cached plans in XML

show_cached_plan_in_xml helps track the query performance in the statement
cache. For a given query, show_cached_plan_in_xml, identified by its object ID
or SSQLID and PlanID, returns:

116 Adaptive Server Enterprise

 CHAPTER 3 Displaying Query Optimization Strategies and Estimates

• The header section, which contains information about the cache statement,
such as the statement ID, object ID and the text:
<?xml version="1.0" encoding="UTF-8"?>
<query>
<statementId>1328134997</statementId>
<text>
<![CDATA[SQL Text: select name from sysobjects where id = 10]]>
</text>
If PlanID is set to 0, show_cached_plan_in_xml displays output for all
available plans associated with the cached statement.
• The plan section which contains the plan ID and these subsections:
• Parameter – returns the plan status, the parameters used to compile the
query, and the parameter values that caused the slowest performance.
<planId>11</planId>
<planStatus> available </planStatus>
<execCount>1371</execCount>
<maxTime>3</maxTime>
<avgTime>0</avgTime>
<compileParameters/>
<execParameters/>
• opTree – returns the operators tree, row count, and logical I/O (lio)
and physical I/O (pio) estimates for every operator. the opTree sub-
section returns query plan and optimizer estimates such as lio, pio and
row count
This is an example of an output for the Emit operator.
<opTree>
<Emit>
<VA>1</VA>
<est>
<rowCnt>10</rowCnt>
<lio>0</lio>
<pio>0</pio>
<rowSz>22.54878</rowSz>
</est>
<act>
<rowCnt>1</rowCnt>
</act>
<arity>1</arity>
<IndexScan>
<VA>0</VA>
<est>

Performance and Tuning Series: Query Processing and Abstract Plans 117
set commands for XML format messages

<rowCnt>10</rowCnt>
<lio>0</lio>
<pio>0</pio>
<rowSz>22.54878</rowSz>
</est>
<act>
<rowCnt>1</rowCnt>
<lio>3</lio>
<pio>0</pio>
</act>
<varNo>0</varNo>
<objName>sysobjects</objName>
<scanType>IndexScan</scanType>
<indName>csysobjects</indName>
<indId>3</indId>
<scanOrder> ForwardScan </scanOrder>
<positioning> ByKey </positioning>
<perKey>
<keyCol>id</keyCol>
<keyOrder> Ascending </keyOrder>
</perKey>
<indexIOSizeInKB>2</indexIOSizeInKB>
<indexBufReplStrategy> LRU </indexBufReplStrategy>
<dataIOSizeInKB>2</dataIOSizeInKB>
<dataBufReplStrategy> LRU </dataBufReplStrategy>
</IndexScan>
</Emit>
<opTree>
• execTree – returns the query plan with the operator internal details.
Details vary, depending on the operator. This is an example of an
output for the Emit operator.
<Emit>
<Details>
<VA>5</VA>
<Vtuple Label="Output Vtuple">
<collection Label="Columns (#2)">
<Column>
<0x0x1462d2838) type:GENERIC_TOKEN len:0 offset:0
valuebuf:0x(nil) status:(0x00000008 (STATNULL))
(constant:0x0x1462d24c0 type:INT4 len:4 maxlen:4 constat: (0x0004
(VARIABLE), 0x0002 (PARAM)))
</Column>
<Column>
(0x0x1462d2878) type:GENERIC_TOKEN len:0 offset:0
valuebuf:0x(nil) status:(0x00000008 (STATNULL))

118 Adaptive Server Enterprise

 CHAPTER 3 Displaying Query Optimization Strategies and Estimates

(constant:0x0x1462d26e8 type:INT4 len:4 maxlen:4 constat: (0x0004

(VARIABLE), 0x0002 (PARAM))
</Column>
<Collection>
<Collection Label="Evals">
<EVAL>
constp: 0x0x1462d2290 status: 0 E_ASSIGN
</EVAL>
<EVAL>
constp: 0x0x1462d2348 status: 0 E_ASSIGN
</EVAL>
<EVAL>
constp: 0x(nil) status: 0 E_END
</EVAL>
</Collection>
</Vtuple>
</Details>

Diagnostic usage scenarios

For the following examples, if dbcc traceon(3604) is set, trace information is
sent to the client’s connection. If dbcc traceon (3605) is set, trace information
is sent to the error log. For Adaptive Server versions 15.0.2 and later, you can
use the set switch on. For example:
set switch on 3604
set switch on 3605
Optimization tracing options (dbcc traceon/off(302,310,317)) from versions of
Adaptive Server earlier than 15.0 are no longer supported.
Use dbcc traceon(3604) or set switch on print_output_to_client to direct trace
output to the client process that would otherwise go to the error log. Use dbcc
traceon(3605) or set switch on print_output_to_errorlog to direct output to the
error log as well as to the client process.
Scenario A To send the execution plan XML to the client as trace output, use:
set plan for show_exec_xml to client on
Then run the queries for which the plan is wanted:
select id from sysindexes where id < 0

Performance and Tuning Series: Query Processing and Abstract Plans 119
Diagnostic usage scenarios

Scenario B To get the execution plan, use the showplan_in_xml function. You can get the
output from the last query, or from any of the first 20 queries in a batch or
stored procedure.
set plan for show_opt_xml to message on
Run the query as:
select id from sysindexes where id < 0
select name from sysobjects where id > 0
go

select showplan_in_xml(0)
go
The example generates two XML documents as text streams. You can run an
XPath query over this built-in as long as the XML option is enabled in
Adaptive Server.
select xmlextract("/", showplan_in_xml(-1))
go
This allows the XPath query “/” to be run over the XML doc produced by the
last query.
Scenario C To set multiple options:
set plan for show_exec_xml, show_opt_xml to client on
go

select name from sysobjects where id > 0

go
This sets up the output from the optimizer and the query execution engine to
send the result to the client, as is done in normal tracing.
set plan for show_exec_xml off
go
select name from sysobjects where id > 0
go
The optimizer’s diagnostics are still available, as show_opt_xml is left on.
Scenario D When running a set of queries in a batch, you can ask for the optimizer plan for
the last query.
set plan for show_opt_xml to message on
go
declare @v int
select @v = 1
select name from sysobjects where id = @v

120 Adaptive Server Enterprise

 CHAPTER 3 Displaying Query Optimization Strategies and Estimates

go

select showplan_in_xml(-1)
go
showplan_in_xml() can also be part of the same batch as it works the same way.
Any message for the showplan_in_xml() function is ignored for logging.
To create a stored procedure:
create proc PP as
declare @v int
select @v = 1
select name from sysobjects where id = @v
go

exec PP
go

select showplan_in_xml(-1)
go
If the stored procedure calls another stored procedure, and the called stored
procedure compiles, and optimizer diagnostics are turned on, you get the
optimizer diagnostics for the new set of statements as well. The same is true if
show_execio_xml is turned on and only the called stored procedure is executed.
Scenario E To query the output of the showplan_in_xml() function for the query execution
plan, which is an XML doc:
set plan for show_exec_xml to message on
go

select name from sysobjects

go

select case when

'/Emit/Scan[@Label=“Scan:myobjectss”]' xmltest
showplan_in_xml(-1)
then “PASSED” else "FAILED" end
go

set plan for show_exec_xml off

go

Performance and Tuning Series: Query Processing and Abstract Plans 121
Permissions for set commands

Scenario F Use show_final_plan_xml to configure Adaptive Server to display the query

plan as XML output. This output does not include the actual LIO costs, PIO
costs, or the row counts. Once show_final_plan_xml is enabled, you can select
the query plan from the last run query (which has a query ID of -1). To enable
show_final_plan_xml:
set plan for show_final_plan_xml to message on
Run your query, for example:
use pubs2
go
select * from titles
go
Select the query plan for the last query run using the showplan_in_xml
parameter:
select showplan_in_xml(-1)

Permissions for set commands

The sa_role has full access to the set commands described above.
For other users, the system administrator must grant and revoke new set tracing
permissions to allow set option and set plan for XML, as well as dbcc traceon/off
(3604,3605).
For more information, see the grant command description in Adaptive Server
Reference Manual: Commands.

Analyzing dynamic parameters

Adaptive Server lets you analyze dynamic parameters (which are indicated by
question marks) before running a query, helping you avoid inefficient query
plans.
Analyze dynamic parameters using:
• @@lwpid global variable – returns the object ID of the most recently
prepared lightweight procedure that corresponds to a dynamic SQL
prepare statement.

122 Adaptive Server Enterprise

 CHAPTER 3 Displaying Query Optimization Strategies and Estimates

• @@plwpid global variable – returns the object ID of the next most recently
prepared lightweight procedure that corresponds to a dynamic SQL
prepare statement.

• show_dynamic_params_in_xml – displays information about parameters in

dynamic SQL statements. See the Reference Manual: Blocks.
Using the value provided by @@plwpid as the value for the
show_dynamic_params_in_xml object_id parameter, Adaptive Server displays
information about the dynamic parameters in the query. Continue refining the
parameters until you find the ones that provide you with the best query plan.
Disable the statement cache before you analyze dynamic parameters for a
query. The statement cache reuses query plans to avoid compilation.
The output of show_dynamic_params_in_xml is similar to:
<!ELEMENT query (parameter*)>
<!ELEMENT parameter (number, type, column?)>
<!ELEMENT number (#PCDATA)>
<!ELEMENT type (#PCDATA)>
<!ELEMENT column (#PCDATA)>
The root element of the document is <query>, which can have zero or more
<parameter> elements. Each <parameter> element includes:
• number – the dynamic parameter's position in the statement (starting at 1).
• type – the datatype.
• column – the name of a table and column (in the format table.column)
associated with the dynamic parameter, if such an association exists. If no
association exists, there is no column information in the output.

Dynamic parameter example analysis

A typical use case for analyzing dynamic parameters is:
1 Disable statement cache:
set statement_cache off
2 Prepare the statement to be examined that contains the dynamic
parameters.
3 Select @@plwpid to determine the object ID of the most recent lightweight
procedure.

Performance and Tuning Series: Query Processing and Abstract Plans 123
Analyzing dynamic parameters

4 Run show_dynamic_params_in_xml using the value of @@plwpid as the

value for object_id, and displaying the result in the local variable named
xmldoc1.

5 Close the statement.

6 Enable the statement cache:
set statement_cache on
7 Analyze the results in xmldoc1, and select values for the parameters.

124 Adaptive Server Enterprise

CH A PTE R 4 Finding Slow Running Queries

Topic Page
Saving diagnostics to a trace file 125
Displaying SQL text 129
Retaining session settings 132

Adaptive Server includes the set show_sqltext, set tracefile, and set
export_options parameters that enable you to collect diagnostic
information about poorly-running queries without having to previously
enable showplan or other investigatory parameters.

Saving diagnostics to a trace file

Once enabled, set tracefile saves all SQL text for the current session to the
specified file, each SQL text batch appending to the previous batch.
The syntax to enable tracing is:
set tracefile file_name [off] [for spid
The syntax to disable tracing is:
set tracefile off [for spid]
Where:

Performance and Tuning Series: Query Processing and Abstract Plans 125
Saving diagnostics to a trace file

• file_name – is the full path to the file in which you are saving the SQL text.
If you do not specify a directory path, Adaptive Server creates the file in
$SYBASE.

Note If file_name contains special characters (“:”, “/”, and so on) other
than numbers and letters, you must include file_name in quotes. For
example, this file_name must be in quotes because of the “/” for the
directory structure:
set tracefile '/tmp/mytracefile.txt' for 25
If file_name does not contain special characters and you want to save it to
$SYBASE, it does not require quotes. For example, this file_name does not
need to be in quotes:
set tracefile mytracefile.txt

• off – disables the tracing for this session or spid.

• spid – server process ID whose SQL text you want saved to a trace file.
Only the users with the SA or SSO role can enable tracing for other spids.
You cannot save the SQL text for system tasks (such as the housekeeper
or the port manager).
Examples • This example opens a trace file named sql_text_file for the the current
session:
set tracefile '/var/sybase/REL1502/text_dir/sql_text_file'
Subsequent outputs from set showplan, set statistics io, and dbcc
traceon(100) are saved in sql_text_file.

• This example does not specify a directory path, so the trace file is saved in
$SYBASE/sql_text_file:
set tracefile 'sql_text_file' for 11
Any SQL run on spid 11 is saved to this tracefile.
• This example saves the SQL text for spid 86:
set tracefile
'/var/sybase/REL1502/text_dir/sql_text_file' for 86
• This example disables set tracefile:
set tracefile off
These are the restrictions for set tracefile:

126 Adaptive Server Enterprise

 CHAPTER 4 Finding Slow Running Queries

• You cannot save the SQL text for system tasks (such as the housekeeper
or the port manager).
• You must have the sa or sso roles, or be granted set tracing permission, to
run enable or disable tracing.
• set tracefile is not allowed to open an existing file as a tracefile.

• During an SA or SSO session, if you enable set tracfile for a specific spid,
all subsequent tracing commands executed take effect on that spid, not the
SA or SSO spid.
• If Adaptive Server runs out of file space while writing the tracefile, it
closes the file and disables the tracing.
• If an isql session starts tracing for a spid, but the isql session quits without
disabling the tracing, another isql session can begin tracing this spid.
• Tracing occurs for the session for which it is enabled only, not for the
session that enabled it.
• You cannot trace more than one session at a time from a single sa or sso
session. If you attempt to open a tracefile for a session for which there is
already a trace file open, Adaptive Server issues this error message:
tracefile is already open for this session.

• You cannot trace the same session from multiple sa or sso sessions.
• The file storing the trace output is closed when the session being traced
quits or when you disable tracing.
• Before you allocate resources for tracing, keep in mind that each tracing
requires one file descriptor per engine.

Set options that save diagnostic information to a trace file

You can use set tracefile in combination with other set commands and options
that provide diagnostic information for a better understanding of slow-running
queries. These are the set commands and options that save diagnostic
information to a file:
• set show_sqltext [on | off]

• set showplan [on | off]

• set statistics io [on | off]

• set statistics time [on | off]

Performance and Tuning Series: Query Processing and Abstract Plans 127
Saving diagnostics to a trace file

• set statistics plancost [on | off]

These are the set options:

• set option show [normal | brief | long | on | off]

• set option show_lop [normal | brief | long | on | off]

• set option show_parallel [normal | brief | long | on | off]

• set option show_search_engine [normal | brief | long | on | off]

• set option show_counters [normal | brief | long | on | off]

• set option show_managers [normal | brief | long | on | off]

• set option show_histograms [normal | brief | long | on | off]

• set option show_abstract_plan [normal | brief | long | on | off]

• set option show_best_plan [normal | brief | long | on | off]

• set option show_code_gen [normal | brief | long | on | off]

• set option show_pio_costing [normal | brief | long | on | off]

• set option show_lio_costing [normal | brief | long | on | off]

• set option show_log_props [normal | brief | long | on | off]

• set option show_elimination [normal | brief | long | on | off]

Which sessions are being traced?

Use sp_helpapptrace to determine which sessions Adaptive Server is tracing.
sp_helpapptrace returns the server process IDs (spids) for all the sessions
Adaptive Server is tracing, the spids of the sessions tracing them, and the name
of the tracefile.
The syntax for sp_helpapptrace is:
sp_helpapptrace
sp_helpapptrace returns these columns:

• traced_spid – spid of the session you are tracing.

• tracer_spid – spid of the session that traced_spid is tracing. Prints “exited”

if the tracer_spid session has exited.
• trace_file – full path to the tracefile.

128 Adaptive Server Enterprise

 CHAPTER 4 Finding Slow Running Queries

For example:
sp_helpapptrace
traced_spid tracer_spid trace_file
-------------- ------------- ----------
11 exited /tmp/myfile1
13 14 /tpcc/sybase.15_0/myfile2

Rebinding a trace
If a session is tracing another session, but quits without disabling the tracing,
Adaptive Server allows a new session to rebind with the earlier trace. This
means that a sa or sso is not required to finish every trace they start, but can
start a trace session, quit, and then rebind to this trace session

Displaying SQL text

set show_sqltext allows you to print the SQL text for ad-hoc queries, stored
procedures, cursors, and dynamic prepared statements. You do not need to
enable the set show_sqltext before you execute the query (as you do with
commands like set showplan on) to collect diagnostic information for a SQL
session. Instead you can enable it while the commands are running to help
determine which query is performing poorly and diagnose their problems.
Before you enable show_sqltext, you must first enable dbcc traceon to display
the output to standard out:
dbcc traceon(3604)
The syntax for set show_sqltext is:
set show_sqltext {on | off}
For example, this enables show_sqltext:
set show_sqltext on
Once set show_sqltext is enabled, Adaptive Server prints all SQL text to
standard out for each command or system procedure you enter. Depending on
the command or system procedure you run, this output can be extensive.
For example, if you run sp_who, Adaptive Server prints all SQL text associated
with this system procedure (the output is abbreviated for space purposes):

Performance and Tuning Series: Query Processing and Abstract Plans 129
Displaying SQL text

sp_who
2007/02/23 02:18:25.77
SQL Text: sp_who
Sproc: sp_who, Line: 0
Sproc: sp_who, Line: 20
Sproc: sp_who, Line: 22
Sproc: sp_who, Line: 25
Sproc: sp_who, Line: 27
Sproc: sp_who, Line: 30
Sproc: sp_who, Line: 55
Sproc: sp_who, Line: 64
Sproc: sp_autoformat, Line: 0
Sproc: sp_autoformat, Line: 165
Sproc: sp_autoformat, Line: 167
Sproc: sp_autoformat, Line: 177
Sproc: sp_autoformat, Line: 188
. . .
Sproc: sp_autoformat, Line: 326
Sproc: sp_autoformat, Line: 332
SQL Text: INSERT
#colinfo_af(colid,colname,usertype,type,typename,collength,maxlength,autoform
at,selected,selectorder,asname,mbyte) SELECT
c.colid,c.name,t.usertype,t.type,t.name,case when c.length < 80 then 80 else
c.length end,0,0,0,0,c.name,0 FROM tempdb.dbo.syscolumns c,tempdb.dbo.systypes
t WHERE c.id=1949946031 AND c.usertype=t.usertype
Sproc: sp_autoformat, Line: 333
Sproc: sp_autoformat, Line: 334
. . .
Sproc: sp_autoformat, Line: 535
Sproc: sp_autoformat, Line: 0
Sproc: sp_autoformat, Line: 393
Sproc: sp_autoformat, Line: 395
. . .
Sproc: sp_autoformat, Line: 686
Sproc: sp_autoformat, Line: 688
SQL Text: UPDATE #colinfo_af SET maxlength=(SELECT
isnull(max(isnull(char_length(convert(varchar(80),fid)),4)),1) FROM
#who1result ), autoformat = 1, mbyte=case when usertype in (24, 25, 34, 35) then
1 else 0 end WHERE colname='fid'
Sproc: sp_autoformat, Line: 689
Sproc: sp_autoformat, Line: 690
. . .
Sproc: sp_autoformat, Line: 815
Sproc: sp_autoformat, Line: 818
SQL Text: SELECT
fid=right(space(80)+isnull(convert(varchar(80),fid),'NULL'),3),

130 Adaptive Server Enterprise

 CHAPTER 4 Finding Slow Running Queries

spid=right(space(80)+isnull(convert(varchar(80),spid),'NULL'),4),
status=SUBSTRING(convert(varchar(80),status),1,8),
loginame=SUBSTRING(convert(varchar(80),loginame),1,8),
origname=SUBSTRING(convert(varchar(80),origname),1,8),
hostname=SUBSTRING(convert(varchar(80),hostname),1,8),
blk_spid=right(space(80)+isnull(convert(varchar(80),blk_spid),'NULL'),8),
dbname=SUBSTRING(convert(varchar(80),dbname),1,6),
tempdbname=SUBSTRING(convert(varchar(80),tempdbname),1,10),
cmd=SUBSTRING(convert(varchar(80),cmd),1,17),
block_xloid=right(space(80)+isnull(convert(varchar(80),block_xloid),'NULL'),1
1) FROM #who1result order by fid, spid, dbname

Sproc: sp_autoformat, Line: 819

Sproc: sp_autoformat, Line: 820
Sproc: sp_autoformat, Line: 826
Sproc: sp_who, Line: 68
Sproc: sp_who, Line: 70
fid spid status loginame origname hostname blk_spid dbname
tempdbnamecmd block_xloid
--- ---- -------- -------- -------- -------- -------- -------------
------------ -----------
0 2 sleeping NULL NULL NULL 0 master tempdb
DEADLOCK TUNE 0
0 3 sleeping NULL NULL NULL 0 master tempdb
ASTC HANDLER 0
0 4 sleeping NULL NULL NULL 0 master tempdb
CHECKPOINT SLEEP 0
0 5 sleeping NULL NULL NULL 0 master tempdb
HK WASH 0
0 6 sleeping NULL NULL NULL 0 master tempdb
HK GC 0
0 7 sleeping NULL NULL NULL 0 master tempdb
HK CHORES 0
0 8 sleeping NULL NULL NULL 0 master tempdb
PORT MANAGER 0
0 9 sleeping NULL NULL NULL 0 master tempdb
NETWORK HANDLER 0
0 10 sleeping NULL NULL NULL 0 master tempdb
LICENSE HEARTBEAT 0
0 1 running sa sa echo 0 master tempdb
INSERT 0

(10 rows affected)

(return status = 0)

Performance and Tuning Series: Query Processing and Abstract Plans 131
Retaining session settings

To disable show_sqltext, enter:

set show_sqltext off
Restrictions for • You must have the sa or sso roles to run show_sqltext.
show_sqltext
• You cannot use show_sqltext to print the SQL text for triggers.
• You cannot use show_sqltext to show a binding variable or a view name.

Retaining session settings

Adaptive Server’s default behavior is to reset any set parameter changes that
are set by a trigger or system procedure after they finish running. Enabling set
export_options allows you to export the session settings set by a system
procedure or trigger to the parent (or issuer) of the stored procedure or trigger.
In this example, Adaptive Server exports set showplan on to outer_proc, but
does not exported it to a parent stored procedure:
create proc inner_proc
as
set showplan on
select * from titles
set export_options on
go

create proc outer_proc

as
exec inner_proc
go

132 Adaptive Server Enterprise

CH A PTE R 5 Parallel Query Processing

This chapter provides an in-depth description of parallel query processing.

Topic Page
Vertical, horizontal, and pipelined parallelism 133
Queries that benefit from parallel processing 134
Enabling parallelism 135
Controlling parallelism at the session level 139
Controlling query parallelism 141
Using parallelism selectively 141
Using parallelism with large numbers of partitions 143
When parallel query results differ 144
Understanding parallel query plans 146
Adaptive Server parallel query execution model 148

Vertical, horizontal, and pipelined parallelism

Adaptive Server supports horizontal and vertical parallelism for query
execution. Vertical parallelism is the ability to run multiple operators
simultaneously by employing different system resources such as CPUs,
disks, and so on. Horizontal parallelism is the ability to run multiple
instances of an operator on the specified portion of the data.
The way you partition your data greatly affects the efficiency horizontal
parallelism. The logical partitioning of data is useful in operational
decision-support systems (DSS) queries where large volumes of data are
being processed.
See Chapter 10, “Partitioning Tables and Indexes,” in the Transact-SQL
User’s Guide and the Performance and Tuning Series: Physical Database
Tuning guide for a more detailed discussion of partitioning on Adaptive
Server. Understanding different types of partitioning is a prerequisite to
understanding this chapter.

Performance and Tuning Series: Query Processing and Abstract Plans 133
Queries that benefit from parallel processing

Adaptive Server also supports pipelined parallelism. Pipelining is a form of

vertical parallelism in which intermediate results are piped to higher operators
in a query tree. The output of one operator is used as input for another operator.
The operator used as input can run simultaneously with the operator feeding
the data, which is an essential element in pipelined parallelism. Use parallelism
only when multiple resources like disks and CPUs are available. Using
parallelism can be detrimental if your system is not configured for resources
that can work in tandem. In addition, data must be spread across disk resources
in a way that closely ties the logical partitioning of the data with the physical
partitioning on parallel devices. The biggest challenge for a parallel system is
to control the correct granularity of parallelism. If parallelism is too finely
grained, communication and synchronization overhead can offset any benefit
that is obtained from parallel operations. Making parallelism too coarse does
not permit proper scaling.

Queries that benefit from parallel processing

When Adaptive Server is configured for parallel query processing, the query
optimizer evaluates each query to determine whether it is eligible for parallel
execution. If it is eligible, and if the optimizer determines that a parallel query
plan can deliver results faster than a serial plan, the query is divided into plan
fragments that are processed simultaneously. The results are combined and
delivered to the client in a shorter period of time than it takes to process the
query serially as a single fragment.
Parallel query processing can improve the performance of:
• select statements that scan large numbers of pages but return relatively few
rows, such as table scans or clustered index scans with grouped or
ungrouped aggregates.
• Table scans or clustered index scans that scan a large number of pages, but
have where clauses that return only a small percentage of rows.
• select statements that include union, order by, or distinct, since these query
operations can make use of parallel sorting or parallel hashing.
• select statements where a reformatting strategy is chosen by the optimizer,
since these can populate worktables in parallel and can make use of
parallel sorting.
• join queries.

134 Adaptive Server Enterprise

 CHAPTER 5 Parallel Query Processing

Commands that return large, unsorted result sets are unlikely to benefit from
parallel processing due to network constraints. In most cases, results can be
returned from the database faster than they can be merged and returned to the
client over the network.
Parallel DMLs like insert, delete, and update are not supported and so do not
benefit from parallelism.

Enabling parallelism
To configure Adaptive Server for parallelism, enable the number of worker
processes and max parallel degree parameters.

To gain optimal performance, be aware of other configuration parameters that

affect the quality of plans generated by Adaptive Server.

number of worker processes

Before you enable parallelism, configure the number of worker processes (also
referred to as threads) available for Adaptive Server by setting the
configuration parameter number of worker processes. Sybase recommends that
you set the value for number of worker processes to one and a half times the
total number required at peak load. You can calculate an approximate number
using the max parallel degree configuration parameter, which indicates the total
number of worker processes that can be used for any query. Depending on the
number of connections to the Adaptive Server and the approximate number of
queries that are run simultaneously, you can use this rule to roughly estimate
the value for the number of worker processes that may be needed at any time:
[number of worker processes] = [max parallel degree] X [the number of
concurrent connections wanting to run queries in parallel] X [1.5]
For example, to set the number of worker processes to 40:
sp_configure "number of worker processes", 40
Any runtime adjustment for the number of threads may have a negative effect
on query performance. Adaptive Server always tries to optimize thread usage,
but it may have already committed to a plan that needs increased resources, and
therefore does not guarantee a linear scaledown when it runs with fewer
threads.

Performance and Tuning Series: Query Processing and Abstract Plans 135
Enabling parallelism

If the query processor has insufficient worker processes, the processor tries to
adjust the query plan during runtime. If a minimal number of worker processes
are required but unavailable, the query aborts with this error message:
Insufficient number of worker processes to execute the
parallel query. Increase the value of the configuration
parameter ‘number of worker processes’

max parallel degree

Use the max parallel degree configuration parameter to configure the maximum
amount of parallelism for a query. This parameter determines the maximum
number of threads Adaptive Server uses when processing a given query. For
example, to set max parallel degree to 10, enter:
sp_configure "max parallel degree", 10
Unlike versions of Adaptive Server earlier than 15.0, this parameter’s value is
not entirely enforced by the query optimizer. A complete enforcement process
is expensive in terms of optimization time. Adaptive Server comes close to the
desired setting of max parallel degree and exceeds it only for semantic reasons.

max resource granularity

The value of max resource granularity configures the maximum percentage of
system resources a query can use. In Adaptive Server version 15.0 and later,
max resource granularity affects only procedure cache. By default, max resource
granularity is 10%. However, this value is not enforced at execution time; it is
only a guide for the query optimizer. The query engine can avoid memory-
intensive strategies, such as hash-based algorithms, when max resource
granularity is set to a low value.

To set max resource granularity to 5%, enter:

sp_configure "max resource granularity", 5
If the the query processor’s seach engine has consumed more than the
configured percentage of procedure cache, and if it has found at least one full
plan, the search engine times out and uses the current best plan for the query.

136 Adaptive Server Enterprise

 CHAPTER 5 Parallel Query Processing

If the query processor does not find a full plan before reaching the value for
max resource granularity as a percentage of procedure cache, the search engine
continues to search until it finds the next full plan. However, if the search
engine reaches 50% of the complete procedure cache and finds no plan, it
aborts the query compilation to avoid shutting down the server.

max repartition degree

Adaptive Server must dynamically repartition intermediate data to match the
partitioning scheme of another operand or to perform an efficient partition
elimination. max repartition degree controls the amount of dynamic
repartitioning Adaptive Server can do. If the value of max repartition degree is
too high, the number of intermediate partitions becomes too large and the
system becomes flooded with worker processes that compete for resources,
which eventually degrades performance. The value for max repartition degree
enforces the maximum number of partitions created for any intermediate data.
Repartitioning is a CPU-intensive operation. The value of max repartition
degree should not exceed the total number of Adaptive Server engines.

If all tables and indexes are unpartitioned, Adaptive Server uses the value for
max repartition degree to provide the number of partitions to create as a result
of repartitioning the data. When the value is set to 1, which is the default case,
the value of max repartition degree is set to the number of online engines.
Use max repartition degree when using the force option to perform a parallel
scan on a table or index.
select * from customers (parallel)
For example, if the customers table is unpartitioned and the force option is used,
Adaptive Server tries to find the inherent partitioning degree of that table or
index, which in this case is 1. It uses the number of engines configured for the
server, or whatever degree is best based on the number of pages in the table or
index that does not exceed the value of max repartition degree.
To set max repartition degree to 5:
sp_configure "max repartition degree", 5

Performance and Tuning Series: Query Processing and Abstract Plans 137
Enabling parallelism

max scan parallel degree

The max scan parallel degree configuration parameter is used only for
backward compatibility, when the data in a partitioned table or index is highly
skewed. If the value of this parameter is greater than 1, Adaptive Server uses
this value to do a hash-based scan. The value of max scan parallel degree cannot
exceed the value of max parallel degree.

prod-consumer overlap factor

prod-consumer overlap factor affects how much pipelined parallelism can be
created in a query plan. The default value is 20%, which means that if two
operators in a parent-child relationship are run by separate worker processes,
there is a 20% overlap. The remaining 80% of the operation is sequential. This
affects the way in which Adaptive Server costs two plan fragments. Consider
the example of a scan operator under a grouping operation. In such a case, if
the scan operator takes N1 seconds and grouping operations take N2 seconds,
the response time of the two operators is:
0.2 * max (N1, N2) + 0.8 * (N1 + N2)
In setting this parameter, consider the number of online engines on which
Adaptive Server is running and the complexity of the queries to be run. As a
general rule, use thread resources to scan on multiple partitions first. Then, if
there are unused thread resources, use them to speed up vertical pipelined
parallelism. Do not exceed a value of 50.

min pages for parallel scan

max pages for parallel scan controls the tables and indexes that can be accessed
in parallel. If the number of pages in a table is below this value, the table is
accessed serially. The default value is 200 pages; page size is not relevant.
Although the tables and indexes of the table are accessed serially, Adaptive
Server tries to repartition the data, if that is appropriate, and to use parallelism
above the scans, if that is appropriate.

138 Adaptive Server Enterprise

 CHAPTER 5 Parallel Query Processing

max query parallel degree

max query parallel degree defines the number of worker processes to use for a
given query. This parameter is relevant only if you do not want to enable
parallelism globally. You must configure the number of worker processes to a
value greater than zero, but max query parallel degree must be set to 1.
When max query parallel degree is set to a value greater than 1, queries are not
compiled to use parallelism. Instead, you can specify parallel hints, using
abstract plans to compile one or more queries using parallelism.
Use use parallel N to define how much parallelism is to be used for a given
query. Alternatively, use create plan to specify the query and the number of
worker processes to use for it.

Controlling parallelism at the session level

The set options let you restrict the degree of parallelism on a session basis, in
stored procedures, or in triggers. These options are useful for tuning
experiments with parallel queries and can also be used to restrict noncritical
queries to run in serial, so that worker processes remain available for other
tasks.
Table 5-1: Session-level parallelism control parameters
Parameter Function
parallel_degree Sets the maximum number of worker processes for a query in a session, stored procedure,
or trigger. Overrides the max parallel degree configuration parameter, but must be less than
or equal to the value of max parallel degree.
scan_parallel_degree Sets the maximum number of worker processes for a hash-based scan during a specific
session, stored procedure, or trigger. Overrides the max scan parallel degree configuration
parameter and must be less than or equal to the value of max scan parallel degree.
resource_granularity Overrides the global value max resource granularity and sets it to a session-specific value,
which influences whether Adaptive Server uses memory-intensive operations.
repartition_degree Sets the value of max repartition degree for a session. This is the maximum degree to which
any intermediate data stream is be repartitioned for semantic purposes.

Performance and Tuning Series: Query Processing and Abstract Plans 139
Controlling parallelism at the session level

If you specify a value that is too large for any of the set options, the value of
the corresponding configuration parameter is used, and a message reports the
value that is in effect. While set parallel_degree, set scan_parallel_degree, set
repartition_degree, or set resource_granularity is in effect during a session, the
plans for any stored procedures that you execute are not placed in the procedure
cache. Procedures executed with these set options in effect may produce less
than optimal plans.

set command examples

This example restricts all queries started in the current session to 5 worker
processes:
set parallel_degree 5
While this command is in effect, any query on a table with more than 5
partitions cannot use a partition-based scan.
To remove the session limit, use:
set parallel_degree 0
or:
set scan_parallel_degree 0
To run subsequent queries in serial mode, use:
set parallel_degree 1
or:
set scan_parallel_degree 1
To set resource granularity to 25% of the total resources available in the
system, use:
set resource_granularity 25
The same is true for repartition degree as well; you can set it to a value of 5. It
cannot, however, exceed the value of max parallel degree.
set repartition_degree 5

140 Adaptive Server Enterprise

 CHAPTER 5 Parallel Query Processing

Controlling query parallelism

The parallel extension to the from clause of a select command allows users to
suggest the number of worker processes used in a select statement. The degree
of parallelism that you specify cannot be more than the value set with
sp_configure or the session limit controlled by a set command. If you specify a
higher value, the specification is ignored, and the optimizer uses the set or
sp_configure limit.

The syntax for the select statement is:

select ...
from tablename [( [index index_name]
[parallel [degree_of_parallelism | 1 ]]
[prefetch size] [lru|mru] ) ] ,
tablename [( [index index_name]
[parallel [degree_of_parallelism | 1]
[prefetch size] [lru|mru] ) ] ...

Query-level parallel clause examples

To specify the degree of parallelism for a single query, include parallel after
the table name. This example executes in serial:
select * from huge_table (parallel 1)
This example specifies the index to use in the query, and sets the degree of
parallelism to 2:
select * from huge_table (index ncix parallel 2)

Using parallelism selectively

Not all queries benefit from parallelism. In general, the optimizer determines
which queries will not benefit from parallelism and attempts to run them
serially. When the query processor makes errors in such cases, it is usually
because of skewed statistics or incorrect costing as a result of imperfect
modeling. Experience will show you whether queries are running better or
worse, and you can decide to keep parallel on or off.
If you keep parallel on, and have identified the queries you want to run in serial
mode, you can attach an abstract plan hint, as follows:

Performance and Tuning Series: Query Processing and Abstract Plans 141
Using parallelism selectively

select count(*) from sysobjects

plan “(use parallel 1)”
The same effect is achieved by creating a query plan:
create plan “select count(*) from sysobjects”
“use parallel 1”
If, however, you notice that parallelism is resource-intensive or that it does not
generate query plans that perform well, use it selectively. To enable parallelism
for selected complex queries:
1 Set the number of worker processes to a number greater than zero, based
on the guidelines in “number of worker processes” on page 135. For
example, to configure 10 worker processes, execute:
sp_configure “number of worker processes”, 10
2 Set max query parallel degree to a value greater than 1. As a starting point,
set it to what you would have used for max parallel degree:
sp_configure “max query parallel degree”, 10
3 The preferred way to force a query to use a parallel plan is to use the
abstract plan syntax:
use parallel N
where N is less than the value of max query parallel degree.
To write a query that uses a maximum of 5 threads, use:
select count (*), S1.id from sysobjects S1, sysindexes S2
where S1.id = S2.id
group by S1.id
plan
“(use parallel 5)”
This query tells the optimizer to use 5 worker processes, if it can. the only
drawback to this approach is that you must alter the actual queries in the
application. To avoid this, use create plan:
create plan
“select count(*), S1.id from sysobjects S1, sysindexes S2
where S1.id = S2.id
group by S1.id”
“(use parallel 5)”
To turn the abstract plan load option on globally, enter:
sp_configure “abstract plan load”, 1

142 Adaptive Server Enterprise

 CHAPTER 5 Parallel Query Processing

See “Creating and Using Abstract Plans” on page 321 for more
information about using abstract plans.

Using parallelism with large numbers of partitions

The information in this section also applies when partitioning is configured for
manageability, and when partitions are created on physical or logical devices
that exhibit little or no parallelism.
For the purpose of this discussion, you have decided to partition a table using
range partitioning that represents each week of a year. The issue here is that the
query optimizer does not know how the underlying disk system will respond to
a 52-way parallel scan. The optimizer must determine the best way to scan the
table. If there are enough worker processes configured, the optimizer uses 52
threads to scan the table, which may well cause serious performance issues and
be even slower than a serial scan.
To prevent this, first find out exactly how much parallelism is supported. If you
know the devices that are used for this table, you can use the following
command on a UNIX system, where the underlying device is called /dev/xx:
time dd if=/dev/xx of=/dev/null bs=2k skip=8 count = 102400 &
Assume that time records as x.
Now run two of the same commands concurrently:
time dd if=/dev/xx of=/dev/null bs=2k skip=8 count = 102400 &
time dd if=/dev/xx of=/dev/null bs=2k skip=8 count = 102400 &
This time, assume that time is y. In a linear scale-up, x is the same as y, which
is probably impossible to achieve. The following identity may suffice:
x <= y <= (N*x)/k
Where N is the number of simultaneous sessions started and k is a constant that
identifies an acceptable improvement level. A good approximation of k might
be 1.4, which says that parallel scan is allowed as long as it delivers 40% better
metrics than a serial scan.

Performance and Tuning Series: Query Processing and Abstract Plans 143
When parallel query results differ

Table 5-2: Parallel scan metrics

Performanc
Number of threads e metrics Acceptable for k=1.4
1 200s
2 245s 245 <= (200*2)/1.4; i.e. 245<=285.71
4 560s 560 <= (200*4)/1.4; i.e. 560<=571.42
5 725s 725 <= (200*5)/1.4; i.e. 725<=714.28

Table 5-2 shows that the disk subsystem did not perform well after four
concurrent accesses; the performance numbers went below the acceptable limit
established by k. In general, read enough data blocks to allow for any skewed
readings.
Having established that 4 threads is optimal, provide this hint by binding it to
the object using sp_chgattribute in this way:
sp_chgattribute <tablename>, “plldegree”, 4
This tells the query optimizer to use a maximum of four threads. It may choose
fewer than four threads if it does not find enough resources. The same
mechanism can be applied to an index. For example, if an index called auth_ind
exists on authors and you want to use two threads to access it, use:
sp_chgattribute “authors.auth_ind”, “plldegree”, 4
You must run sp_chgatttribute from the current database.

When parallel query results differ

When a query does not include scalar aggregates or require a final sorting step,
a parallel query might return results in a different order from the same query
run in serial, and subsequent executions of the same query in parallel might
return results in different orders. The relative speed of the different worker
processes leads to differences in result-set ordering. Each parallel scan behaves
differently, due to pages already in cache, lock contention, and so forth. Parallel
queries always return the same set of results, just not in the same order.

Note If you need a dependable ordering of results, use order by, or run the
query in serial mode.

144 Adaptive Server Enterprise

 CHAPTER 5 Parallel Query Processing

In addition, due to the pacing effects of multiple worker processes reading data
pages, two types of queries accessing the same data may return different results
when an aggregate or a final sort is not done. They are:
• Queries that use set rowcount
• Queries that select a column into a local variable without sufficiently
restrictive query clauses

Queries that use set rowcount

The set rowcount option stops processing from continuing after a certain
number of rows are returned to the client. With serial processing, the results are
consistent in repeated executions as long as the query plans are the same. In
serial mode, given the same query plan, the same rows are returned in the same
order for a given rowcount value, because a single process reads the data pages
in the same order every time. With parallel queries, the order of the results and
the set of rows returned can differ, because worker processes may access pages
sooner or later than other processes. To get consistent results, either use a
clause that performs a final sort step, or run the query in serial.

Queries that set local variables

This query sets the value of a local variable in a select statement:
select @tid = title_id from titles
where type = "business"
The where clause matches multiple rows in the titles table, so the local variable
is always set to the value from the last matching row returned by the query. The
value is always the same in serial processing, but for parallel query processing,
the results depend on which worker process finishes last. To achieve a
consistent result, use a clause that performs a final sort step, execute the query
in serial mode, or add clauses so that the query arguments select only single
rows.

Performance and Tuning Series: Query Processing and Abstract Plans 145
Understanding parallel query plans

Understanding parallel query plans

The key to understanding parallel query processing in Adaptive Server is to
understand the basic building blocks in a parallel query plan.

Note See Chapter 2, “Using showplan,” which explains how to display a query
plan in a text-based format for each SQL statement in a batch or stored
procedure.

A compiled query plan contains a tree of execution operators that closely

resembles the relational semantics of the query. Each query operator
implements a relational operation using a specific algorithm. For example, a
query operator called the nested-loop join implements the relational join
operation. In Adaptive Server, the primary operator for parallelism is the
exchange operator, which is a control operator that does not implement any
relational operation. An exchange operator is to create new worker processes
that can handle a fragment of the data. During optimization, Adaptive Server
strategically places the exchange operator to create operator tree fragments that
can run in parallel. All operators found below the exchange operator (down to
the next exchange operator) are executed by worker threads that clone the
fragment of the operator tree to produce data in parallel. The exchange operator
can then redistribute this data to the parent operator above it in the query plan.
The exchange operator handles the pipelining and rerouting of data.
In the following sections, the word “degree” is used in two different contexts.
When “degree N” of a table or index is referred to, it references the number of
partitions contained in a table or index. When the “degree of an operation” or
“the degree of a configuration parameter” is referred to, it references the
number of partitions generated in the intermediate data stream.
The following example shows how operators in the query processor work in
serial with the following query run in the pubs2 database. The table titles is
hash-partitioned three ways on the column pub_id.
select * from titles
QUERY PLAN FOR STATEMENT 1 (at line 1).

1 operator(s) under root

The type of query is SELECT.

ROOT:EMIT Operator

146 Adaptive Server Enterprise

 CHAPTER 5 Parallel Query Processing

|SCAN Operator
| FROM TABLE
| titles
| Table Scan.
| Forward Scan.
| Positioning at start of table.
| Using I/O Size 2 Kbytes for data pages.
| With LRU Buffer Replacement Strategy for data
pages.
As this example illustrates, the titles table is being scanned by the scan
operator, the details of which appear in the showplan output. The emit operator
reads the data from the scan operator and sends it to the client application. A
given query can create an arbitrarily complex tree of such operators.
When parallelism turned on, Adaptive Server can perform a simple scan in
parallel using the exchange operator above the scan operator. exchange
produces three worker processes (based on the three partitions), each of which
scans the three disjointed parts of the table and sends the output to the
consumer process. The emit operator at the top of the tree does not know that
the scans are done in parallel.
Example A:
select * from titles
Executed in parallel by coordinating process and 3 worker processes.

4 operator(s) under root

The type of query is SELECT.

ROOT:EMIT Operator

|EXCHANGE Operator (Merged)

|Executed in parallel by 3 Producer and 1 Consumer processes.

|
| |EXCHANGE:EMIT Operator
| |
| | |RESTRICT Operator
| | |
| | | |SCAN Operator
| | | | FROM TABLE
| | | | titles
| | | | Table Scan.
| | | | Forward Scan.
| | | | Positioning at start of table.

Performance and Tuning Series: Query Processing and Abstract Plans 147
Adaptive Server parallel query execution model

| | | | Executed in parallel with a 3-way partition scan.

| | | | Using I/O Size 2 Kbytes for data pages.
| | | | With LRU Buffer Replacement Strategy for data pages.
The operator called EXCHANGE: EMIT is placed under an EXCHANGE operator
to funnel data. See “EXCHANGE operator” on page 148.

Adaptive Server parallel query execution model

One of the key components of the parallel query execution model is the
EXCHANGE operator. You can see it in the showplan output of a query.

EXCHANGE operator
The EXCHANGE operator marks the boundary between a producer and a
consumer operator (the operators below the EXCHANGE operator produce data
and those above it consume data). Example A, which showed parallel scan of
the titles table (select * from titles), the EXCHANGE: EMIT and the SCAN
operator produce data. This is shown briefly.
select * from titles

The type of query is SELECT.

ROOT:EMIT Operator

|EXCHANGE Operator (Merged)

|Executed in parallel by 3 Producer and 1 Consumer
processes.
|
| |EXCHANGE:EMIT Operator
| |
| | |RESTRICT Operator
| | |
| | | |SCAN Operator
| | | | FROM TABLE
| | | | titles
| | | | Table Scan.

148 Adaptive Server Enterprise

 CHAPTER 5 Parallel Query Processing

In this example, one consumer process reads data from a pipe (which is used
as a medium to transfer data across process boundaries) and passes the data to
the emit operator, which in turn routes the result to the client. The exchange
operator also spawns worker processes, which are called producer threads. The
exchange:emit operator writes the data into a pipe managed by the exchange
operator.
Figure 5-1: Binding of thread to plan fragments in query plan

Figure 5-1 shows the process boundary between a producer and a consumer
process. There are two plan fragments in this query plan. The plan fragment
with the scan and the exchange:emit operators are cloned three ways and then
a three-to-one exchange operator writes it into a pipe. The emit operator and the
exchange operator are run by a single process, which means there is a single
clone of that plan fragment.

Pipe management
The four types of pipes managed by the exchange operator are distinguished by
how they split and merge data streams. You can determine which type of pipe
is being managed by the exchange operator by looking at its description in the
showplan output, where the number of producers and consumers are shown.
The four pipe types are described below.

Performance and Tuning Series: Query Processing and Abstract Plans 149
Adaptive Server parallel query execution model

Many-to-one In this case, the exchange operator spawns multiple producer threads and has
one consumer task that reads the data from a pipe, to which multiple producer
threads write. The exchange operator in the previous example implements a
many-to-one exchange. A many-to-one exchange operator can be
order-preserving and this technique is employed particularly when doing a
parallel sort for an order by clause and the resultant data stream merged to
generate the final ordering. The showplan output shows more than one
producer process and one consumer process.
|EXCHANGE Operator (Merged)
|Executed in parallel by 3 Producer and 1
Consumer processes
One-to-many In this case, there is one producer and multiple consumer threads. The producer
thread writes data to multiple pipes according to a partitioning scheme devised
at query optimization, and then routes data to each of these pipes. Each
consumer thread reads data from one of the assigned pipes. This kind of data
split can preserve the ordering of the data. The showplan output shows one
producer process and more than one consumer processes:
|EXCHANGE Operator (Repartitioned)
|Executed in parallel by 1 Producer
and 4 Consumer processes
Many-to-many Many-to-many means there are multiple producers and multiple consumers.
Each producer writes to multiple pipes, and each pipe has multiple consumers.
Each stream is written to a pipe. Each consumer thread reads data from one of
the assigned pipes.
|EXCHANGE Operator (Repartitioned)
|Executed in parallel by 3 Producer and 4
Consumer processes
Replicated exchange In this case, the producer thread writes all of its data to each pipe that the
operators exchange operator configures. The producer thread makes a number of copies
of the source data (the number is specified by the query optimizer) equal to the
number of pipes in the exchange operator. Each consumer thread reads data
from one of the assigned pipes. The showplan output shows this as follows:
|EXCHANGE (Replicated)
|Executed in parallel by 3 Producers and 4
Consumer processes

150 Adaptive Server Enterprise

 CHAPTER 5 Parallel Query Processing

Worker process model

A parallel query plan is composed of different operators, at least one of which
is an exchange operator. At runtime, a parallel query plan is bound to a set of
server processes that, together, execute the query plan in a parallel fashion.
The server process associated with the user connection is called the alpha
process because it is the source process from which parallel execution is
initiated. In particular, each worker process involved in the execution of the
parallel query plan is spawned by the alpha process.
In addition to spawning worker processes, the alpha process initializes all the
worker processes involved in the execution of the plan, and creates and
destroys the pipes necessary for worker processes to exchange data. The alpha
process is, in effect, the global coordinator for the execution of a parallel query
plan.
At runtime, Adaptive Server associates each exchange operator in the plan with
a set of worker processes. The worker processes execute the query plan
fragment located immediately below the exchange operator.
For the query in Example A, represented in “EXCHANGE operator” on page
148, the exchange operator is associated with three worker processes. Each
worker process executes the plan fragment made of the exchange:emit operator
and of the scan operator.

Performance and Tuning Series: Query Processing and Abstract Plans 151
Adaptive Server parallel query execution model

Figure 5-2: Query execution plan with one exchange operator

titles (3 partitions)

Each exchange operator is also associated with a server process named the beta
process, which can be either the alpha process or a worker process. The beta
process associated with a given exchange operator is the local coordinator for
the execution of the plan fragment below the exchange operator. In the example
above, the beta process is the same process as the alpha process, because the
plan to be executed has only one level of exchange operators.
Next, use this query to illustrate what happens when the query plan contains
multiple exchange operators:
select count(*),pub_id, pub_date
from titles
group by pub_id, pub_date

152 Adaptive Server Enterprise

 CHAPTER 5 Parallel Query Processing

Figure 5-3: Query execution plan with two exchange operators

There are two levels of exchange operators marked as EXCHANGE-1 and

EXCHANGE-2 in Figure 5-3. Worker process T4 is the beta process
associated with the exchange operator EXCHANGE-2.
The beta process locally orchestrate execution of the plan fragment below the
exchange operator; it dispatches query plan information that is needed by the
worker processes, and synchronizes the execution of the plan fragment.
A process involved in the execution of a parallel query plan that is neither the
alpha process nor a beta process is called a gamma process.
A given parallel query plan is bound at runtime to a unique alpha process, to
one or more beta processes, and to at least one gamma process. Any Adaptive
Server parallel plan needs at least two different processes (alpha and gamma)
to be executed in parallel.
To find out the mapping between exchange operators and worker processes, as
well as to figure out which process is the alpha process, and which processes
are the beta processes, use dbcc traceon(516):

Performance and Tuning Series: Query Processing and Abstract Plans 153
Adaptive Server parallel query execution model

=======Thread to XCHg Map BEGINS=======

ALFA thread spid: 17
XCHG = 2 <- refers to Xchg-2
Comp Count = 2 Exec Count = 2
Range Adjustable
Consumer XCHG = 5
Parent thread spid: 34 <- refers to T4
Child thread 0: spid:37 <- refers to T1
Child thread 1: spid: 38 <- refers to T2
Child thread 2: spid: 36 <- refers to T3
Scheduling level: 0
XCHG = 5 <- refers to Xchg-1
Comp Count = 3 Exec Count = 3
Bounds Adjustable
Consumer XCHG -1
Parent thread spid: 17 <- refers to Alpha
Child thread 0: spid: 34 <- refers to T4
Child thread 1: spid: 35 <- refers to T5

Scheduling level: 0
=======Thread to XCHg Map BEGINS=======

Using parallelism in SQL operations

Partition tables or indexes in any way that best reflects the needs of your
application. Sybase recommends that you put partitions on segments that use
different physical disks so that enough I/O parallelism is present. For example,
you can have a well-defined partition based on hashing of certain columns of a
table, or certain ranges, or a list of values ascribed to a partition. Hash, range,
and list partitions belong to the category of “semantic-based” partitioning—
given a row, you can determine to which partition the row belongs.
Round-robin partitioning has no semantics associated with its partitioning. A
row can occur in any of its partitions. The choice of columns to partition and
the type of partitioning used can have a significant impact on the performance
of the application. Think of partitions as a low-cardinality index; the columns
on which partitioning must be defined are based on the queries in the
application.

154 Adaptive Server Enterprise

 CHAPTER 5 Parallel Query Processing

The query processing engine and its operators take advantage of the Adaptive
Server partitioning strategy. Partitioning defined on table and indexes is called
static partitioning. In addition, Adaptive Server dynamically repartitions data
to match the needs for relational operations like joins, vector aggregation,
distincts, unions, and so on. Repartitioning is done in streaming mode and no
storage is associated with it. Repartitioning is not the same as issuing the alter
table repartition command, where static repartitioning is done.

A query plan consists of query execution operators. In Adaptive Server,

operators belong to one of two categories:
• Attribute-insensitive operators include scans, union alls, and scalar
aggregation. Underlying partitions do not affect attribute-insensitive
operators.
• Attribute-sensitive operators (for example, join, distinct, union, and vector
aggregation) allow for an operation on a given amount of data to be broken
into a smaller number of operations on smaller fragments of the data using
semantics-based partitioning. Afterwards, a simple union all provides the
final result set. The union all is implemented using a many-to-one
exchange operator.

The following sections discuss these two classes of operators. The examples in
these sections use the following table with enough data to trigger parallel
processing.
create table RA2(a1 int, a2 int, a3 int)

Parallelism of attribute-insensitive operation

This section discusses the attribute-insensitive operations, which include scans
(serial and parallel), scalar aggregations, and union alls.

Table scan
For horizontal parallelism, either at least one of the tables in the query must be
partitioned, or the configuration parameter max repartition degree must be
greater than 1. If max repartition degree is set to 1, Adaptive Server uses the
number of online engines as a hint. When Adaptive Server runs horizontal
parallelism, it runs multiple versions of one or more operators in parallel. Each
clone of an operator works on its partition, which can be statically created or
dynamically built at execution.
Serial table scan The following example below shows the serial execution of a query where the
table RA2 is scanned using the table scan operator. The result of this operation
is routed to the emit operator, which forwards the result to the client.

Performance and Tuning Series: Query Processing and Abstract Plans 155
Adaptive Server parallel query execution model

select * from RA2

QUERY PLAN FOR STATEMENT 1 (at line 1).

1 operator(s) under root

The type of query is SELECT.

ROOT:EMIT Operator

|SCAN Operator
| FROM TABLE
| RA2
| Table Scan.
| Forward Scan.
| Positioning at start of table.
| Using I/O Size 2 Kbytes for data pages.
| With LRU Buffer Replacement Strategy for data
pages.
In versions earlier than 15.0, Adaptive Server did not try to scan an
unpartitioned table in parallel using a hash-based scan unless a force option was
used. Figure 5-4 shows a scan of an allpages-locked table executed in serial
mode by a single task T1. The task follows the page chain of the table to read
each page, while doing physical I/O if the needed pages are not in the cache.
Figure 5-4: Serial task scans data pages

Single page chain

7T1

Parallel table scan You can force a parallel table scan of an unpartitioned table using the Adaptive
Server force option. In this case, Adaptive Server uses a hash-based scan.

156 Adaptive Server Enterprise

 CHAPTER 5 Parallel Query Processing

Hash-based table Figure 5-5 shows how three worker processes divide the work of accessing
scans data pages from an allpages-locked table during a hash-based table scan. Each
worker process performs a logical I/O on every page, but each process
examines rows on one-third of the pages, as indicated by differently shaded
lines. Hash-based table scans are used only if the user forces a parallel degree.
See “Partition skew” on page 198.
With one engine, the query still benefits from parallel access because one work
process can execute while others wait for I/O. If there are multiple engines,
some of the worker processes can be running simultaneously.
Figure 5-5: Multiple worker processes scans unpartitioned table

WP1 Single Page Chain

WP2
WP3

Multiple worker processes

Hash-based scans increase the logical I/O for the scan, since each worker
process must access each page to hash on the page ID. For a data-only-locked
table, hash-based scans hash either on the extent ID or the allocation page ID,
so that only a single worker process scans a page and logical I/O does not
increase.
Partitioned-based If you partition this table as follows:
table scans
alter table RA2 partition by range(a1, a2)
(p1 values <= (500,100), p2 values <= (1000, 2000))
With the following query, Adaptive Server may choose a parallel scan of the
table. Parallel scan is chosen only if there are sufficient pages to scan and the
partition sizes are similar enough that the query will benefit from parallelism.
select * from RA2

QUERY PLAN FOR STATEMENT 1 (at line 1).

Executed in parallel by coordinating process and 2

Performance and Tuning Series: Query Processing and Abstract Plans 157
Adaptive Server parallel query execution model

worker processes.

3 operator(s) under root

The type of query is SELECT.

ROOT:EMIT Operator

|EXCHANGE Operator (Merged)

|Executed in parallel by 2 Producer and 1 Consumer
processes.

|
| |EXCHANGE:EMIT Operator
| |
| | |SCAN Operator
| | | FROM TABLE
| | | RA2
| | | Table Scan.
| | | Forward Scan.
| | | Positioning at start of table.
| | | Executed in parallel with a 2-way
partition scan.
| | | Using I/O Size 2 Kbytes for data pages.
| | | With LRU Buffer Replacement Strategy
for data pages.
After partitioning the table, showplan output includes two additional operators,
exchange and exchange:emit. This query includes two worker processes, each
of which scans a given partition and passes the data to the exchange:emit
operator, as illustrated in Figure 5-1 on page 149.
Figure 5-6 shows how a query scans a table that has three partitions on three
physical disks. With a single engine, this query can benefit from parallel
processing because one worker process can execute while others sleep, waiting
for I/O or waiting for locks held by other processes to be released. If multiple
engines are available, the worker processes can run simultaneously on multiple
engines. Such a configuration can perform extremely well.

158 Adaptive Server Enterprise

 CHAPTER 5 Parallel Query Processing

Figure 5-6: Multiple worker processes access multiple partitions

Index scan
Indexes, like tables, can be partitioned or unpartitioned. Local indexes inherit
the partitioning strategy of the table. Each local index partition scans data in
only one partition. Global indexes have a different partitioning strategy from
the base table; they reference one or more partitions.
Global nonclustered Adaptive Server supports global indexes that are nonclustered and
indexes unpartitioned for all table partitioning strategies. Global indexes are supported
for compatibility with Adaptive Server versions earlier than 15.0; they are also
useful in OLTP environments. The index and the data partitions can reside on
the same or different storage areas.
Noncovered scan of To create an unpartitioned global nonclustered index on table RA2, which is
global nonclustered partitioned by range, enter:
index using hashing
create index RA2_NC1 on RA2(a3)
This query has a predicate that uses the index key of a3:
select * from RA2 where a3 > 300
QUERY PLAN FOR STATEMENT 1 (at line 1).
. . . . . . . . . . . . . .
The type of query is SELECT.

ROOT:EMIT Operator

|EXCHANGE Operator (Merged)

Performance and Tuning Series: Query Processing and Abstract Plans 159
Adaptive Server parallel query execution model

|Executed in parallel by 3 Producer and 1

Consumer processes.
|
| |EXCHANGE:EMIT Operator
| |
| | |SCAN Operator
| | | FROM TABLE
| | | RA2
| | | Index : RA2_NC1
| | | Forward Scan.
| | | Positioning by key.
| | | Keys are:
| | | a3 ASC
| | | Executed in parallel with a 3-way
hash scan.
| | | Using I/O Size 2 Kbytes for index
leaf pages.
| | | With LRU Buffer Replacement Strategy
for index leaf pages.
| | | Using I/O Size 2 Kbytes for data
pages.
| | | With LRU Buffer Replacement Strategy
for data pages.
Adaptive Server uses an index scan using the index RA2_NC1 using three
producer threads spawned by the exchange operator. Each producer thread
scans all qualifying leaf pages and uses a hashing algorithm on the row ID of
the qualifying data and accesses the data pages to which it belongs. The
parallelism in this case is exhibited at the data page level.

160 Adaptive Server Enterprise

 CHAPTER 5 Parallel Query Processing

Figure 5-7: Hash-based parallel scan of global nonclustered index

Legend for Figure 5-7:

If the query does not need to access the data page, then it does not execute in
parallel. However, the partitioning columns must be added to the query;
therefore, it becomes a noncovered scan:
select a3 from RA2 where a3 > 300

Performance and Tuning Series: Query Processing and Abstract Plans 161
Adaptive Server parallel query execution model

QUERY PLAN FOR STATEMENT 1 (at line 1).

Executed in parallel by coordinating process and 2
worker processes.
3 operator(s) under root

The type of query is SELECT.

ROOT:EMIT Operator

|EXCHANGE Operator (Merged)

|Executed in parallel by 2 Producer and 1 Consumer
processes.

|
| |EXCHANGE:EMIT Operator
| |
| | |SCAN Operator
| | | FROM TABLE
| | | RA2
| | | Index : RA2_NC1
| | | Forward Scan.
| | | Positioning by key.
| | | Keys are:
| | | a3 ASC
| | | Executed in parallel with a 2-way hash
scan.
| | | Using I/O Size 2 Kbytes for index leaf
pages.
| | | With LRU Buffer Replacement Strategy for
index leaf pages.
| | | Using I/O Size 2 Kbytes for data pages.
| | | With LRU Buffer Replacement Strategy for
data pages.
Covered scan using If there is a nonclustered index that includes the partitioning column, there is
nonclustered global no reason for Adaptive Server to access the data pages and the query executes
index
in serial:
create index RA2_NC2 on RA2(a3,a1,a2)

select a3 from RA2 where a3 > 300

QUERY PLAN FOR STATEMENT 1 (at line 1).

1 operator(s) under root

The type of query is SELECT.

162 Adaptive Server Enterprise

 CHAPTER 5 Parallel Query Processing

ROOT:EMIT Operator

| SCAN Operator
| FROM TABLE
| RA2
| Index : RA2_NC2
| Forward Scan.
| Positioning by key.
| Index contains all needed columns. Base table
will not be read.
| Keys are:
| a3 ASC
| Using I/O Size 2 Kbytes for index leaf pages.
| With LRU Buffer Replacement Strategy for index
leaf pages.
Clustered index scans With a clustered index on an all-pages-locked table, a hash-based scan strategy
is not permitted. The only allowable strategy is a partitioned scan. Adaptive
Server uses a partitioned scan if that is necessary. For a data-only-locked table,
a clustered index is usually a placement index, which behaves as a nonclustered
index. All discussions pertaining to a nonclustered index on an all-pages-
locked table apply to a clustered index on a data-only-locked table as well.
Local indexes Adaptive Server supports clustered and nonclustered local indexes.
Clustered indexes on Local clustered indexes allow multiple threads to scan each data partition in
partitioned tables parallel, which can greatly improve performance. To take advantage of this
parallelism, use a partitioned clustered index. On a local index, data is sorted
separately within each partition. The information in each data partition
conforms to the boundaries established when the partitions were created, which
makes it possible to enforce unique index keys across the entire table.
Unique, clustered local indexes have the following restrictions:
• Index columns must include all partition columns.
• Partition columns must have the same order as the index definition's
partition key.
• Unique, clustered local indexes cannot be included on a round-robin table
with more than one partition.
Nonclustered indexes Adaptive Server supports local, nonclustered indexes on partitioned tables.
on partitioned tables
There is, however, a slight difference when using local indexes. When doing a
covered index scan of a local nonclustered index, Adaptive Server can still use
a parallel scan because the index pages are partitioned as well.

Performance and Tuning Series: Query Processing and Abstract Plans 163
Adaptive Server parallel query execution model

To illustrate the difference, this example creates a local nonclustered index:

create index RA2_NC2L on RA2(a3,a1,a2) local index

select a3 from RA2 where a3 > 300

QUERY PLAN FOR STATEMENT 1 (at line 1).
Executed in parallel by coordinating process and 2
worker processes.

3 operator(s) under root

The type of query is SELECT.

ROOT:EMIT Operator

|EXCHANGE Operator (Merged)

|Executed in parallel by 2 Producer and 1 Consumer
processes.

|
| |EXCHANGE:EMIT Operator
| |
| | |SCAN Operator
| | | FROM TABLE
| | | RA2
| | | Index : RA2_NC2L
| | | Forward Scan.
| | | Positioning by key.
| | | Index contains all needed columns. Base
table will not be read.
| | | Keys are:
| | | a3 ASC
| | | Executed in parallel with a 2-way
partition scan.
| | | Using I/O Size 2 Kbytes for index leaf
pages.
| | | With LRU Buffer Replacement Strategy
for index leaf pages.
Sometimes, Adaptive Server chooses a hash-based scan on a local index. This
occurs when a different parallel degree is needed or when the data in the
partition is skewed such that a hash-based parallel scan is preferred.

164 Adaptive Server Enterprise

 CHAPTER 5 Parallel Query Processing

Scalar aggregation
The Transact-SQL scalar aggregation operation can be done in serial or in
parallel.

Two-phased scalar aggregation

In a parallel scalar aggregation, the aggregation operation is performed in two
phases, using two scalar aggregate operators. In the first phase, the lower scalar
aggregation operator performs aggregation on the data stream. The result of
scalar aggregation from the first phase is merged using a many-to-one
exchange operator, and this stream is aggregated a second time.

In case of a count(*) aggregation, the second phase aggregation performs a

scalar sum. This is highlighted in the showplan output of the next example.
select count(*) from RA2
QUERY PLAN FOR STATEMENT 1 (at line 1).
Executed in parallel by coordinating process and 2
worker processes.

5 operator(s) under root

The type of query is SELECT.

ROOT:EMIT Operator

|SCALAR AGGREGATE Operator

| Evaluate Ungrouped SUM OR AVERAGE AGGREGATE.
|
| |EXCHANGE Operator (Merged)
| |Executed in parallel by 2 Producer and 1
Consumer processes.

| |
| | |EXCHANGE:EMIT Operator
| | |
| | | |SCALAR AGGREGATE Operator
| | | | Evaluate Ungrouped COUNT AGGREGATE.
| | | |
| | | | |SCAN Operator
| | | | | FROM TABLE
| | | | | RA2
| | | | | Table Scan.
| | | | | Forward Scan.
| | | | | Positioning at start of table.

Performance and Tuning Series: Query Processing and Abstract Plans 165
Adaptive Server parallel query execution model

| | | | | Executed in parallel with a

2-way partition scan.
| | | | | Using I/O Size 2 Kbytes for data
pages.
| | | | | With LRU Buffer Replacement
Strategy for data pages.

Serial aggregation
Adaptive Server may also choose to do the aggregation in serial. If the amount
of data to be aggregated is not enough to guarantee a performance advantage,
a serial aggregation may be the preferred technique. In case of a serial
aggregation, the result of the scan is merged using a many-to-one exchange
operator. This is shown in the example below, where a selective predicate has
been added to minimize the amount of data flowing into the scalar aggregate
operator. In such a case, it probably does not make sense to do the aggregation
in parallel.
select count(*) from RA2 where a2 = 10
QUERY PLAN FOR STATEMENT 1 (at line 1).
Executed in parallel by coordinating process and 2
worker processes.

4 operator(s) under root

The type of query is SELECT.

ROOT:EMIT Operator

|SCALAR AGGREGATE Operator

| Evaluate Ungrouped COUNT AGGREGATE.
|
| |EXCHANGE Operator (Merged)
| |Executed in parallel by 2 Producer
and 1 Consumer processes.

| |
| | |EXCHANGE:EMIT Operator
| | |
| | | |SCAN Operator
| | | | FROM TABLE
| | | | RA2
| | | | Table Scan.
| | | | Forward Scan.
| | | | Positioning at start of table.
| | | | Executed in parallel with a 2-way

166 Adaptive Server Enterprise

 CHAPTER 5 Parallel Query Processing

partition scan.
| | | | Using I/O Size 2 Kbytes for data
pages.
| | | | With LRU Buffer Replacement
Strategy for data pages.

union all
union all operators are implemented using a physical operator by the same
name. union all is a fairly simple operation and should be used in parallel only
when the query is moving a lot of data.

Parallel union all

The only condition to generating a parallel union all is that each of its operands
must be of the same degree, irrespective of the type of partitioning they have.
The following example (using table HA2) shows a union all operator being
processed in parallel. The position of the exchange operator above the union all
operator signifies that it is being processed by multiple threads:
create table HA2(a1 int, a2 int, a3 int)
partition by hash(a1, a2) (p1, p2)

select * from RA2

union all
select * from HA2

QUERY PLAN FOR STATEMENT 1 (at line 1).

Executed in parallel by coordinating process and 2
worker processes.

The type of query is SELECT.

ROOT:EMIT Operator

|EXCHANGE Operator (Merged)

|Executed in parallel by 2 Producer and 1 Consumer
processes.

|
| |EXCHANGE:EMIT Operator
| |
| | |UNION ALL Operator has 2 children.
| | |
| | | |SCAN Operator
| | | | FROM TABLE

Performance and Tuning Series: Query Processing and Abstract Plans 167
Adaptive Server parallel query execution model

| | | | RA2
| | | | Table Scan.
. . . . . . . . . . . . . . . . . . .
| | | | Executed in parallel with a 2-way
partition scan.
. . . . . . . . . . . . . . . . . . .
| | |
| | | |SCAN Operator
| | | | FROM TABLE
| | | | HA2
| | | | Table Scan.
. . . . . . . . . . . . . . . . . . .
| | | | Executed in parallel with a 2-way
partition scan.

Serial union all

In the next example, the data from each side of the union operator is restricted
by selective predicates on either side. The amount of data being sent through
the union all operator is small enough that Adaptive Server decides not to run
the unions in parallel. Instead, each scan of the tables RA2 and HA2 are
organized by putting 2-to-1 exchange operators on each side of the union. The
resultant operands are then processed in parallel by the union all operator:
select * from RA2
where a2 > 2400
union all
select * from HA2
where a3 in (10,20)
Executed in parallel by coordinating process and 4
worker processes.

7 operator(s) under root

The type of query is SELECT.

ROOT:EMIT Operator

|UNION ALL Operator has 2 children.

|
| |EXCHANGE Operator (Merged)
| |Executed in parallel by 2 Producer and 1
Consumer processes.

| |
| | |EXCHANGE:EMIT Operator

168 Adaptive Server Enterprise

 CHAPTER 5 Parallel Query Processing

| | |
| | | |SCAN Operator
| | | | FROM TABLE
| | | | RA2
| | | | Table Scan.
| | | | Executed in parallel with a 2-way
partition scan.
|
| |EXCHANGE Operator (Merged)
| |Executed in parallel by 2 Producer and 1
Consumer processes.

| |
| | |EXCHANGE:EMIT Operator
| | |
| | | |SCAN Operator
| | | | FROM TABLE
| | | | HA2
| | | | Table Scan.
| | | | Executed in parallel with a 2-way
partition scan.

Parallelism of attribute-sensitive operation

Attribute-sensitive operations include joins, vector aggregations, and unions.

join
If two tables are joined in parallel, Adaptive Server tries to use semantics-based
partitioning to make the join more efficient, depending on the amount of data
being joined and the type of partitioning that each of the operands have. If the
amount of data to be joined is small, but the number of pages to scan for each
of the tables is quite significant, Adaptive Server serializes the parallel streams
from each side and the join is done in serial mode. In this case, the query
optimizer determines that it is suboptimal to run a join operation in parallel. In
general, one or both of the operands used for the join operators may be any
intermediate operator, like another join or a grouping operator, but the examples
used show only scans as operands.
Tables with same The partitioning of each operand of a join is useful only with respect to the join
useful partitioning predicate. If two tables have the same partitioning, and the partitioning
columns are a subset of the join predicate, the tables are said to be
equipartitioned. For example, if you create another table, RB2, which is
partitioned similarly to that of RA2, using the following command:

Performance and Tuning Series: Query Processing and Abstract Plans 169
Adaptive Server parallel query execution model

create table RB2(b1 int, b2 int, b3 int)

partition by range(b1,b2)
(p1 values <= (500,100), p2 values <= (1000, 2000))
Then join RB2 with RA2; the scans and the join can be done in parallel without
additional repartitioning. Adaptive Server can join the first partition of RA2
with the first partition of RB2, then join the second partition of RA2 with the
second partition of RB2. This is called an equipartitioned join and is possible
only if the two tables join on columns a1, b1 and a2, b2 as shown below:
select * from RA2, RB2
where a1 = b1 and a2 = b2 and a3 < 0
QUERY PLAN FOR STATEMENT 1 (at line 1).
Executed in parallel by coordinating process and 2
worker processes.

7 operator(s) under root

The type of query is SELECT.

ROOT:EMIT Operator

|EXCHANGE Operator (Merged)

|Executed in parallel by 2 Producer
and 1 Consumer processes.

|
| |EXCHANGE:EMIT Operator
| |
| | |NESTED LOOP JOIN Operator
(Join Type: Inner Join)
| | |
| | | |RESTRICT Operator
| | | |
| | | | |SCAN Operator
| | | | | FROM TABLE
| | | | | RB2
| | | | | Table Scan.
| | | | | Forward Scan.
| | | | | Positioning at start of table.
| | | | | Executed in parallel with a
2-way partition scan.
| | |
| | | |RESTRICT Operator
| | | |
| | | | |SCAN Operator

170 Adaptive Server Enterprise

 CHAPTER 5 Parallel Query Processing

| | | | | FROM TABLE
| | | | | RA2
| | | | | Table Scan.
| | | | | Forward Scan.
| | | | | Positioning at start of table.
| | | | | Executed in parallel with a
2-way partition scan.
The exchange operator is shown above the nested-loop join. This implies that
exchange spawns two producer threads: the first scans the first partition of RA2
and RB2 and performs the nested-loop join; the second scans the second
partition of RA2 and RB2 to do the nested-loop join. The two threads merge the
results using a many-to-one (in this case, two-to-one) exchange operator.
One of the tables with In this example, the table RB2 is repartitioned to a three-way hash partitioning
useful partitioning on column b1 using the alter table command.
alter table RB2 partition by hash(b1) (p1, p2, p3)
Now, take a slightly modified join query as shown below:
select * from RA2, RB2 where a1 = b1
The partitioning on table RA2 is not useful because the partitioned columns are
not a subset of the joining columns (that is, given a value for the joining column
a1, you cannot specify the partition to which it belongs). However, the
partitioning on RB2 is helpful because it matches the joining column b1 of RB2.
In this case, the query optimizer repartitions table RA2 to match the partitioning
of RB2 by using hash partitioning on column a1 of RA2 (the joining column,
which is followed by a three-way merge join). The many-to-many (two-to-
three) exchange operator above the scan of RA2 does this dynamic
repartitioning. The exchange operator above the merge join operator merges the
result using a many-to-one (three-to-one, in this case) exchange operator. The
showplan output for this query is shown in the following example:
select * from RA2, RB2 where a1 = b1
QUERY PLAN FOR STATEMENT 1 (at line 1).
Executed in parallel by coordinating process and 5
worker processes.

10 operator(s) under root

The type of query is SELECT.

ROOT:EMIT Operator
|EXCHANGE Operator (Merged)
|Executed in parallel by 3 Producer and 1 Consumer

Performance and Tuning Series: Query Processing and Abstract Plans 171
Adaptive Server parallel query execution model

processes.

|
| |EXCHANGE:EMIT Operator
| |
| | |MERGE JOIN Operator (Join Type: Inner
Join)
| | | Using Worktable3 for internal storage.
| | | Key Count: 1
| | | Key Ordering: ASC
| | |
| | | |SORT Operator
| | | |Using Worktable1 for internal storage.
| | | |
| | | | |EXCHANGE Operator (Repartitioned)
| | | | |Executed in parallel by 2 Producer
and 3 Consumer processes.

| | | | |
| | | | | |EXCHANGE:EMIT Operator
| | | | | |
| | | | | | |RESTRICT Operator
| | | | | | |
| | | | | | | |SCAN Operator
| | | | | | | | FROM TABLE
| | | | | | | | RA2
| | | | | | | | Table Scan.
| | | | | | | | Forward Scan.
| | | | | | | | Positioning at start
of table.
| | | | | | | | Executed in parallel
with a 2-way
partition scan.
| | |
| | | |SORT Operator
| | | |Using Worktable2 for internal storage.
| | | |
| | | | |SCAN Operator
| | | | | FROM TABLE
| | | | | RB2
| | | | | Table Scan.
| | | | | Forward Scan.
| | | | | Positioning at start of table.
| | | | | Executed in parallel with a
3-way partition scan.

172 Adaptive Server Enterprise

 CHAPTER 5 Parallel Query Processing

Both tables with The next example uses a join where the native partitioning of the tables on both
useless partitioning sides is useless. The partitioning on table RA2 is on columns (a1,a2) and that
of RB2 is on (b1). The join predicate is on different sets of columns, and the
partitioning for both tables does not help at all. One option is to dynamically
repartition both sides of the join. By repartitioning table RA2 using a M-to-N
(two-to-three) exchange operator, Adaptive Server chooses column a3 of table
RA2 for repartitioning, as it is involved in the join with table RB2. For identical
reasons, table RB2 is also repartitioned three ways on column b3. The
repartitioned operands of the join are equipartitioned with respect to the join
predicate, which means that the corresponding partitions from each side will
join. In general, when repartitioning needs to be done on both sides of the join
operator, Adaptive Server employs a hash-based partitioning scheme.
select * from RA2, RB2 where a3 = b3
QUERY PLAN FOR STATEMENT 1 (at line 1).
Executed in parallel by coordinating process and 8
worker processes.

12 operator(s) under root

The type of query is SELECT.

ROOT:EMIT Operator

|EXCHANGE Operator (Merged)

|Executed in parallel by 3 Producer and 1 Consumer
processes.

|
| |EXCHANGE:EMIT Operator
| |
| | |MERGE JOIN Operator
(Join Type: Inner Join)
| | | Using Worktable3 for internal storage.
| | | Key Count: 1
| | | Key Ordering: ASC
| | |
| | | |SORT Operator
| | | |Using Worktable1 for internal
storage.
| | | |
| | | | |EXCHANGE Operator (Repartitioned)
| | | | |Executed in parallel by 2
Producer and 3 Consumer
processes.

Performance and Tuning Series: Query Processing and Abstract Plans 173
Adaptive Server parallel query execution model

| | | | |
| | | | | |EXCHANGE:EMIT Operator
| | | | | |
| | | | | | |RESTRICT Operator
| | | | | | |
| | | | | | | |SCAN Operator
| | | | | | | | FROM TABLE
| | | | | | | | RA2
| | | | | | | | Table Scan.
| | | | | | | | Forward Scan.
| | | | | | | | Positioning at
start of table.
| | | | | | | | Executed in
parallel with
a 2-way
partition scan.
| | |
| | | |SORT Operator
| | | |Using Worktable2 for internal
storage.
| | | |
| | | | |EXCHANGE Operator (Repartitioned)
| | | | |Executed in parallel by 3
Producer and 3 Consumer
processes.

| | | | |
| | | | | |EXCHANGE:EMIT Operator
| | | | | |
| | | | | | |SCAN Operator
| | | | | | | FROM TABLE
| | | | | | | RB2
| | | | | | | Table Scan.
| | | | | | | Forward Scan.
| | | | | | | Positioning at start
of table.
| | | | | | | Executed in parallel
with a 3-way
partition scan.
In general, all joins, including nested-loop, merge, and hash joins, behave in a
similar way. nested-loop joins display one exception, which is that the inner
side of a nested-loop join cannot be repartitioned. This limitation occurs
because, in the case of a nested-loop join, a column value for the joining
predicate is pushed from the outer side to the inner side.

174 Adaptive Server Enterprise

 CHAPTER 5 Parallel Query Processing

Replicated join A replicated join is useful when an index nested-loop join needs to be used.
Consider the case where a large table has a useful index on the joining column,
but useless partitioning, and joins to a small table that is either partitioned or
not partitioned. The small table can be replicated N ways to that of the inner
table, where N is the number of partitions of the large table. Each partition of
the large table is joined with the small table and, because no exchange operator
is needed on the inner side of the join, an index nested-loop join is allowed.
create table big_table(b1 int, b2 int, b3 int)
partition by hash(b3) (p1, p2)

create index big_table_nc1 on big_table(b1)

create table small_table(s1 int, a2 int, s3 int)

select * from small_table, big_table

where small_table.s1 = big_table.b1
QUERY PLAN FOR STATEMENT 1 (at line 1).
Executed in parallel by coordinating process and 3
worker processes.

7 operator(s) under root

The type of query is SELECT.

ROOT:EMIT Operator

|EXCHANGE Operator (Merged)

|Executed in parallel by 2 Producer and 1
Consumer processes.

|
| |EXCHANGE:EMIT Operator
| |
| | |NESTED LOOP JOIN Operator (Join Type:
Inner Join)
| | |
| | | |EXCHANGE Operator (Replicated)
| | | |Executed in parallel by 1 Producer
and 2 Consumer processes.

| | | |
| | | | |EXCHANGE:EMIT Operator
| | | | |
| | | | | |SCAN Operator
| | | | | | FROM TABLE

Performance and Tuning Series: Query Processing and Abstract Plans 175
Adaptive Server parallel query execution model

| | | | | | small_table
| | | | | | Table Scan.
| | |
| | | |SCAN Operator
| | | | FROM TABLE
| | | | big_table
| | | | Index : big_table_nc1
| | | | Forward Scan.
| | | | Positioning by key.
| | | | Keys are:
| | | | b1 ASC
| | | | Executed in parallel with a
2-way hash scan.
Parallel reformatting Parallel reformatting is especially useful when you are working with a
nested-loop join. Usually, reformatting refers to materializing the inner side of
a nested join into a worktable, then creating an index on the joining predicate.
With parallel queries and nested-loop join, reformatting is also helpful when
there is no useful index on the joining column or nested-loop join is the only
viable option for a query because of the server/session/query level settings.
This is an important option for Adaptive Server. The outer side may have
useful partitioning and, if not, it can be repartitioned to create that useful
partitioning. But for the inner side of a nested-loop join, any repartitioning
means that the table must be reformatted into a worktable that uses the new
partitioning strategy. The inner scan of a nested-loop join must then access the
worktable.
In this next example, partitioning for tables RA2 and RB2 is on columns (a1,
a2) and (b1, b2) respectively. The query is run with merge and hash join turned
off for the session.
select * from RA2, RB2 where a1 = b1 and a2 = b3

QUERY PLAN FOR STATEMENT 1 (at line 1).

Executed in parallel by coordinating process and 12
worker processes.

17 operator(s) under root

The type of query is SELECT.

ROOT:EMIT Operator

|SEQUENCER Operator has 2 children.

|
| |EXCHANGE Operator (Merged)

176 Adaptive Server Enterprise

 CHAPTER 5 Parallel Query Processing

| |Executed in parallel by 4 Producer

and 1 Consumer processes.

| |
| | |EXCHANGE:EMIT Operator
| | |
| | | |STORE Operator
| | | | Worktable1 created, in allpages
locking mode, for REFORMATTING.
| | | | Creating clustered index.
| | | |
| | | | |INSERT Operator
| | | | | The update mode is direct.
| | | | |
| | | | | |EXCHANGE Operator
(Repartitioned)
| | | | | |Executed in parallel by
2 Producer and 4
Consumer processes.
| | | | | |
| | | | | | |EXCHANGE:EMIT Operator
| | | | | | |
| | | | | | | |RESTRICT Operator
| | | | | | | |
| | | | | | | | |SCAN Operator
| | | | | | | | | FROM TABLE
| | | | | | | | | RB2
| | | | | | | | | Table Scan.
| | | | | | | | | Executed in
parallel
with a
2-way
partition
scan.
| | | | |
| | | | | TO TABLE
| | | | | Worktable1.
|
| |EXCHANGE Operator (Merged)
| |Executed in parallel by 4 Producer
and 1 Consumer processes.

| |
| | |EXCHANGE:EMIT Operator
| | |
| | | |NESTED LOOP JOIN Operator

Performance and Tuning Series: Query Processing and Abstract Plans 177
Adaptive Server parallel query execution model

(Join Type: Inner Join)

| | | |
| | | | |EXCHANGE Operator (Repartitioned)
| | | | |Executed in parallel by 2
Producer and 4 Consumer
processes.
| | | | |
| | | | | |EXCHANGE:EMIT Operator
| | | | | |
| | | | | | |RESTRICT Operator
| | | | | | |
| | | | | | | |SCAN Operator
| | | | | | | | FROM TABLE
| | | | | | | | RA2
| | | | | | | | Table Scan.
| | | | | | | | Executed in
parallel with
a 2-way
partition scan.
| | | | |SCAN Operator
| | | | | FROM TABLE
| | | | | Worktable1.
| | | | | Using Clustered Index.
| | | | | Forward Scan.
| | | | | Positioning by key.

The sequence operator executes all of its child operators but the last, before
executing the last child operator. In this case, the sequence operator executes
the first child operator, which reformats table RB2 into a worktable using a
four-way hash partitioning on columns b1 and b3. The table RA2 is also
repartitioned four ways to match the stored partitioning of the worktable.
Serial join Sometimes, it may not make sense to run a join in parallel because of the
amount of data that needs to be joined. If you run a query similar to that of the
earlier join queries, but now have predicates on each of the tables (RA2 and
RB2) such that the amount of data to be joined is not enough, the join may be
done in serial mode. In such a case, it does not matter how these tables are
partitioned. The query still benefits from scanning the tables in parallel.
select * from RA2, RB2 where a1=b1 and a2 = b2
and a3 = 0 and b2 = 20

QUERY PLAN FOR STATEMENT 1 (at line 1).

Executed in parallel by coordinating process and 4

178 Adaptive Server Enterprise

 CHAPTER 5 Parallel Query Processing

worker processes.

11 operator(s) under root

The type of query is SELECT.

ROOT:EMIT Operator

|MERGE JOIN Operator (Join Type: Inner Join)

| Using Worktable3 for internal storage.
| Key Count: 1
| Key Ordering: ASC
|
| |SORT Operator
| | Using Worktable1 for internal storage.
| |
| | |EXCHANGE Operator (Merged)
| | |Executed in parallel by 2 Producer and
1 Consumer processes.

| | |
| | | |EXCHANGE:EMIT Operator
| | | |
| | | | |RESTRICT Operator
| | | | |
| | | | | |SCAN Operator
| | | | | | FROM TABLE
| | | | | | RA2
| | | | | | Table Scan.
| | | | | | Executed in parallel with
a 2-way partition scan.
| |SORT Operator
| |Using Worktable2 for internal storage.
| |
| | |EXCHANGE Operator (Merged)
| | |Executed in parallel by 2 Producer and
1 Consumer processes.

| | |
| | | |EXCHANGE:EMIT Operator
| | | |
| | | | |RESTRICT Operator
| | | | |
| | | | | |SCAN Operator
| | | | | | FROM TABLE
| | | | | | RB2

Performance and Tuning Series: Query Processing and Abstract Plans 179
Adaptive Server parallel query execution model

| | | | | | Table Scan.
| | | | | | Executed in parallel with
a 2-way partition scan.
Semijoins Semijoins, which result from flattening of in/exist subqueries, behave the same
way as regular inner joins. However, replicated joins are not used for semijoins,
because an outer row can match more than one time in such a situation.
Outer joins In terms of parallel processing for outer joins, replicated joins are not
considered. Everything else behaves in a similar way as regular inner joins.
One other point of difference is that no partition elimination is done for any
table in an outer join that belongs to the outer group.

Vector aggregation
Vector aggregation refers to queries with group-bys. There are different ways
Adaptive Server can perform vector aggregation. The actual algorithms are not
described here; only the technique for parallel evaluation is shown in the
following sections.
In-partitioned vector If any base or intermediate relation requires a grouping and is partitioned on a
aggregation subset, or the same columns as that of the columns in the group by clause, the
grouping operation can be done in parallel on each of the partitions and the
resultant grouped streams merged using a simple N-to-1 exchange. This is
because a given group cannot appear in more than one stream. The same
restriction applies to grouping over any SQL query as long as you use
semantics-based partitioning on the grouping columns or a subset of them. This
method of parallel vector aggregation is called in-partitioned aggregation.
The following query uses a parallel in-partitioned vector aggregation since
range partitioning is defined on the columns a1 and a2, which also happens to
be the column on which the aggregation is needed.
select count(*), a1, a2 from RA2 group by a1,a2
QUERY PLAN FOR STATEMENT 1 (at line 1).
Executed in parallel by coordinating process and 2
worker processes.

4 operator(s) under root

The type of query is SELECT.

ROOT:EMIT Operator

|EXCHANGE Operator (Merged)

|Executed in parallel by 2 Producer and

180 Adaptive Server Enterprise

 CHAPTER 5 Parallel Query Processing

1 Consumer processes.

|
| |EXCHANGE:EMIT Operator
| |
| | |HASH VECTOR AGGREGATE Operator
| | | GROUP BY
| | | Evaluate Grouped COUNT AGGREGATE.
| | | Using Worktable1 for internal storage.
| | |
| | | |SCAN Operator
| | | | FROM TABLE
| | | | RA2
| | | | Table Scan.
| | | | Forward Scan.
| | | | Positioning at start of table.
| | | | Executed in parallel with a 2-way
partition scan.
| | | | Using I/O Size 2 Kbytes for data
pages.
| | | | With LRU Buffer Replacement
Strategy for data pages.
Repartitioned vector Sometimes, the partitioning of the table or the intermediate results may not be
aggregation useful for the grouping operation. It may still be worthwhile to do the grouping
operation in parallel by repartitioning the source data to match the grouping
columns, then applying the parallel vector aggregation. Such a scenario is
shown below, where the partitioning is on columns (a1, a2), but the query
requires a vector aggregation on column a1.
select count(*), a1 from RA2 group by a1

QUERY PLAN FOR STATEMENT 1 (at line 1).

Executed in parallel by coordinating process and 4
worker processes.

6 operator(s) under root

The type of query is SELECT.

ROOT:EMIT Operator

|EXCHANGE Operator (Merged)

|Executed in parallel by 2 Producer and 1 Consumer
processes.

Performance and Tuning Series: Query Processing and Abstract Plans 181
Adaptive Server parallel query execution model

|
| |EXCHANGE:EMIT Operator
| |
| | |HASH VECTOR AGGREGATE Operator
| | | GROUP BY
| | | Evaluate Grouped COUNT AGGREGATE.
| | | Using Worktable1 for internal storage.
| | |
| | | |EXCHANGE Operator (Repartitioned)
| | | |Executed in parallel by 2 Producer
and 2 Consumer processes.

| | | |
| | | | |EXCHANGE:EMIT Operator
| | | | |
| | | | | |SCAN Operator
| | | | | | FROM TABLE
| | | | | | RA2
| | | | | | Table Scan.
| | | | | | Forward Scan.
| | | | | | Positioning at start of
table.
| | | | | | Executed in parallel with
a 2-way partition scan.
Two-phased vector For the query in the previous example, repartitioning may be expensive.
aggregation Another possibility is to do a first level of grouping, merge the data using a
N-to-1 exchange operator, then do another level of grouping. This is called a
two-phased vector aggregation. Depending on the number of duplicates for the
grouping column, Adaptive Server can reduce the cardinality of the data
streaming through the N-to-1 exchange, which reduces the cost of the second
level of grouping.
select count(*), a1 from RA2 group by a1

QUERY PLAN FOR STATEMENT 1 (at line 1).

Executed in parallel by coordinating process and 2
worker processes.

5 operator(s) under root

The type of query is SELECT.

ROOT:EMIT Operator

|HASH VECTOR AGGREGATE Operator

182 Adaptive Server Enterprise

 CHAPTER 5 Parallel Query Processing

| GROUP BY
| Evaluate Grouped SUM OR AVERAGE AGGREGATE.
| Using Worktable2 for internal storage.
|
| |EXCHANGE Operator (Merged)
| |Executed in parallel by 2 Producer and
1 Consumer processes.
| |
| | |EXCHANGE:EMIT Operator
| | |
| | | |HASH VECTOR AGGREGATE Operator
| | | | GROUP BY
| | | | Evaluate Grouped COUNT AGGREGATE.
| | | | Using Worktable1 for internal
storage.
| | | |
| | | | |SCAN Operator
| | | | | FROM TABLE
| | | | | RA2
| | | | | Table Scan.
| | | | | Executed in parallel with
a 2-way partition scan.
Serial vector As with some of the earlier examples, if the amount of data flowing into the
aggregation grouping operator is restricted by using a predicate, executing that query in
parallel may not make much sense. In such a case, the partitions are scanned in
parallel and an N-to-1 exchange operator is used to serialize the stream
followed by a serial vector aggregation:
select count(*), a1, a2 from RA2
where a1 between 100 and 200
group by a1, a2

QUERY PLAN FOR STATEMENT 1 (at line 1).

Executed in parallel by coordinating process and
2 worker processes.

4 operator(s) under root

The type of query is SELECT.

ROOT:EMIT Operator

|HASH VECTOR AGGREGATE Operator

| GROUP BY
| Evaluate Grouped COUNT AGGREGATE.

Performance and Tuning Series: Query Processing and Abstract Plans 183
Adaptive Server parallel query execution model

| Using Worktable1 for internal storage.

|
| |EXCHANGE Operator (Merged)
| |Executed in parallel by 2 Producer and 1
Consumer processes.
| |
| | |EXCHANGE:EMIT Operator
| | |
| | | |SCAN Operator
| | | | FROM TABLE
| | | | RA2
| | | | Positioning at start of table.
| | | | Executed in parallel with a 2-way
partition scan.
You cannot always group on the partitioning columns, or take advantage of a
table that is already partitioned on the grouping columns. The query optimizer
determines if it is better to repartition and perform the grouping in parallel, or
merge the data stream in a partitioned table and do the grouping in serial or a
two-phased aggregation.
distinct Queries with distinct operations are the same as grouped vector aggregation
without the aggregation part. For example:
select distinct a1, a2 from RA2
is same as:
select a1, a2 from RA2 group by a1, a2
All of the methodologies that are applicable to vector aggregates are applicable
here as well.
Queries with an in list Adaptive Server uses an optimized technique to handle an in list. This is a
common SQL construct. So, a construct like:
col in (value1, value2,..valuek)
is same as:
col = value1 OR col = value2 OR .... col = valuek
The values in the in list are put into a special in-memory table and sorted for
removal of duplicates.The table is then joined back with the base table using an
index nested-loop join. The following example illustrates this with two values
in the in list that correspond to two values in the or list:
SCAN Operator
FROM OR List
OR List has up to 2 rows of OR/IN values.

184 Adaptive Server Enterprise

 CHAPTER 5 Parallel Query Processing

select * from RA2 where a3 in (1425, 2940)

QUERY PLAN FOR STATEMENT 1 (at line 1).

Executed in parallel by coordinating process and 2
worker processes.

6 operator(s) under root

The type of query is SELECT.

ROOT:EMIT Operator

|EXCHANGE Operator (Merged)

|Executed in parallel by 2 Producer and 1
Consumer processes.

|
| |EXCHANGE:EMIT Operator
| |
| | |NESTED LOOP JOIN Operator (Join Type:
Inner Join)
| | |
| | | |SCAN Operator
| | | | FROM OR List
| | | | OR List has up to 2 rows of OR/IN
values.
| | |
| | | |RESTRICT Operator
| | | |
| | | | |SCAN Operator
| | | | | FROM TABLE
| | | | | RA2
| | | | | Index : RA2_NC1
| | | | | Forward Scan.
| | | | | Positioning by key.
| | | | | Keys are:
| | | | | a3 ASC
| | | | | Executed in parallel with a
2-way hash scan.

Performance and Tuning Series: Query Processing and Abstract Plans 185
Adaptive Server parallel query execution model

Queries with or Adaptive Server takes a disjunctive predicate like an or clause and applies each
clauses side of the disjunction separately to qualify a set of row IDs (RIDs). The set of
conjunctive predicates on each side of the disjunction must be indexable. Also,
the conjunctive predicates on each side of the disjunction cannot have further
disjunction within them; that is, it makes little sense to use an arbitrarily deep
nesting of disjunctive and conjunctive clauses. In the next example, a
disjunctive predicate is taken on the same column (you can have predicates on
different columns as long as you have indexes that can do inexpensive scans),
but the predicates may qualify an overlapping set of data rows. Adaptive Server
uses the predicates on each side of the disjunction separately and qualifies a set
of row IDs. These row IDs are then subjected to duplicate elimination.
select a3 from RA2 where a3 = 2955 or a3 > 2990

QUERY PLAN FOR STATEMENT 1 (at line 1).

Executed in parallel by coordinating process and 2
worker processes.

8 operator(s) under root

The type of query is SELECT.

ROOT:EMIT Operator

|EXCHANGE Operator (Merged)

|Executed in parallel by 2 Producer and 1
Consumer processes.

|
| |EXCHANGE:EMIT Operator
| |
| | |RID JOIN Operator
| | | Using Worktable2 for internal storage.
| | |
| | | |HASH UNION Operator has 2 children.
| | | | Using Worktable1 for internal storage.
| | | |
| | | | |SCAN Operator
| | | | | FROM TABLE
| | | | | RA2
| | | | | Index : RA2_NC1
| | | | | Forward Scan.
| | | | | Positioning by key.
| | | | | Index contains all needed
columns.Base table will not
be read.

186 Adaptive Server Enterprise

 CHAPTER 5 Parallel Query Processing

| | | | | Keys are:
| | | | | a3 ASC
| | | | | Executed in parallel with a
2-way hash scan.
| | | |
| | | | |SCAN Operator
| | | | | FROM TABLE
| | | | | RA2
| | | | | Index : RA2_NC1
| | | | | Forward Scan.
| | | | | Positioning by key.
| | | | | Index contains all needed
columns. Base table will
not be read.
| | | | | Keys are:
| | | | | a3 ASC
| | | | | Executed in parallel with a
2-way hash scan.
| | | |RESTRICT Operator
| | | |
| | | | |SCAN Operator
| | | | | FROM TABLE
| | | | | RA2
| | | | | Using Dynamic Index.
| | | | | Forward Scan.
| | | | | Positioning by Row IDentifier
(RID.)
| | | | | Using I/O Size 2 Kbytes for
data pages.
| | | | | With LRU Buffer Replacement
Strategy for data pages.
Two separate index scans are employed using the index RA2_NC1, which is
defined on the column a3. The qualified set of row IDs are then checked for
duplicate row IDs, and finally, joined back to the base table. Note the line
Positioning by Row Identifier (RID). You can use different indexes
for each side of the disjunction, depending on what the predicates are, as long
as they are indexable. One way to easily identify this is to run the query
separately with each side of the disjunction to make sure that the predicates are
indexable. Adaptive Server may not choose an index intersection if it seems
more expensive than a single scan of the table.

Performance and Tuning Series: Query Processing and Abstract Plans 187
Adaptive Server parallel query execution model

Queries with an order If a query requires sorted output because of the presence of an order by clause,
by clause Adaptive Server can apply the sort in parallel. First, Adaptive Server tries to
avoid the sort if there is some inherent ordering available. If Adaptive Server
is forced to do the sort, it sees if the sort can be done in parallel. To do that,
Adaptive Server may repartition an existing data stream or it may use the
existing partitioning scheme, then apply the sort to each of the constituent
streams. The resultant data is merged using an N-to-1 order, preserving the
exchange operator.
select * from RA2 order by a1, a2

QUERY PLAN FOR STATEMENT 1 (at line 1).

Executed in parallel by coordinating process and 2
worker processes.

4 operator(s) under root

The type of query is SELECT.

ROOT:EMIT Operator

|EXCHANGE Operator (Merged)

|Executed in parallel by 2 Producer and
1 Consumer processes.

|
| |EXCHANGE:EMIT Operator
| |
| | |SORT Operator
| | | Using Worktable1 for internal storage.
| | |
| | | |SCAN Operator
| | | | FROM TABLE
| | | | RA2
| | | | Index : RA2_NC2L
| | | | Forward Scan.
| | | | Positioning at index start.
| | | | Executed in parallel with a
2-way partition scan.
Depending upon the volume of data to be sorted, and the available resources,
Adaptive Server may repartition the data stream to a higher degree than the
current degree of the stream, so that the sort operation is faster. The degrees of
sorting depends on whether the benefit obtained from doing the sort in parallel
far outweighs the overheads of repartitioning.

188 Adaptive Server Enterprise

 CHAPTER 5 Parallel Query Processing

Subqueries
Adaptive Server uses different methods to reduce the cost of processing
subqueries. Parallel optimization depends on the type of subquery:
• Materialized subqueries – parallel query methods are not considered for
the materialization step.
• Flattened subqueries – parallel query optimization is considered only
when the subquery is flattened to a regular inner join or a semijoin.
• Nested subqueries – parallel operations are considered for the outermost
query block in a query containing a subquery; the inner, nested queries
always execute serially. This means that all tables in nested subqueries are
accessed serially. In the following example, the table RA2 is accessed in
parallel, but the result is that the table is serialized using a two-to-one
exchange operator before accessing the subquery. The table RB2 inside the
subquery is accessed in parallel.
select count(*) from RA2 where not exists
(select * from RB2 where RA2.a1 = b1)

QUERY PLAN FOR STATEMENT 1 (at line 1).

Executed in parallel by coordinating process and 2
worker processes.

8 operator(s) under root

The type of query is SELECT.

ROOT:EMIT Operator

|SCALAR AGGREGATE Operator

| Evaluate Ungrouped COUNT AGGREGATE.
|
| |SQFILTER Operator has 2 children.
| |
| | |EXCHANGE Operator (Merged)
| | |Executed in parallel by 2 Producer
and 1 Consumer processes.

| | |
| | | |EXCHANGE:EMIT Operator
| | | |
| | | | |RESTRICT Operator
| | | | |

Performance and Tuning Series: Query Processing and Abstract Plans 189
Adaptive Server parallel query execution model

| | | | | |SCAN Operator
| | | | | | FROM TABLE
| | | | | | RA2
| | | | | | Index : RA2_NC2L
| | | | | | Forward Scan.
| | | | | | Executed in parallel with
a 2-way partition scan.
| |
| | Run subquery 1 (at nesting level 1).
| |
| | QUERY PLAN FOR SUBQUERY 1 (at nesting
level 1 and at line 2).
| |
| | Correlated Subquery.
| | Subquery under an EXISTS predicate.
| |
| | |SCALAR AGGREGATE Operator
| | | Evaluate Ungrouped ANY AGGREGATE.
| | | Scanning only up to the first
qualifying row.
| | |
| | | |SCAN Operator
| | | | FROM TABLE
| | | | RB2
| | | | Table Scan.
| | | | Forward Scan.
| |
| | END OF QUERY PLAN FOR SUBQUERY 1.
The following example shows an in subquery flattened into a semijoin.
Adaptive Server converts this into an inner join to provide greater flexibility in
shuffling the tables in the join order. As seen below, the table RB2, which was
originally in the subquery, is now being accessed in parallel.
select * from RA2 where a1 in (select b1 from RB2)

QUERY PLAN FOR STATEMENT 1 (at line 1).

Executed in parallel by coordinating process and 5
worker processes.

10 operator(s) under root

The type of query is SELECT.

ROOT:EMIT Operator

190 Adaptive Server Enterprise

 CHAPTER 5 Parallel Query Processing

|EXCHANGE Operator (Merged)

|Executed in parallel by 3 Producer and 1 Consumer
processes.

|
| |EXCHANGE:EMIT Operator
| |
| | |MERGE JOIN Operator (Join Type: Inner Join)
| | | Using Worktable3 for internal storage.
| | | Key Count: 1
| | | Key Ordering: ASC
| | |
| | | |SORT Operator
| | | | Using Worktable1 for internal
storage.
| | | |
| | | | |SCAN Operator
| | | | | FROM TABLE
| | | | | RB2
| | | | | Table Scan.
| | | | | Executed in parallel with a
3-way partition scan.
| | |
| | | |SORT Operator
| | | | Using Worktable2 for internal
storage.
| | | |
| | | | |EXCHANGE Operator (Merged)
| | | | |Executed in parallel by 2
Producer and 3 Consumer
processes.

| | | | |
| | | | | |EXCHANGE:EMIT Operator
| | | | | |
| | | | | | |RESTRICT Operator
| | | | | | |
| | | | | | | |SCAN Operator
| | | | | | | | FROM TABLE
| | | | | | | | RA2
| | | | | | | | Index : RA2_NC2L
| | | | | | | | Forward Scan.
| | | | | | | | Positioning at
index start.
| | | | | | | | Executed in
parallel with

Performance and Tuning Series: Query Processing and Abstract Plans 191
Adaptive Server parallel query execution model

a 2-way
partition scan.

select into clauses

Queries with select into clauses create a new table in which to store the query’s
result set. Adaptive Server optimizes the base query portion of a select into
command in the same way it does a standard query, considering both parallel
and serial access methods. A select into statement that is executed in parallel:
• Creates the new table using the columns specified in the select into
statement.
• Creates N partitions in the new table, where N is the degree of parallelism
that the optimizer chooses for the insert operation in the query.
• Populates the new table with query results, using N worker processes.
• Unpartitions the new table, if no specific destination partitioning is
required.
Performing a select into statement in parallel requires more steps than an
equivalent serial query plan. This is a simple select into done in parallel:
select * into RAT2 from RA2

QUERY PLAN FOR STATEMENT 1 (at line 1).

Executed in parallel by coordinating process and 2
worker processes.

4 operator(s) under root

The type of query is INSERT.

ROOT:EMIT Operator

|EXCHANGE Operator (Merged)

|Executed in parallel by 2 Producer and 1
Consumer processes.

|
| |EXCHANGE:EMIT Operator
| |
| | |INSERT Operator
| | | The update mode is direct.
| | |
| | | |SCAN Operator

192 Adaptive Server Enterprise

 CHAPTER 5 Parallel Query Processing

| | | | FROM TABLE
| | | | RA2
| | | | Table Scan.
| | | | Forward Scan.
| | | | Positioning at start of table.
| | | | Executed in parallel with a 2-way
partition scan.
| | |
| | | TO TABLE
| | | RAT2
| | | Using I/O Size 2 Kbytes for data
pages.
Adaptive Server does not try to increase the degree of the stream coming from
the scan of table RA2, and uses it to do a parallel insert into the destination
table. The destination table is initially created using round-robin partitioning of
degree two. After the insert, the table is unpartitioned.
If the data set to be inserted is not big enough, Adaptive Server may choose to
insert this data in serial. The scan of the source table can still be done in
parallel. The destination table is then created as an unpartitioned table.
The select into allows destination partitioning to be specified. In such a case,
the destination table is created using that partitioning, and Adaptive Server
finds the most optimal way to insert data. If the destination table must be
partitioned the same way as the source data, and there is enough data to insert,
the insert operator executes in parallel.
The next example shows the same partitioning for source and destination table,
and demonstrates that Adaptive Server recognizes this scenario and chooses
not to repartition the source data.
select * into new_table
partition by range(a1, a2)
(p1 values <= (500,100), p2 values <= (1000, 2000))
from RA2

QUERY PLAN FOR STATEMENT 1 (at line 1).

Executed in parallel by coordinating process and 2
worker processes.

4 operator(s) under root

The type of query is INSERT.

ROOT:EMIT Operator

Performance and Tuning Series: Query Processing and Abstract Plans 193
Adaptive Server parallel query execution model

|EXCHANGE Operator (Merged)

|Executed in parallel by 2 Producer and 1 Consumer
processes.

|
| |EXCHANGE:EMIT Operator
| |
| | |INSERT Operator
| | | The update mode is direct.
| | |
| | | |SCAN Operator
| | | | FROM TABLE
| | | | RA2
| | | | Table Scan.
| | | | Forward Scan.
| | | | Positioning at start of table.
| | | | Executed in parallel with a 2-way
partition scan.
| | |
| | | TO TABLE
| | | RRA2
| | | Using I/O Size 16 Kbytes for data
pages.
If the source partitioning does not match that of the destination table, the source
data must be repartitioned. This is illustrated in the next example, where the
insert is done in parallel using two worker processes after the data is
repartitioned using a 2-to-2 exchange operator that converts the data from
range partitioning to hash partitioning.
select * into HHA2
partition by hash(a1, a2)
(p1, p2)
from RA2

QUERY PLAN FOR STATEMENT 1 (at line 1).

Executed in parallel by coordinating process and 4
worker processes.

6 operator(s) under root

The type of query is INSERT.

ROOT:EMIT Operator

194 Adaptive Server Enterprise

 CHAPTER 5 Parallel Query Processing

|EXCHANGE Operator (Merged)

|Executed in parallel by 2 Producer and 1
Consumer processes.

|
| |EXCHANGE:EMIT Operator
| |
| | |INSERT Operator
| | | The update mode is direct.
| | |
| | | |EXCHANGE OperatorEXCHANGE Operator (
Merged)
| | | |Executed in parallel by 2 Producer
and 2 Consumer processes.

| | | |
| | | | |EXCHANGE:EMIT Operator
| | | | |
| | | | | |SCAN Operator
| | | | | | FROM TABLE
| | | | | | RA2
| | | | | | Table Scan.
| | | | | | Forward Scan.
| | | | | | Positioning at start of table.
| | | | | | Executed in parallel with a
2-way partition scan.
| | |
| | | TO TABLE
| | | HHA2
| | | Using I/O Size 16 Kbytes for data
pages.

insert/delete/update
insert, delete, and update operations are done in serial in Adaptive Server.
However, tables other than the destination table used in the query to qualify
rows to be deleted or updated, can be accessed in parallel.
delete from RA2
where exists
(select * from RB2
where RA2.a1 = b1 and RA2.a2 = b2)

QUERY PLAN FOR STATEMENT 1 (at line 1).

Executed in parallel by coordinating process and 3
worker processes.

Performance and Tuning Series: Query Processing and Abstract Plans 195
Adaptive Server parallel query execution model

9 operator(s) under root

The type of query is DELETE.

ROOT:EMIT Operator

|DELETE Operator
| The update mode is deferred.
|
| |NESTED LOOP JOIN Operator (Join Type: Inner Join)
| |
| | |SORT Operator
| | | Using Worktable1 for internal storage.
| | |
| | | |EXCHANGE Operator (Merged)
| | | |Executed in parallel by 3 Producer
and 1 Consumer processes.

| | | |
| | | | |EXCHANGE:EMIT Operator
| | | | |
| | | | | |RESTRICT Operator
| | | | | |
| | | | | | |SCAN Operator
| | | | | | | FROM TABLE
| | | | | | | RB2
| | | | | | | Table Scan.
| | | | | | | Forward Scan.
| | | | | | | Positioning at start of
table.
| | | | | | | Executed in parallel with
a 3-way partition scan.
| | | | | | | Using I/O Size 2 Kbytes
for data pages.
| | | | | | | With LRU Buffer Replacement
Strategy for data pages.
| |
| | |RESTRICT Operator
| | |
| | | |SCAN Operator
| | | | FROM TABLE
| | | | RA2
| | | | Index : RA2_NC1
| | | | Forward Scan.
| | | | Positioning by key.

196 Adaptive Server Enterprise

 CHAPTER 5 Parallel Query Processing

| | | | Keys are:
| | | | a3 ASC
|
| TO TABLE
| RA2
| Using I/O Size 2 Kbytes for data pages.
The table RB2, which is being deleted, is scanned and deleted in serial.
However, table RA2 was scanned in parallel. The same scenario is true for
update or insert statements.

Partition elimination
One of the advantages of semantic partitioning is that the query processor may
be able to take advantage of this and be able to disqualify range, hash, and list
partitions at compile time. With hash partitions, only equality predicates can be
used, whereas for range and list partitions, equality and in-equality predicates
can be used to eliminate partitions. For example, consider table RA2 with its
semantic partitioning defined on columns a1, a2 where (p1 values <=
(500,100) and p2 values <= (1000, 2000)). If there are predicates on columns
a1 or columns a1, a2, then it would be possible to do some partition
elimination. For example, this statement does not qualify any data:
select * from RA2 where a1 > 1500
You can see this in the showplan output.
QUERY PLAN FOR STATEMENT 1 (at line 1).
................................
| | |SCAN Operator
| | | FROM TABLE
| | | RA2
| | | [ Eliminated Partitions : 1 2 ]
| | | Index : RA2_NC2L
The phrase Eliminated Partitions identifies the partition in accordance
with how it was created and assigns an ordinal number for identification. For
table RA2, the partition represented by p1 where (a1, a2) <= (500, 100) is
considered to be partition number one and p2 where (a1, a2) > (500, 100) and
<= (1000, 2000) is identified as partition number two.
Consider an equality query on a hash-partitioned table where all keys in the
hash partitioning have an equality clause. This can be shown by taking table
HA2, which is hash-partitioned two ways on columns (a1, a2). The ordinal
numbers refer to the order in which partitions are listed in the output of sp_help.

Performance and Tuning Series: Query Processing and Abstract Plans 197
Adaptive Server parallel query execution model

select * from HA2 where a1 = 10 and a2 = 20

QUERY PLAN FOR STATEMENT 1 (at line 1).

................................

|SCAN Operator
| FROM TABLE
| HA2
| [ Eliminated Partitions : 1 ]
| Table Scan.

Partition skew
Partition skew plays an important part in determining whether a parallel
partitioned scan can be used. Adaptive Server partition skew is defined as the
ratio of the size of the largest partition to the average size of a partition.
Consider a table with four partitions of sizes 20, 20, 35, and 80 pages. The size
of the average partition is (20 + 20 + 35 + 85)/4 = 40 pages. The biggest
partition has 85 pages so partition skew is calculated as 85/40 = 2.125. In
partitioned scans, the cost of doing a parallel scan is as expensive as doing the
scan on the largest partition. Instead, a hash-based partition may turn out to be
fast, as each worker process may hash on a page number or an allocation unit
and scan its portion of the data. The penalty paid in terms of loss of
performance by skewed partitions is not always at the scan level, but rather as
more complex operators like several join operations are built over the data. The
margin of error increases exponentially in such cases.
Run sp_help on a table to see the partition skews:
sp_help HA2

........
name type partition_type partitions partition_keys
------ -------------------- -------------- -----------
HA2 base table hash 2 a1, a2

partition_name partition_id pages segment

create_date
-------------------------- ------------ ----------- --
------------
--------------------------
HA2_752002679 752002679 324 default
Aug 10 2005 2:05PM
HA2_768002736 768002736 343 default

198 Adaptive Server Enterprise

 CHAPTER 5 Parallel Query Processing

Aug 10 2005 2:05PM

Partition_Conditions
--------------------
NULL

Avg_pages Max_pages Min_pages Ratio(Max/Avg)

Ratio(Min/Avg)
----------- ----------- ----------- ------------------
---------
---------------------------
333 343 324 1.030030
0.972973
Alternatively, you can calculate skew by querying the systabstats system
catalog, where the number of pages in each partition is listed.

Why queries do not run in parallel

Adaptive Server runs a query in serial when:
• There is not enough data to benefit from parallel access.
• The query contains no equijoin predicates like:
select * from RA2, RB2
where a1 > b1
• There are not enough resources, such as thread or memory, to run a query
in parallel.
• Uses a covered scan of a global nonclustered index.
• Tables and indexes are accessed inside a nested subquery that cannot be
flattened.

Runtime adjustment
If there are not enough worker processes available at runtime, the execution
engine attempts to reduce the number of worker processes used by the
exchange operators present in the plan:

Performance and Tuning Series: Query Processing and Abstract Plans 199
Adaptive Server parallel query execution model

• First, by attempting to reduce the worker process usage of certain

exchange operators in the query plan without resorting to serial
recompilation of the query. Depending on the semantics of the query plan,
certain exchange operators are adjustable and some are not. Some are
limited in the way they can be adjusted.
• Parallel query plans need a minimum number of worker processes to run.
When enough worker processes are not available, the query is recompiled
serially. When recompilation is impossible, the query is aborted and the
appropriate error message is generated.
It does so in two ways:
Adaptive Server supports serial recompilation for all:
• Ad hoc select queries, except select into, alter table, and execute immediate
queries.
• Stored procedures, except select into and alter table queries.

Recognizing and managing runtime adjustments

Adaptive Server provides two mechanisms to help you observe runtime
adjustments of query plans:
• set process_limit_action allows you to abort batches or procedures when
runtime adjustments take place.
• showplan prints an adjusted query plan when runtime adjustments occur,
and showplan is effect.

Using set process_limit_action

Use process_limit_action with the set command to monitor the use of adjusted
query plans at a session or stored procedure level. When you set
process_limit_action to “abort,” Adaptive Server records error 11015 and aborts
the query, if an adjusted query plan is required. When you set
process_limit_action to “warning,” Adaptive Server records error 11014 but still
executes the query. For example, this command aborts the batch when a query
is adjusted at runtime:
set process_limit_action abort

200 Adaptive Server Enterprise

 CHAPTER 5 Parallel Query Processing

By examining the occurrences of errors 11014 and 11015 in the error log, you
can determine the degree to which Adaptive Server uses adjusted query plans
instead of optimized query plans. To remove the restriction and allow runtime
adjustments, use:
set process_limit_action quiet
See set in the Reference Manual: Commands.

Using showplan
When you use showplan, Adaptive Server displays the optimized plan for a
given query before it runs the query. When the query plan involves parallel
processing, and a runtime adjustment is made, showplan displays this message,
followed by the adjusted query plan:
AN ADJUSTED QUERY PLAN IS BEING USED FOR STATEMENT 1
BECAUSE NOT ENOUGH WORKER PROCESSES ARE CURRENTLY
AVAILABLE.

ADJUSTED QUERY PLAN:

When using set notexec, Adaptive Server does not attempt to execute a query,
so runtime plans are never displayed.

Reducing the likelihood of runtime adjustments

To reduce the number of runtime adjustments, increase the number of worker
processes that are available to parallel queries. You can do this either by adding
more total worker processes to the system or by restricting or eliminating
parallel execution for noncritical queries, using either:
• set parallel_degree to set session-level limits on the degree of parallelism,
or
• The query-level parallel 1 and parallel N clauses to limit the worker process
usage of individual statements.
To reduce the number of runtime adjustments for system procedures, recompile
the procedures after changing the degree of parallelism at the server or session
level. See sp_recompile in Adaptive Server Reference Manual: Procedures.

Performance and Tuning Series: Query Processing and Abstract Plans 201
Adaptive Server parallel query execution model

202 Adaptive Server Enterprise

CH A PTE R 6 Eager and Lazy Aggregation

This chapter discusses eager and lazy aggregation in Adaptive Server.

Topic Page
Overview 203
Aggregation and query processing 205
Examples 208
Using eager aggregation 215

Overview
Aggregate processing is one of the most useful operations in DBMS
environments. It summarizes large amounts of data with an aggregated
value, including:
• The minimum, maximum, sum, or average value of a column in a
specified set of rows
• The count of rows that match a condition
• Other statistical functions
In SQL, aggregate processing is performed using the aggregation
functions min(), max(), count(), sum(), and avg(), and group by and having
clauses. The SQL language implements two aggregate processing types,
vector aggregation and scalar aggregation. A select-project-join (SPJ)
query illustrates these two types of aggregate processing:
select r1, s1
from r, s
where r2 = s2
Vector aggregation In vector aggregation, the SPJ result set is grouped on the group by clause
expressions, and then the select clause aggregation functions are applied
to each group. The query produces one result row per group:
select r1, sum (s1)
from r, s

Performance and Tuning Series: Query Processing and Abstract Plans 203
Overview

where r2 = s2
group by r1
Scalar aggregation In scalar aggregation, there is no group by clause and the entire SPJ result set is
aggregated, as a single group, by the same select clause aggregate functions.
The query produces a single result row:
select sum (s1)
from r, s
where r2 = s2

Eager aggregation
Eager aggregation transforms the internal representation of queries such as
those discussed above, and processes them as if the aggregation is performed
incrementally: first locally, over each table, producing intermediate aggregate
results over smaller local subgroups, and then globally after the join, thus
combining the local aggregation results to produce the final result set.
These queries, which return the same result set over any data set, are derived
table SQL-level rewrites of the vector and scalar aggregation examples above.
They illustrate the eager aggregation transformations that Adaptive Server
performs on the internal representation of the queries.
Vector aggregation select r1, sum(sum_s1 * count_r)
from
(select
r1,r2,
count_r = count(*)
from r
group by r1,r2
)gr
,
(select
s2,
sum_s1 = sum(s1)
from s
group by s2
)gs
where r2-s2
group by r1
Scalar aggregation select sum(sum_s1 * count_r)
from
(select
r2,

204 Adaptive Server Enterprise

 CHAPTER 6 Eager and Lazy Aggregation

count_r = count(*)
from r
group by r2
)gr
,
(select
s2,
sum_s1 = sum(s1)
from s
group by s2
)gs
where r2 = s2
Eager aggregation plans are generated and costed by the optimizer, and can be
chosen as the best plan. This form of advanced query optimization can result
in orders of magnitude of performance gain, relative to the original SQL query.

Aggregation and query processing

From the perspective of query processing (QP), aggregation can be both a
costly operation, and an operation whose placement has an important impact
on query performance.
• Aggregation generally computes an aggregated value. Vector aggregation
is costly, compared to scalar aggregation, as rows must be grouped
together to obtain the aggregated result over a group; this implies, in
general, a reordering of the rows through sorting or hashing, both costly
operations.
• Aggregating after applying cardinality-reducing operators (such as filters)
to the input set reduces the cost of aggregation and can thus improve
overall query performance.
• Aggregating before applying cardinality-increasing operators (such as
joins and unions) to the input set reduces the cost of aggregation and can
thus improve overall query performance.
• Aggregating early can reduce the costs of parent operators through
reducing their input set cardinality, and can thus improve overall query
performance.
• Aggregation can dramatically reduce the cardinality of the input set in the
result set, when the grouping columns have relatively few distinct value
combinations.

Performance and Tuning Series: Query Processing and Abstract Plans 205
Aggregation and query processing

• Some properties of the aggregation’s input set, as already grouped (for

example, when the aggregation is ordered on the grouping columns),
reduce the cost of vector aggregation; in scalar aggregation, rows ordered
on the aggregated column allow computing a min or max without
accessing each input row.
• Plan fragment physical properties have a big impact on aggregation cost.
The naive QP implementation of aggregation places the scalar or vector
aggregate operator, as indicated by the SQL query, over the SPJ part of its
query block. However, there are algebraic transformations that preserve the
semantics of the query and allow aggregation at other places in the operators
tree:
• Pushing the aggregation down toward the leaves, to aggregate early
(called eager aggregation).
• Pulling the aggregation up toward the root, to aggregate late (called lazy
aggregation).
Plans obtained through such transformations differ greatly in performance.
More importantly to distributed query processing (DQP), the cardinality of
intermediate results can be greatly reduced by eager aggregation. Such orders-
of-magnitude cardinality reduce cross-node data transfer cost, thus removing
the main shortcoming of DQP as opposed to traditional QP.
Adaptive Server 15.0.2 and later implements eager aggregation over the leaves
of a query plan, which means over the scan operators.
This query illustrates the QP implications of eager aggregation:
select r1, sum(s1)
from r,s
where r2 = s2
group by r1

206 Adaptive Server Enterprise

 CHAPTER 6 Eager and Lazy Aggregation

Figure 6-1: Typical query execution plan

GroupHashing: r1; sum(s1)

MergeJoin: r2=s2

IndScan: ir2, r IndScan: is2, s

The two index scans, on r(r2) and s(s2) provide the orderings needed by the
“r2=s2” merge join. Hash-based grouping is done over the join, as the query
specifies it.
The optimizer also generates query plans that perform eager aggregation, also
called the push-down of grouping, early grouping, or eager grouping. The SQL
representation of the transform using derived tables is:
select r1, sum(sum_s1 * cnt_r)
from
(select r1, r2, cnt_r = count(*)
from r
group by r1, r2
) as gr
,
(select s2, sum_s1 = sum(s1)
from s
group by s2
) as gs
where r2 = s2
group by r1

Performance and Tuning Series: Query Processing and Abstract Plans 207
Examples

Figure 6-2: Possible eager aggregation plan

GroupHashing: r1; sum(sum_s1*cnt_r)

MergeJoin: r2=s2

GroupSorted: r1, r2; GroupSorted: s2;

cnt_r = count(*) cnt_s1 = sum(s1)

IndScan: ir21, r IndScan: is2, s

The two eager GroupSorted operators group on the local grouping columns.
GroupSorted operators apply to any column projected out for a reason other
than that it is an aggregation function argument. These columns include:
• The main grouping columns in the group by clause
• Columns needed by predicates not yet applied
To place the cheap GroupSorted operator, the child plan fragment must provide
ordering on all the local grouping columns; hence the ir21 index on r(r2,
r1).

Examples
Online data archiving The most compelling reason to implement eager aggregation is online data
archiving, which is a distributed query processing (DQP) installation where
recent OLTP read-write data is on an Adaptive Server and historical read-only
data is on another server, either Adaptive Server or ASIQ.
The following view, v, offers decision support system (DSS) applications
transparent access to local Adaptive Server data in ase_tab and, through the
Component Integration Services (CIS) proxy_asiq_tab, to remote historical
data on an ASIQ server.

208 Adaptive Server Enterprise

 CHAPTER 6 Eager and Lazy Aggregation

create view v(v1, v2)

as
select a1, a2 from ase_tab
union all
select q1, q2 from proxy_asiq_tab
The DSS applications ignore the distributed nature of the data and use such
union-in-view tables as the base tables of their complex queries, typically using
aggregation:
select t1, sum(v1)
from t,v
where t2=v2
group by t1
After view and union resolution, the following operator tree is obtained:
Figure 6-3: SQL query rewrite

group:t1;sum(v1)

join:t2=v2

t union

ase_tab proxy_aseq_tab

As this tree uses a CIS proxy table, the CIS layer uses a specialized remote scan
operator to generate and ship a plan fragment to the remote site.

Performance and Tuning Series: Query Processing and Abstract Plans 209
Examples

Figure 6-4: Suboptimal classical CIS behavior

ASIQ
GROUP "select
q1, q2
ASE from
asiq_tab"
JOIN

scan: t union

scan: scan:
ase_tab proxy_asiq_t
scan:
asiq_tab

As such, this mechanism is suboptimal: the entire history table is shipped

through the CIS layer to the Adaptive Server side, incurring a large network
cost; furthermove, the advanced ASIQ bitmap-based grouping algorithms are
not used.

210 Adaptive Server Enterprise

 CHAPTER 6 Eager and Lazy Aggregation

Figure 6-5: Classical processing of aggregation

Grouping done here, on the ASE side,

group according to the semantic of the query.

join Intermediate operators:

• Block the grouping pushdown
• Need to switch with grouping
Scan: t union

ASIQ
Scan: ase tab Scan: proxy_asiq_tab

Grouping needed here, on the ASIQ side:

• ASIQ uses bitmap
• Less result data

Ideally, transformations are performed on the operators tree and grouping on

the ASIQ side, so that only aggregated data is transferred.

Performance and Tuning Series: Query Processing and Abstract Plans 211
Examples

Figure 6-6: Desired aggregation processing layout

Grouping completed here, according to
group the semantic of the query.

join Intermediate operators that preserve

? the semantics

Scan: t union
group
ASIQ
Scan: ase tab
Scan: proxy_asiq_tab

Eager grouping that preserves

the semantics

In this example, there are two operators between the group and the CIS proxy:
a join and a union. The next transform pushes grouping below the join and the
union, achieving eager aggregation:

212 Adaptive Server Enterprise

 CHAPTER 6 Eager and Lazy Aggregation

Figure 6-7: Eager aggregation

group

join Eager Aggregation achieved

Scan: t union

group group

Scan: ase tab Scan: proxy_asiq_tab

Grouping is now adjacent to the CIS proxy. The CIS layer can now send a
grouped query to ASIQ and return aggregated data.

Performance and Tuning Series: Query Processing and Abstract Plans 213
Examples

Figure 6-8: Optimal CIS behavior with eager aggregation

ASE select q2, ASIQ

sum(q1)
group from
asiq_tab
join group by q2
group
scan:t union
scan:
group NetScan: asiq_tab
ASIQ
Scan:
ase_tab
DSS/DQP The efficient execution of complex aggregated DSS queries in a distributed
environment is a challenge not only in online data archivings; it is, in general,
a generic DSS/DQP problem in online data archiving DSS/DQP:
• A typical query involves complex joins and unions, and aggregation is
performed at the top of the query tree.
• The data is distributed across the nodes, and the intermediate results must
be shipped from a producer node to a consumer node for further
processing.
Single-node DSS Although the examples above are for DQP in general, and online data archiving
in particular, the eager aggregation performance impact goes beyond shipping
intermediate results between DQP nodes.
Eager aggregation enhances the performance of aggregated complex queries
by reducing intermediate result sets. Since aggregated complex queries are
typical, eager aggregation enhances Adaptive Server performance in all DSS
applications.

214 Adaptive Server Enterprise

 CHAPTER 6 Eager and Lazy Aggregation

Using eager aggregation

Eager aggregation is an internal query processing feature. You need not change
anything at the SQL level when you enable eager aggregation; queries that use
aggregation have eager aggregation-based plans automatically enumerated and
costed by the optimizer.

Enabling eager aggregation

Eager aggregation is controlled by the advanced_aggregation optimizer setting,
which is off by default in all optimization goals except allrows_dss, where it is
on. Eager aggregation can be enabled, disabled, or reset to the optimizer goal’s
default value at either the connection or query level.
For example, to enable at the connection level:
set advanced_aggregation on
To enable at the query level:
select r1, sum (s1)
from r, s
where r2 = s2
group by r1
plan
"(use advanced_aggregation on)"

Alternatively, if the optimization goal is set to allrows_dss, eager aggregation

is implicitly enabled. In this example, an abstract plan sets allrows_dss at the
query level:
select r1, sum (s1)
from r, s
where r2 = s2
group by r1
plan
"(use optgoal allrows_dss)"

Checking for eager aggregation

When eager aggregation is enabled, the optimizer determines cost, depending
on whether the estimated cheapest plan uses eager aggregation or not.

Performance and Tuning Series: Query Processing and Abstract Plans 215
Using eager aggregation

Output from the showplan aggregation:

1> select r1, sum(s1)
2> from r, s
3> where r2=s2
4> group by r1
5> go
QUERY PLAN FOR STATEMENT 1 (at line 1).
STEP 1
The type of query is SELECT.
6 operator(s) under root
|ROOT:EMIT Operator
|
| |HASH VECTOR AGGREGATE Operator
| | GROUP BY
| | Evaluate Grouped SUM OR AVERAGE AGGREGATE.
| | Using Worktable2 for internal storage.
| | Key Count: 1
| |
| | |MERGE JOIN Operator (Join Type: Inner Join)
| | | Using Worktable1 for internal storage.
| | | Key Count: 1
| | | Key Ordering: ASC
| | |
| | | |GROUP SORTED Operator
| | | | Evaluate Grouped COUNT AGGREGATE.
| | | |
| | | | |SCAN Operator
| | | | | FROM TABLE
| | | | | r
| | | | | Index : ir21
| | | | | Forward Scan.
| | | | | Positioning at index start.
| | | | | Index contains all needed columns. Base table will not be read.
| | | | | Using I/O Size 2 Kbytes for index leaf pages.
| | | | | With LRU Buffer Replacement Strategy for index leaf pages.
| | |
| | | |GROUP SORTED Operator
| | | | Evaluate Grouped SUM OR AVERAGE AGGREGATE.
| | | |
| | | | |SCAN Operator
| | | | | FROM TABLE
| | | | | s
| | | | | Index : is21
| | | | | Forward Scan.
| | | | | Positioning at index start.

216 Adaptive Server Enterprise

 CHAPTER 6 Eager and Lazy Aggregation

| | | | | Index contains all needed columns. Base table will not be read.
| | | | | Using I/O Size 2 Kbytes for index leaf pages.
| | | | | With LRU Buffer Replacement Strategy for index leaf pages.
r1
----------- -----------
1 2
2 4
(2 rows affected)
As the query performs vector aggregation over the join of r and s, the hash
vector aggregate operator at the top of the query tree is expected in all cases.
However, the group sorted operators over the scans of r and of s are not part of
the query; they perform the eager aggregation.
When advanced_aggregation is off, the plan does not contain the eager
aggregation operators group sorted:
1> set advanced_aggregation off
2> go
1> select r1, sum(s1)
2> from r, s
3> where r2=s2
4> group by r1
5> go

QUERY PLAN FOR STATEMENT 1 (at line 1).

STEP 1
The type of query is SELECT.
4 operator(s) under root
|ROOT:EMIT Operator
|
| |HASH VECTOR AGGREGATE Operator
| | GROUP BY
| | Evaluate Grouped SUM OR AVERAGE AGGREGATE.
| | Using Worktable2 for internal storage.
| | Key Count: 1
| |
| | |MERGE JOIN Operator (Join Type: Inner Join)
| | | Using Worktable1 for internal storage.
| | | Key Count: 1
| | | Key Ordering: ASC
| | |
| | | |SCAN Operator
| | | | FROM TABLE
| | | | r
| | | | Index : ir21
| | | | Forward Scan.

Performance and Tuning Series: Query Processing and Abstract Plans 217
Using eager aggregation

| | | | Positioning at index start.

| | | | Index contains all needed columns. Base table will not be read.
| | | | Using I/O Size 2 Kbytes for index leaf pages.
| | | | With LRU Buffer Replacement Strategy for index leaf pages.
| | |
| | | |SCAN Operator
| | | | FROM TABLE
| | | | s
| | | | Index : is21
| | | | Forward Scan.
| | | | Positioning at index start.
| | | | Index contains all needed columns. Base table will not be read.
| | | | Using I/O Size 2 Kbytes for index leaf pages.
| | | | With LRU Buffer Replacement Strategy for index leaf pages.
r1
----------- -----------
1 2
2 4
(2 rows affected)

Forcing eager aggregation with abstract plans

The optimizer opportunistically enumerates the cheap GroupSorted-based
eager aggregation plans when the child plan fragment provides an ordering on
the local grouping columns.
This limitation avoids increasing the optimization search space and time.
However, in some cases hash-based eager aggregation produces the cheapest
plan. abstract plans can be used in such cases to force eager aggregation.
advanced_grouping must be enabled to use such an abstract plan; otherwise the
eager aggregation abstract plan is rejected.
In the example above, if r has no index on (r1, r2), and if r is large but has few
r1--r2 distinct pairs of values, a hash join with eager grouping over r is the best
plan, forced by this abstract plan:
1> select r1, sum(s1)
2> from r, s
3> where r2=s2
4> group by r1
5> plan
6> "(group_hashing
7> (h_join
8> (group_hashing
9> (t_scan r)

218 Adaptive Server Enterprise

 CHAPTER 6 Eager and Lazy Aggregation

10> )
11> (t_scan s)
12> )
13> )"
14> go

QUERY PLAN FOR STATEMENT 1 (at line 1).

Optimized using the Abstract Plan in the PLAN clause.
STEP 1
The type of query is SELECT.
5 operator(s) under root
|ROOT:EMIT Operator
|
| |HASH VECTOR AGGREGATE Operator
| | GROUP BY
| | Evaluate Grouped SUM OR AVERAGE AGGREGATE.
| | Using Worktable3 for internal storage.
| | Key Count: 1| |
| | |HASH JOIN Operator (Join Type: Inner Join)
| | | Using Worktable2 for internal storage.
| | | Key Count: 1
| | |
| | | |HASH VECTOR AGGREGATE Operator
| | | | GROUP BY
| | | | Evaluate Grouped COUNT AGGREGATE.
| | | | Using Worktable1 for internal storage.
| | | | Key Count: 2
| | | |
| | | | |SCAN Operator
| | | | | FROM TABLE
| | | | | r
| | | | | Table Scan.
| | | | | Forward Scan.
| | | | | Positioning at start of table.
| | | | | Using I/O Size 2 Kbytes for data pages.
| | | | | With LRU Buffer Replacement Strategy for data pages.
| | |
| | | |SCAN Operator
| | | | FROM TABLE
| | | | s
| | | | Table Scan.
| | | | Forward Scan.
| | | | Positioning at start of table.
| | | | Using I/O Size 2 Kbytes for data pages.
| | | | With LRU Buffer Replacement Strategy for data pages.
r1

Performance and Tuning Series: Query Processing and Abstract Plans 219
Using eager aggregation

----------- -----------
1 2
2 4

(2 rows affected)
The hash vector aggregate operator eagerly aggregates the scan of r, as
requested by the abstract plan.

220 Adaptive Server Enterprise

CH A PTE R 7 Controlling Optimization

This chapter describes query processing options that affect the query
processor’s choice of join order, index, I/O size, and cache strategy.
Topic Page
Special optimizing techniques 221
Specifying query processor choices 231
Specifying table order in joins 232
Specifying the number of tables considered by the query processor 233
Specifying query index 234
Specifying I/O size in a query 236
Specifying cache strategy 239
Controlling large I/O and cache strategies 241
Asynchronous log service 241
Enabling and disabling merge joins 244
Enabling and disabling join transitive closure 245
Controlling literal parameterization 246
Suggesting a degree of parallelism for a query 248
Concurrency optimization for small tables 258

Special optimizing techniques

Sybase recommends that, before using the tools discussed in this chapter,
you read the Performance and Tuning Series: Basics. It will help you
understand the material in this chapter.
Use the optimization techniques with caution, as they allow you to
override the decisions made by the Adaptive Server query processor and
if misused, can have an extremely negative effect on performance. You
should understand the impact on the performance of both your individual
query and the possible implications for overall system performance.

Performance and Tuning Series: Query Processing and Abstract Plans 221
Viewing current optimizer settings

In most situations, Adaptive Server advanced, cost-based query processor

produces excellent query plans. However, there are times when the query
processor does not choose the proper index for optimal performance, or
chooses a suboptimal join order, and you must control the access methods for
the query. The optimization techniques allow you to take that control.
In addition, while you are tuning, you may want to see the effects of a different
join order, I/O size, or cache strategy. Some of the optimization options let you
specify query processing or access strategy without costly reconfiguration.
Adaptive Server provides tools and query clauses that affect query
optimization and advanced query analysis tools that let you understand why the
query processor makes the choices that it does.

Note This chapter suggests workarounds for certain optimization problems. If

the workarounds do not adequately address these problems, call Sybase
Technical Support.

Viewing current optimizer settings

sp_options allows you to view the current optimizer settings for these options:

• set plan dump / load

• set plan exists check

• set forceplan

• set plan optgoal

• set [optCriteria]

• set plan opttimeoutlimit

• set plan replace

• set statistics simulate

• set metrics_capture

• set prefetch

• set parallel_degree number

• set process_limit_action

222 Adaptive Server Enterprise

 CHAPTER 7 Controlling Optimization

• set resource_granularity number

• set scan_parallel_degree number

• set repartition_degree number

sp_options queries the sysoptions fake table, which stores information about
each set option, its category, and its current and default settings. sysoptions also
contains a bitmap that provides detailed information for each option
The syntax for sp_options is:
sp_options [ [show | help
[, option_name | category_name |null
[, dflt | non_dflt | null
[, spid] ] ] ] ]
where:
• show – lists the current and default values of all options, grouped
according to their category. Issuing sp_options show with an option name
specified shows you the current and default value for the individual option.
You can also specify a session ID, and whether you want to view options
with default settings or options with nondefault settings.
• help – show usage information. Achieve the same result by issuing
sp_options with no parameters.

• null – indicates the option for which you want to view the settings.

• dflt | non_dflt | null – indicates whether to show options with default settings
or to show options with non-default settings.
• spid – specifies the session ID. Use the session ID to view other session
settings.
For example, to display the current optimizer settings shown below, enter:
sp_options show
Category: Query Tuning
name
currentsetting defaultsetting scope
--------------------------------------------------------------------
------------------ --------------------- ---------
optlevel
ase_default ase_current 3
optgoal
allrows_mix allrows_mix 3
opttimeoutlimit
10 10 2
repartition_degree

Performance and Tuning Series: Query Processing and Abstract Plans 223
Viewing current optimizer settings

1 1 2
scan_parallel_degree
0 1 2
resource_granularity
10 10 2

. . .

outer_join_costing: outer join row counts and histogramming

0 0 7
join_duplicate_estimates: avoid overestimates of dups in joins
0 0 7
imdb_costing: 0 PIO costing for scans for in-memory database
1 1 7
auto_temptable_stats: auto generation of statistics for #temptables
0 0 7
use_mixed_dt_sarg_under_specialor: allow special OR in case of mixed . . .
0 0 7
timeout_cart_product: timeout queries involving cartesian product and more . . .
0 0 7

(81 rows affected)

See Adaptive Server Reference Manual: Procedures.

Any user can query sysoptions:
You can also use string manipulation or a cast. For example, if an option is
numeric, you can query sysoptions by entering:
if (isnumeric(currentsetting))
select@int_val = convert(int, currentsetting)
...
else
select@char_val = currentsetting
...
For more infomation about sysoptions, see Adaptive Server Reference Manual:
Tables.

224 Adaptive Server Enterprise

 CHAPTER 7 Controlling Optimization

Setting the optimization level

By default, Adaptive Server does not enable some performance related
optimizer settings, and you must enable them using the set command. Because
these optimizer settings are not enabled, any applications already running
efficiently are not effected by, nor benefit from, optimizer changes when
upgrading to the latest version of Adaptive Server.
Adaptive Server allows you to set the optimization level globally (that is, the
optimization levels are set across the server) and at the session level. Enabling
these changes increases query performance for many applications, though
Sybase recommends additional performance testing.
Use the @@optlevel global variable to determine the current optimization
level settings:
select @@optlevel
The optimization settings are organized according to which optimization
changes are available for each Adaptive Server release. Table 7-1 describes the
settings:
Table 7-1: Opimization level
Parameter Description
ase_current Enables all optimizer changes through the current release
ase_default Disables all optimizer changes since version 1503 ESD #1
ase1503esd2 Enables all optimizer changes through version 15.0.3 ESD #2
ase1503esd3 Enables all optimizer changes through version 15.0.3 ESD #3

These optimization level criteria are enabled by default:

Performance and Tuning Series: Query Processing and Abstract Plans 225
Viewing current optimizer settings

Table 7-2: Optimization criteria enabled by default

Setting Description
cr421607 Support NULL=NULL merge and hash join keys
cr467566 Allow abstract plans and statement caches to work together
cr487450 Improves distinct costing for multitable outer joins and
semijoins
cr497066 Infer the nullability of isnull by observing its parameters
cr500736 Support nocase sort order columns in merge join and hash
join keys
cr531199 Increase the number of useful nested loop join plans the
optimizer considereds
cr534175 Compute group by worktables in nested subqueries only
once, when possible
cr544485 Mark subquery join predicates with distinct view as SARGs
cr545059 Reduce buffer manager optimization sort usage
cr545180 Avoid reformating with no SARGs if a useful index exists
cr545379 Disallow reformatting on user-forced index scan
cr545585 Covered iscan CPU costing too expensive
cr545653 Avoid inner table buffer estimate starvation
cr545771 Improve multi-table outer join and semijoin costing
cr546125 Allow a nonunique index scan for implicitly updatable
cursors
cr552795 Eliminate unnecessary duplicate rows during reformatting
cr562947 Allow cursor table scans
data_page_prefetch_ Adds clustered row bias
costing
mru_buffer_costing Wash size buffer limit for MRU

These optimization level criteria are enabled when you enable ase1503esd2:

226 Adaptive Server Enterprise

 CHAPTER 7 Controlling Optimization

Table 7-3: Optimization criteria enabled with ase1503esd2

Setting Description
cr556728 Facilitate merge joins between small tables
cr559034 Avoid preferring non-covering over covered index
scans
allow_wide_top_sort Allow top sorts to exceed max row size
avoid_bmo_sorts Avoid sorts used only for buffer manager
optimization
conserve_tempdb_space Keep estimated temporary databases below
resource granularity
distinct_exists_transform Transform distinct to semi-join
join_duplicate_estimates Avoid overestimates of join duplicates
outer_join_costing Outer join row counts and histogramming
search_engine_timeout_factor Open cursor command takes a long time with a
complex select statement
timeout_cart_product Timeout queries involving cartesian product and
more than 5 tables.

Table 7-4 lists optimization criteria that are enabled when you enable
ase1503esd3 or ase_current

Table 7-4: Optimization criteria enabled with ase1503esd3 and

ase_current
Setting Description
auto_template_stats Automatically generate statistics for
temporary tables
use_mixed_dt_sarg_under_specialor Allow special or for mixed datatype SARGs
in an in or or list

These optimization criteria are off by default:

Table 7-5: Optimization criteria that are disabled by default
Setting Description
full_index_filter Eliminate noncovered full index scan strategies
no_stats_distinctness Allow duplicate estimates without statistics

Set the optimization and criteria levels using one of these methods:
• At the session level – use set plan optlevel to set the optimization level for
the current session. This configures the session to use the all optimization
changes up through the current version of Adaptive Server:
set plan optlevel ase_current

Performance and Tuning Series: Query Processing and Abstract Plans 227
Viewing current optimizer settings

• For individual logins – use sp_modifylogin to set the optimization level for
logins. sp_modifylogin calls a user-created stored procedure the defines the
optimization level. For example, if you create this stored procedure to
enable the ase1503esd2 optimization level, but disable the optimization
level for cr545180:
create proc login_proc
as
set plan optlevel ase1503esd2
set cr545180 off
go
You may apply these optimization settings to any login. This applies the
settings from login_proc to user joe:
sp_modifylogin joe, 'login script', login_proc
• Across the server – use the sp_configure “optimizer level” parameter to set
the optimization level globally. This sets enables all optimizer changes up
to the current version of Adaptive Server:
sp_configure 'optimizer level', 0, 'ase_current'
• Within an abstract plan – use the optlevel abstract plan to set the optimizer
level. This example enables all optimizer changes up to the current version
of Adaptive Server for the current plan:
select name from sysdatabases
plan '(use optlevel ase_current)'
• During a session with the set command – use the set command to change
the optimizer criteria level for the current session. Optimizer criteria
changes may improve the performance for some queries while
deteriorating others. You may find better performance by applying the
optimizer criteria level at a fine grain level (perhaps at the query level).
Adaptive Server denotes some optimizer criteria levels by their CR
numbers, while other more recent optimizer changes are specified with
descriptive names. Use sp_options to view the list of available options in
the current release. This enables the optimization criteria for CR 545180:

228 Adaptive Server Enterprise

 CHAPTER 7 Controlling Optimization

set CR545180 on

Note When you enable an optimization criteria, Adaptive Server retains

the previous optimization level.

Optimizer Diagnostic Utility

The sp_opt_querystats system procedure lets you analyze the query plan
generated by the Adaptive Server optimizer and the factors that influenced its
choice of a query plan. This analysis may help determine if elements in the
query or the execution environment affect how Adaptive Server executes the
query and its performance. You need not run the selected query to perform the
analysis.
sp_opt_querystats output includes:

• The query plan generated by showplan

• Enabled trace flags and switches
• I/O activity for the query generated by set statistics io
• Missing statistics found for any of the tables involved in the query
• The estimated plan cost calculated by the optimizer
• The final plan and cost estimations calculated by the optimizer
• The abstract plan for the query
• The result of the query if the result set is executed (for example, if noexec
is not on)
• The logical operator tree for the query generated by set option show
• Query execution time generated by set statistics time
• After you execute the query, the query execution time generated by set
statistics time

Performance and Tuning Series: Query Processing and Abstract Plans 229
Optimizer Diagnostic Utility

Configuring Adaptive Server to run sp_opt_querystats

1 Install the Job Scheduler. See the Job Scheduler Users Guide.

Note This may require that you restart Adaptive Server.

2 If Adaptive Server is unnamed, set the server name then restart Adaptive
Server:
sp_addserver server_name, local
3 Any login that runs sp_opt_querystats must have the js_user_role role, and
must use a non-Null password to log in to Adaptive Server.
sp_opt_querystats require the sa_role:

grant role role_name to login_name

4 Create an external login on the loopback server for all users who run
sp_opt_querystats:

sp_addexternlogin loopback, <login>, <password>,

<password>
5 Sybase recommends that you set the value for maximum job output to
1000000 to ensure that Job Scheduler can capture all diagnostic output for
your queries.

Note If sp_opt_querystats truncates the output, increase the value for

maximum job output. Increasing this value consumes no additional
resources.

Running sp_opt_querystats
Log in to Adaptive Server using the login that has the js_user_role. Run
sp_opt_querystats to analyze a query (the diagnostic information begins with
the phrase [BEGIN QUERY ANALYSIS] and ends with [END QUERY
ANALYSIS]).

The syntax is:

sp_opt_querystats "query_text" | help [, "diagnostic_options" | null
[, database_name]]
For example:
sp_opt_querystats "select * from pubs2..authors where

230 Adaptive Server Enterprise

 CHAPTER 7 Controlling Optimization

au_id = "172-32-1176"
See the Reference Manual: Procedures.

Specifying query processor choices

Adaptive Server lets you specify these optimization choices by including
commands in a query batch or in the text of the query:
• The order of tables in a join
• The number of tables evaluated at one time during join optimization
• The index used for a table access
• The I/O size
• The cache strategy
• The degree of parallelism
Under some circumstances, the query processor does not choose the best plan.
Occasionally, the plan choosen by the query processor is only slightly more
expensive than the “best” plan, so you must weigh the cost of maintaining
forced options against the slower performance of a less than optimal plan.
The commands to specify join order, index, I/O size, or cache strategy, coupled
with the query-reporting commands like statistics io and showplan, can help you
determine why the query processor makes its choices.

Warning! Use the options described in this chapter with caution. Forced query
plans may be inappropriate in some situations and may cause poor
performance. If you include these options in your applications, regularly check
query plans, I/O statistics, and other performance data.

These options are generally intended for use as tools for tuning and
experimentation, not as long-term solutions to optimization problems.

Performance and Tuning Series: Query Processing and Abstract Plans 231
Specifying table order in joins

Specifying table order in joins

Adaptive Server optimizes join orders to minimize I/O. In most cases, the order
that the query processor chooses does not match the order of the from clauses
in your select command. To force Adaptive Server to access tables in the order
they are listed, use:
set forceplan [on|off]
The query processor still chooses the best access method for each table. If you
use forceplan and specify a join order, the query processor may use different
indexes on tables than it would with a different table order, or it may not be able
to use existing indexes.
You might use this command as a debugging aid if other query analysis tools
lead you to suspect that the query processor is not choosing the best join order.
Always verify that the order you are forcing reduces I/O and logical reads by
using set statistics io on and comparing I/O both with and without forceplan.
If you use forceplan, your routine performance maintenance checks should
include verifying that the queries and procedures that use forceplan still require
the option to improve performance.
You can include forceplan in the text of stored procedures.
set forceplan forces only join order, and not join type. There is no command for
specifying the join type; you can disable merge joins at the server or session
level.
You can disable hash joins at the session level. Also remember that an abstract
plan allows full plan specification, including join order and join types.
See Chapter 12, “Creating and Using Abstract Plans,” and “Enabling and
disabling merge joins” on page 244.
Forcing join order has these risks:
• Misuse can lead to extremely expensive queries. Always test the query
thoroughly with statistics io, and with and without forceplan.
• It requires maintenance. You must regularly check queries and stored
procedures that include forceplan. Also, each new version of Adaptive
Server may eliminate the problems that lead you to incorporate index
forcing, so check all queries that use forced query plans each time a new
version is installed.
Before you use forceplan:

232 Adaptive Server Enterprise

 CHAPTER 7 Controlling Optimization

• Check the showplan output to determine whether index keys are used as
expected.
• Use set option show normal to look for other optimization problems.
• Run update statistics on the index.
• Use update statistics to add statistics for search arguments on unindexed
search clauses in the query, especially for search arguments that match
minor keys in compound indexes.
• Use set option show_missing_stats on to look for columns that may need
statistics.
• If the query joins more than four tables, use set table count, which may
result in an improved join order.
See “Specifying the number of tables considered by the query processor”
on page 233.

Specifying the number of tables considered by the

query processor
In versions earlier than 15.0, Adaptive Server optimized joins by considering
two to four permutations at a time. Versions 15.0 and later do not limit the
query processor to two or four permutations. Instead, the new search engine
introduces a timeout mechanism to avoid excessive time spent optimizing a
query. The set table count setting discussed later in this section still affects the
initial join order looked at by the search engine, and thus affects the final join
order when timeout does occur. If you suspect that an inefficient join order is
being chosen when the search engine times out, use set table count to increase
the number of tables that are considered, which affects the initial join order
considered by the search engine in starting the permutation.
Adaptive Server still optimizes joins by considering permutations of two to
four tables at a time, but if you suspect that an inefficient join order is being
chosen for a join query, use set table count to increase the number of tables that
are considered at the same time:
set table count int_value
Valid values are 0 though 8; 0 restores the default behavior.
For example, to specify four-at-a-time optimization, use:

Performance and Tuning Series: Query Processing and Abstract Plans 233
Specifying query index

set table count 4

As you decrease the value, you reduce the chance that the query processor
considers all possible join orders. Increasing the number of tables considered at
one time during join ordering can greatly increase the time it takes to optimize
a query.
Since the time it takes to optimize the query is increased with each additional
table, set table count is most useful when the execution savings from improved
join order outweighs the extra optimizing time. Some examples are:

• If you think that a more optimal join order can shorten total query
optimization and execution time, especially for stored procedures that you
expect to be executed many times once a plan is in the procedure cache
• When saving abstract plans for later use
Use statistics time to check parse and compile time, and statistics io to verify that
the improved join order is reducing physical and logical I/O.
If increasing the table count produces an improvement in join optimization, but
unacceptably increases CPU time, rewrite the from clause in the query,
specifying the tables in the join order indicated by showplan output, and use
forceplan to run the query. Be sure that your routine performance maintenance
checks include verifying that the join order you are forcing still improves
performance.

Specifying query index

You can use the (index index_name clause in select, update, and delete
statements to specify the index to use for a query. You can also force a query to
perform a table scan by specifying the table name. The syntax is:
select select_list
from table_name [correlation_name]
(index {index_name | table_name } )
[, table_name ...]
where ...

delete table_name
from table_name [correlation_name]
(index {index_name | table_name }) ...

234 Adaptive Server Enterprise

 CHAPTER 7 Controlling Optimization

update table_name set col_name = value

from table_name [correlation_name]
(index {index_name | table_name})...
For example:
select pub_name, title
from publishers p, titles t (index date_type)
where p.pub_id = t.pub_id
and type = "business"
and pubdate > "1/1/93"
Specifying an index in a query may be helpful when you suspect that the query
processor is choosing a suboptimal query plan. When you do specify the index:
• Always check statistics io for the query to see whether the index you
choose requires less I/O than the query processor’s choice.
• Test a full range of valid values for the query clauses, especially if you are:
• Tuning queries on tables that have skewed data distribution
• Performing range queries, since the access methods for these queries
are sensitive to the size of the range
Use (index index_name only after testing when you are certain that the query
performs better with the specified index option. Once you include an index
specification in a query, regularly verify that the resulting plan is still better
than other choices made by the query processor.
If a unclustered index has the same name as the table, specifying a table name
causes the unclustered index to be used. You can force a table scan using select
select_list from tablename (0).
Specifying indexes has these risks:
• Changes in the distribution of data could make the forced index less
efficient than other choices.
• Dropping the index means that all queries and procedures that specify the
index print an informational message indicating that the index does not
exist. The query is optimized using the best alternative access method.
• Increased maintenance, since you must periodically check all queries
using this option. Also, each new version of Adaptive Server may
eliminate the problems that lead you to incorporate index forcing, so you
should check all queries using forced indexes each time you install a new
version.

Performance and Tuning Series: Query Processing and Abstract Plans 235
Specifying I/O size in a query

• The index must exist at the time the query using it is optimized. You
cannot create an index and then use it in a query in the same batch.
Before specifying an index in queries:
• Check showplan output for the “Keys are” message to be sure that the
index keys are being used as expected.
• Use dbcc traceon(3604) or set option show normal to look for other
optimization problems.
• Run update statistics on the index.
• If the index is a composite index, run update statistics on the minor keys in
the index, if they are used as search arguments. This can greatly improve
query processor cost estimates. Creating statistics for other columns
frequently used for search clauses can also improve estimates.
• Use set option show_missing_stats on to look for columns that may need
statistics.

Specifying I/O size in a query

If your Adaptive Server is configured for large I/Os in the default data cache or
in named data caches, the query processor may decide to use large I/O for:
• Queries that scan entire tables
• Range queries using clustered indexes, such as queries using >, <, > x and
< y, between, and like “charstring %”
• Queries that scan a large number of index leaf pages
If the cache used by the table or index is configured for 16K I/O, a single I/O
can read up to 8 pages simultaneously. Each named data cache can have several
pools, each with a different I/O size. Specifying the I/O size in a query causes
the I/O for that query to take place in the pool that is configured for that size.
See the System Administration Guide: Volume 2 for information on configuring
named data caches.
To specify an I/O size that is different from the one chosen by the query
processor, add the prefetch specification to the index clause of a select, delete,
or update statement. The syntax is:
select select_list
from table_name

236 Adaptive Server Enterprise

 CHAPTER 7 Controlling Optimization

( [index {index_name | table_name} ]

prefetch size)
[, table_name ...]
where ...

delete table_name from table_name

( [index {index_name | table_name} ]
prefetch size)
...

update table_name set col_name = value

from table_name
( [index {index_name | table_name} ]
prefetch size)
...
The valid prefetch size depends on the page size. If no pool of the specified size
exists in the data cache used by the object, the query processor chooses the best
available size.
If there is a clustered index on au_lname, this query performs 16K I/O while it
scans the data pages:
select *
from authors (index au_names prefetch 16)
where au_lname like "Sm%"
If a query normally performs large I/O, and you want to check its I/O
performance with 2K I/O, you can specify a size of 2K:
select type, avg(price)
from titles (index type_price prefetch 2)
group by type

Note Reference to large I/Os are on a 2K logical page size server. If you have
an 8K page size server, the basic unit for the I/O is 8K. If you have a 16K page
size server, the basic unit for the I/O is 16K.

Index type and large I/O size

When you specify an I/O size with prefetch, the specification can affect both
the data pages and the leaf-level index pages. Table 7-6 shows the effects.

Performance and Tuning Series: Query Processing and Abstract Plans 237
Specifying I/O size in a query

Table 7-6: Access methods and prefetching

Access method Large I/O performed on
Table scan Data pages
Clustered index Data pages only, for allpages-locked tables
Data pages and leaf-level index pages for
data-only-locked tables
Nonclustered index Data pages and leaf pages of nonclustered index

showplan reports the I/O size used for both data and leaf-level pages.

See “I/O size messages” on page 56.

When prefetch specification cannot be followed

In most cases, when you specify an I/O size in a query, the query processor
incorporates the I/O size into the query’s plan. However, there are times when
the specification cannot be followed, either for the query as a whole or for a
single, large I/O request.
You cannot use large I/O for the query if:
• The cache is not configured for I/O of the specified size. The query
processor substitutes the best size available.
• sp_cachestrategy has been used to disable large I/O for the table or index.

You cannot use large I/O for a single buffer if:

• Any of the pages included in that I/O request are in another pool in the
cache.
• The page is on the first extent in an allocation unit. This extent holds the
allocation page for the allocation unit, and only seven data pages.
• No buffers are available in the pool for the requested I/O size.
When a large I/O cannot be performed, Adaptive Server performs 2K I/O on
the specific page or pages in the extent that are needed by the query.
To determine whether the prefetch specification is followed, use showplan to
display the query plan and statistics io to see the results on I/O for the query.
sp_sysmon reports on the large I/Os requested and denied for each cache.

See the Performance and Tuning Series: Monitoring Adaptive Server with
sp_sysmon.

238 Adaptive Server Enterprise

 CHAPTER 7 Controlling Optimization

setting prefetch
By default, a query uses large I/O whenever a large I/O pool is configured and
the query processor determines that large I/O would reduce the query cost. To
disable large I/O during a session, use:
set prefetch off
To reenable large I/O, use:
set prefetch on
If large I/O is turned off for an object using sp_cachestrategy, set prefetch on
does not override that setting.
If large I/O is turned off for a session using set prefetch off, you cannot override
the setting by specifying a prefetch size as part of a select, delete, or insert
statement.
The set prefetch command takes effect in the same batch in which it is run, so
you can include it in a stored procedure to affect the execution of the queries in
the procedure.

Specifying cache strategy

For queries that scan a table’s data pages or the leaf level of an unclustered
index (covered queries), the Adaptive Server query processor chooses one of
two cache replacement strategies: the fetch-and-discard, most-recently used
(MRU) strategy or the least recently used (LRU) strategy.
See the Performance and Tuning Series: Physical Database Tuning.
The query processor may choose the MRU strategy for:
• Any query that performs table scans
• A range query that uses a clustered index
• A covered query that scans the leaf level of a nonclustered index
• An inner table in a nested-loop join, if the inner table is larger than the
cache
• The outer table of a nested-loop join, since it needs to be read only once
• Both tables in a merge join.
To affect the cache strategy for objects:

Performance and Tuning Series: Query Processing and Abstract Plans 239
Specifying cache strategy

• Specify lru or mru in a select, update, or delete statement

• Use sp_cachestrategy to disable or reenable the mru strategy
If you specify the MRU strategy, and a page is already in the data cache, the
page is placed at the MRU end of the cache, rather than at the wash marker.
Specifying the cache strategy affects only data pages and the leaf pages of
indexes. Root and intermediate pages always use the LRU strategy.

In select, delete, and update statements

You can use lru or mru in a select, delete, or update command to specify the I/O
size for the query. (You get only sizes based on caches you have configured
correctly. For example, if you specify 4K but Adaptive Server does not use a
4K page size, the command returns 2K):
select select_list
from table_name
(index index_name prefetch size [lru|mru])
[, table_name ...]
where ...

delete table_name from table_name (index index_name

prefetch size [lru|mru]) ...

update table_name set col_name = value

from table_name (index index_name
prefetch size [lru|mru]) ...
For example, to add the LRU replacement strategy to a 16K I/O specification,
enter:
select au_lname, au_fname, phone
from authors (index au_names prefetch 16 lru)
See “Specifying I/O size in a query” on page 236.

240 Adaptive Server Enterprise

 CHAPTER 7 Controlling Optimization

Controlling large I/O and cache strategies

Status bits in the sysindexes table identify whether you should consider a table
or an index for large I/O prefetch or for MRU replacement strategy. By default,
both are enabled. To disable or reenable these strategies, use sp_cachestrategy:
sp_cachestrategy dbname , [ownername.]tablename
[, indexname | "text only" | "table only"
[, { prefetch | mru }, { "on" | "off"}]]
For example, to turn off the large I/O prefetch strategy for the au_name_index
of the authors table, enter:
sp_cachestrategy pubtune, authors, au_name_index,
prefetch, "off"
To reenable MRU replacement strategy for the titles table:
sp_cachestrategy pubtune, titles, "table only",
mru, "on"
Only a system administrator or the object owner can change or view the cache
strategy status of an object.

Getting information on cache strategies

To see the cache strategy that is in effect for a given object, execute
sp_cachestrategy, with the database and object name:

sp_cachestrategy pubtune, titles

object name index name large IO MRU
---------------- ---------------- -------- --------
titles NULL ON ON

showplan output shows the cache strategy used for each object, including
worktables.

Asynchronous log service

Asynchronous log service (ALS) increases scalability in Adaptive Server and
provides higher throughput in logging subsystems for high-end symmetric
multiprocessor systems.

Performance and Tuning Series: Query Processing and Abstract Plans 241
Asynchronous log service

You cannot use ALS if you have fewer than four engines; if you attempt to do
so, with fewer than 4 online engines an error message appears.
You can enable, disable, or configure ALS using the sp_dboption stored
procedure:
sp_dboption <db Name>, "async log service",
"true|false"
After issuing sp_dboption, you must issue a checkpoint in the database for
which you are setting the ALS option:
sp_dboption "mydb", "async log service", "true"
use mydb
checkpoint
You can use the checkpoint to identify the one or more databases or use an all
clause:
checkpoint [all | [dbname[, dbname[, dbname.....]]]
To disable ALS, enter:
sp_dboption "mydb", "async log service", "false"
use mydb
checkpoint
-------------
Before you disable ALS, make sure there are no active users in the database. If
there are active users in the database when you disable ALS, you see this error
message:
Error 3647: Cannot put database in single-user mode.
Wait until all users have logged out of the database and
issue a CHECKPOINT to disable "async log service".
Use sp_helpdb to see whether ALS is enabled in a specified database:
sp_helpdb "mydb"
----------
mydb 3.0 MB sa 2
July 09, 2002
select into/bulkcopy/pllsort, trunc log on chkpt,
async log service
For more information on these stored procedures, see the Adaptive Server
Reference Manual: Procedures.

242 Adaptive Server Enterprise

 CHAPTER 7 Controlling Optimization

Understanding the user log cache (ULC) architecture

The Adaptive Server logging architecture features the user log cache, or ULC,
by which each task owns its own log cache. No other task can write to this
cache, and the task continues writing to the user log cache whenever a
transaction generates a log record. When the transaction commits or aborts, or
when the log cache fills up, the ULC is flushed to the common log cache,
shared by all the current tasks, which is then written to the disk.
Flushing the ULC is the first part of a commit or abort operation, requiring the
following steps, each of which can cause delay or increase contention:
1 Obtaining a lock on the last log page.
2 Allocating new log pages if necessary.
3 Copying the log records from the ULC to the log cache.
The processes in steps 2 and 3 require you to hold a lock on the last log
page, which prevents any other tasks from writing to the log cache or
performing commit or abort operations.
4 Flush the log cache to disk.
Step 4 requires repeated scanning of the log cache to issue write commands
on dirty buffers.
Repeated scanning can cause contention on the buffer cache spinlock to
which the log is bound. Under a large transaction load, contention on this
spinlock can be significant.

When to use ALS

You can enable ALS on any specified database that has at least one of the
following performance issues, if your system runs four or more online engines:
• Heavy contention on the last log page
You can tell that the last log page is under contention when the sp_sysmon
output in the Task Management Report section shows a significantly high
value. For example:
Task Management per sec per xact count % of total
----------------------- --------- ---------- -------- --------
Log Semaphore Contention 58.0 0.3 34801 73.1%

Performance and Tuning Series: Query Processing and Abstract Plans 243
Enabling and disabling merge joins

• Underutilized bandwidth in the log device

Note Use ALS only when you identify a single database with high transaction
requirements, since setting ALS for multiple databases may cause unexpected
variations in throughput and response times. To configure ALS on multiple
databases, first check that throughput and response times are satisfactory.

Using the ALS

Two threads scan the dirty buffers (buffers full of data not yet written to the
disk), copy the data, and write it to the log. These threads are:
• The user log cache (ULC) flusher – The ULC flusher is a system task
thread that is dedicated to flushing the user log cache of a task into the
general log cache. When a task is ready to commit, the user enters a
commit request into the flusher queue. Each entry has a handle, by which
the ULC flusher can access the ULC of the task that queued the request.
The ULC flusher task continuously monitors the flusher queue, removing
requests from the queue and servicing them by flushing ULC pages into
the log cache.
• The log writer – Once the ULC flusher has finished flushing the ULC
pages into the log cache, it queues the task request into a wakeup queue.
The log writer patrols the dirty buffer chain in the log cache, issuing a write
command if it finds dirty buffers, and monitors the wakeup queue for tasks
whose pages are all written to disk. Since the log writer patrols the dirty
buffer chain, it knows when a buffer is ready to write to disk.

Enabling and disabling merge joins

By default, merge joins are enabled at the server level, for allrows mix and for
allrows_dss optgoal, and are disabled at the server level for other optgoals,
including allrows_oltp. When merge joins are disabled, the server costs only the
other join types that are not disabled. To enable merge joins server-wide, set
enable merge join to 1. The enable sort-merge joins and JTC configuration
parameter from versions of Adaptive Server earlier than 15.0 does not affect
the 15.0 and later query processor.

244 Adaptive Server Enterprise

 CHAPTER 7 Controlling Optimization

The command set merge_join on overrides the server level to allow use of
merge joins in a session or stored procedure.
To enable merge joins, use:
set merge_join on
To disable merge joins, use:
set merge_join off

Enabling and disabling hash joins

By default, hash joins are enabled only when you run allrows_dss optgoal. To
override the server level setting, and allow use of hash join in a session or
stored procedure, use set hash_join on.
To enable hash joins, use:
set hash_join on
To disable hash joins, use:
set hash_join off

Enabling and disabling join transitive closure

In Adaptive Server version 15.0 and later, join transitive closure is always on
and cannot be disabled. The search engine uses the timeout mechanism to
avoid excessive optimization time. Although the timeout setting no longer
affects the actual use of transitive closure for the query processor, it can still
affect the initial join order with which the search engine begins the permutation
when the timeout occurs. You may find this discussion useful when you suspect
that a suboptimal join order is being chosen at timeout.
By default, join transitive closure is not enabled at the server level, since it can
increase optimization time. You can enable join transitive closure at a session
level with set jtc on. The session-level command overrides the server-level
setting for the enable sort-merge joins and JTC configuration parameter
(available for versions of Adaptive Server earlier than 15.0).

Performance and Tuning Series: Query Processing and Abstract Plans 245
Controlling literal parameterization

For queries that execute quickly, even when several tables are involved, join
transitive closure may increase optimization time with little improvement in
execution cost. For example, with join transitive closure applied to this query,
the number of possible joins is multiplied for each added table:
select * from t1, t2, t3, t4, ... tN
where t1.c1 = t2.c1
and t1.c1 = t3.c1
and t1.c1 = t4.c1
...
and t1.c1 = tN.c1
For joins on very large tables, however, the additional optimization time
involved in costing the join orders added by join transitive closure may result in
a join order that greatly improves the response time.
Use set statistics time to see how long Adaptive Server takes to optimize the
query. If running queries with set jtc on greatly increases optimization time, but
also improves query execution by choosing a better join order, check the
showplan, set option show_search_engine normal, or set option show_search_engine
long output. Explicitly add the useful join orders to the query text. Run the
query without join transitive closure, and get the improved execution time,
without the increased optimization time of examining all possible join orders
generated by join transitive closure.
You can also enable join transitive closure and save abstract plans for queries
that benefit. If you then execute those queries with loading from the saved
plans enabled, the saved execution plan is used to optimize the query, making
optimization time extremely short.
See Performance and Tuning: Optimizer and Abstact Plans for more
information on using abstract plans and configuring join transitive closure
server-wide.

Controlling literal parameterization

Adaptive Server version 15.0.1 and later allow you to automatically convert
literal values in SQL queries to parameter descriptions (similar to variables).
To enable or disable enable literal autoparam server-wide, use:
sp_configure "enable literal autoparam", [0 | 1]

246 Adaptive Server Enterprise

 CHAPTER 7 Controlling Optimization

Where 1 automatically converts literal values to parameter descriptions, and 0

(the default) disables the feature.
Set literal parameterization at the session level using:
set literal_autoparam [off | on]
In versions of Adaptive Server earlier than 15.0.1, two queries that were
identical except for one or more literal values resulted in the statement cache
storing two separate query plans, or two additional rows, in sysqueryplans. For
example, the query plans for these queries were stored separately, even though
they are almost identical:
select count(*) from titles where total_sales > 100
select count(*) from titles where total_sales > 200
Examples If you enable automatic literal parameterization, the SQL text of the select
count(*) example referred to above is converted to:

select count(*) from titles where total_sales > @@@V0_INT

Where @@@V0_INT is an internally generated name for the parameter that
represents the literal values 100 and 200.
All instances of literal values in the SQL text are replaced by internally
generated parameters. For example:
select substring(name, 3, 4) from sysobjects where name in
('systypes', 'syscolumns')
is transformed to:
select substring(name, 3, 4) from sysobjects where name in
(@@@V0_VCHAR1,@@@V1_VCHAR1)
Any combination of values that replace the literals, 3, 4, systypes and
syscolumns is transformed to the same SQL text with the same parameters
and shares the same query plan when you enable the statement cache.
Automatic literal parameterization:
• Reduces compilation time on the second—and subsequent—executions of
the query, regardless of the literal values in the query.
• Reduces the amount of SQL text storage space, including memory usage
in the statement cache and the number of rows in sysqueryplans for
abstract plans and query metrics.
• Reduces the amount of procedure cache used to store query plans.
• Occurs automatically within Adaptive Server, when enabled: you need not
change the applications that submit the queries to Adaptive Server.

Performance and Tuning Series: Query Processing and Abstract Plans 247
Suggesting a degree of parallelism for a query

Usage issues for automatic literal parameterization include:

• Adaptive Server parameterizes the literals only for select, delete, update,
and insert. For insert statements, Adaptive Server parameterizes only insert
... select statements, not insert ... values statements.

• Adaptive Server does not parameterize literals in queries that include a

derived table.
• Adaptive Server does not parameterize queries similar to select id + 1
from sysobjects group by id + 1 or select id + 1 from
sysobjects order by id + 1 because of the expressions (“id + 1”)
in the group by and order by clauses.
• Adaptive Server does not cache SQL statements with text longer than
16384 bytes in the statement cache (SQL statements over 16K are not
cached). Transforming literals in the SQL statement into variables can
significantly expand the size of the SQL text (especially if there was a
large number of literals). Enabling automatic literal parameterization may
result in Adaptive Server not caching some SQL statements that it would
otherwise have cached.
• If two SQL statements are the same except that their literal values have
different datatypes, they are not transformed into matching SQL texts. For
example, the following two SQL statements return the same results, but
are parameterized differently because they use the different datatypes:
select name from sysobjects where id = 1
select name from sysobjects where id = 1.0
The parameterized versions of these statements are:
select name from sysobjects where id = @@@V0_INT
select name from sysobjects where id = @@@V0_NUMERIC

Suggesting a degree of parallelism for a query

The parallel and degree_of_parallelism extensions to the from clause of a select
command allow users to restrict the number of worker processes used in a scan.
For a parallel partition scan to be performed, the degree_of_parallelism must
be equal to or greater than the number of partitions. For a parallel index scan,
specify any value for the degree_of_parallelism.
The syntax for the select statement is:

248 Adaptive Server Enterprise

 CHAPTER 7 Controlling Optimization

select...
[from {tablename}
[(index index_name
[parallel [degree_of_parallelism | 1]]
[prefetch size] [lru|mru])],
{tablename} [([index_name]
[parallel [degree_of_parallelism | 1]
[prefetch size] [lru|mru])] ...
Table 7-7 shows how to combine the index and parallel keywords to obtain
serial or parallel scans.
Table 7-7: Optimizer hints for serial and parallel execution
Use: To specify a:
(index tablename parallel N) Parallel partition scan
(index index_name parallel N) Parallel index scan
(index tablename parallel 1) Serial table scan
(index index_name parallel 1) Serial index scan
(parallel N) Parallel scan, with the choice of table or index scan left to the optimizer
(parallel 1) Serial scan, with the choice of table or index scan left to the optimizer

When you specify the parallel degree for a table in a merge join, it affects the
degree of parallelism used for both the scan of the table and the merge join.
You cannot use the parallel option if you have disabled parallel processing
either at the session level with the set parallel_degree 1 command, or at the
server level with the parallel degree configuration parameter. The parallel
option cannot override these settings.
If you specify a degree_of_parallelism that is greater than the maximum
configured degree of parallelism, Adaptive Server ignores the hint.
The optimizer ignores hints that specify a parallel degree if any of the
following conditions is true:
• The from clause is used in the definition of a cursor.
• parallel is used in the from clause of an inner query block of a subquery, and
the optimizer does not move the table to the outermost query block during
subquery flattening.
• The table is a view, a system table, or a virtual table.
• The table is the inner table of an outer join.
• The query specifies exists, min, or max on the table.

Performance and Tuning Series: Query Processing and Abstract Plans 249
Optimization goals

• The value for the max scan parallel degree configuration parameter is set
to 1.
• An unpartitioned clustered index is specified or is the only parallel option.
• A nonclustered index is covered.
• The query is processed using the OR strategy.
• The select statement is used for an update or insert.

Query level parallel clause examples

To specify the degree of parallelism for a single query, include parallel after the
table name. This example executes in serial:
select * from titles (parallel 1)
This example specifies the index to be used in the query, and sets the degree of
parallelism to 5:
select * from titles
(index title_id_clix parallel 5)
where ...
To force a table scan, use the table name instead of the index name.

Optimization goals
Adaptive Server lets you choose a query optimization goal that best suits your
query environment:
• fastfirstrow – optimizes queries so that Adaptive Server returns the first few
rows as quickly as possible.
• allrows_oltp – optimizes queries so that Adaptive Server uses a limited
number of optimization criteria (described in “Optimization criteria” on
page 252) to find a good query plan. allrows_oltp is most useful for purely
OLTP queries.
• allrows_mixed – optimizes queries so that Adaptive Server uses most
available optimization techniques, including merge_join and parallel, to
find the best query plan. allrows_mixed, which is the default strategy, is
most useful in a mixed-query environment.

250 Adaptive Server Enterprise

 CHAPTER 7 Controlling Optimization

• allrows_dss – optimizes queries so that Adaptive Server uses all available

optimization techniques to find the best query plan, including hash join,
advanced aggregates processing, and bushy tree plan. allrows_dss is most
useful in a DSS environment.

Setting optimization goals

You can set the optimization goal at the server, session, or query level. The
server-level optimization goal is overridden at the session level, which is
overridden at the query level—which means you can set a different
optimization goal at each level.
At the server level To set the optimization goal at the server level, you can:
• Use the sp_configure command
• Modify the optimization goal configuration parameter in the Adaptive
Server configuration file
For example, to set the optimization level for the server to fastfirstrow, enter:
sp_configure "optimization goal", 0, "fastfirstrow"
At the session level To set the optimization goal at the session level, use set plan optgoal. For
example, to modify the optimization goal for the session to allrows, enter:
set plan optgoal allrows_oltp
To verify the current optimization goal at the session level, enter:
select @@optgoal
At the query level To set the optimization goal at the query level, use the select or other DML
command. For example, to change the optimization goal to allrows_oltp for the
current query, enter:
select * from A order by A.a plan "(use optgoal allrows_oltp)"
At the query level only, you can specify the number of rows that Adaptive
Server quickly returns when you set fastfirstrow as the optimization goal. For
example, enter:
select * from A order by A.a plan "(use optgoal fastfirstrow 5)"
Some exceptions In general, you can set query-level optimization goals using select, update, and
delete statements. However:

• You cannot set query-level optimization goals in pure insert statements,

although you can set optimization goals in select ... insert statements.

Performance and Tuning Series: Query Processing and Abstract Plans 251
Optimization criteria

• fastfirstrow is relevant only for select statements; it incurs an error when

used with other DML statements.

Optimization criteria
You can set specific optimization criteria for each session. The optimization
criteria represent specific algorithms or relational techniques that may or may
not be considered when Adaptive Server creates a query plan. By setting
individual optimization criteria on or off, you can fine-tune the query plan for
the current session.

Note Each optimization goal has default settings for each optimization
criterion. Resetting optimization criteria may interfere with the default settings
of the current optimization goal and produce an error message—although
Adaptive Server applies the new setting.

Sybase recommends that you set individual optimization criteria only rarely
and with caution if you must fine-tune a particular query. Overriding
optimization goal settings can overly complicate query administration. Always
set optimization criteria after setting any existing session level optgoal setting;
an explicit optgoal setting may return an optimization criteria to its default
value.

See “Default optimization criteria” on page 254.

Setting optimization Use the set command to enable or disable individual criteria.
criteria
For example, to enable the hash join algorithm, enter:
set hash_join 1
To disable the hash join algorithm, enter:
set hash_join 0
To enable one option and disable another, enter:
set hash_join 1, merge_join 0
Criteria descriptions Most criteria described here decides whether a particular query engine operator
can be used in the final plan chosen by the optimizer.
The optimization criteria are:

252 Adaptive Server Enterprise

 CHAPTER 7 Controlling Optimization

• hash_join – determines whether the query processor may use the hash join
algorithm. Hash joins may consume more runtime resources, but are
valuable when the joining columns do not have useful indexes or when a
relatively large number of rows satisfy the join condition, compared to the
product of the number of rows in the joined tables.
• hash_union_distinct – determines whether the query processor may use the
hash union distinct algorithm, which is not efficient if most rows are
distinct.
• merge_join – determines whether the query processor may use the merge
join algorithm, which relies on ordered input. merge_join is most valuable
when input is ordered on the merge key—for example, from an index scan.
merge_join is less valuable if sort operators are required to order input.

• merge_union_all – determines whether the query processor may use the

merge algorithm for union all. merge_union_all maintains the ordering of
the result rows from the union input. merge_union_all is particularly
valuable if the input is ordered and a parent operator (such as merge join)
benefits from that ordering. Otherwise, merge_union_all may require sort
operators that reduce efficiency.
• merge_union_distinct – determines whether the query processor may use
the merge algorithm for union. merge_union_distinct is similar to
merge_union_all, except that duplicate rows are not retained.
merge_union_distinct requires ordered input and provides ordered output.

• multi_table_store_ind – determines whether the query processor may use

reformatting on the result of a multiple table join. Using
multi_tablet_store_ind may increase the use of worktables.

• nl_join – determines whether the query processor may use the nested-loop-
join algorithm.
• opportunistic_distinct_view – determines whether the query processor may
use a more flexible algorithm when enforcing distinctness.
• parallel_query – determines whether the query processor may use parallel
query optimization.
• store_index – determines whether the query processor may use
reformatting, which may increase the use of worktables.
• append_union_all – determines whether the query processor may use the
append union all algorithm.

Performance and Tuning Series: Query Processing and Abstract Plans 253
Optimization criteria

• bushy_search_space – determines whether the query processor may use

bushy-tree-shaped query plans, which may increase the search space, but
provide more query plan options to improve performance.
• distinct_hashing – determines whether the query processor may use a
hashing algorithm to eliminate duplicates, which is very efficient when
there are few distinct values compared to the number of rows.
• distinct_sorted – determines whether the query processor may use a single-
pass algorithm to eliminate duplicates. distinct_sorted relies on an ordered
input stream, and may increase the number of sort operators if its input is
not ordered.
• group-sorted – determines whether the query processor may use an on-the-
fly grouping algorithm. group-sorted relies on an input stream sorted on the
grouping columns, and it preserves this ordering in its output.
• distinct_sorting – determines whether the query processor may use the
sorting algorithm to eliminate duplicates. distinct_sorting is useful when
the input is not ordered (for example, if there is no index) and the output
ordering generated by the sorting algorithm could benefit; for example, in
a merge join.
• group_hashing – determines whether the query processor may use a group
hashing algorithm to process aggregates.
• index_intersection – determines whether the query processor may use the
intersection of multiple index scans as part of the query plan in the search
space.
If all the algorithms of a relational operator are disabled, the query processor
reenables a default algorithm. For example, if all join algorithms (nl_join,
m_join, and h_join) are disabled, the query processor enables nl_join.

The query processor can also reenable nl_join for semantic reasons: for
example, if the joining tables are not connected through equijoins.
Default optimization Each optimization goal— fastfirstrow, allrows_oltp, allrows_mixed,
criteria allrows_dss—has a default setting (on (1)or off (0)) for each optimization
criterion. For example, the default setting for merge_join is off (0) for
fastfirstrow and allrows_oltp, and on (1) for allrows_mixed and allrows_dss. See
Table 7-8 for a list of default settings for each optimization criteria.
Sybase recommends that you reset the optimization goal and evaluate
performance before changing optimization criteria. Change optimization
criteria only if you must fine-tune a particular query.

254 Adaptive Server Enterprise

 CHAPTER 7 Controlling Optimization

Table 7-8: Default settings for optimization criteria

Optimization criteria fastfirstrow allrows_oltp allrows_mixed allrows_dss
append_union_all 1 1 1 1
bushy_search_space 0 0 0 1
distinct_sorted 1 1 1 1
distinct_sorting 1 1 1 1
group_hashing 1 1 1 1
group_sorted 1 1 1 1
hash_join 0 0 0 1
hash_union_distinct 1 1 1 1
index_intersection 0 0 0 1
merge_join 0 0 1 1
merge_union_all 1 1 1 1
multi_gt_store_ind 0 0 0 1
nl_join 1 1 1 1
opp_distinct_view 1 1 1 1
parallel_query 1 0 1 1
store_index 1 1 1 1

Limiting optimization time

You can use the optimization timeout limit configuration parameter to restrict the
amount of time Adaptive Server spends optimizing a query. optimization
timeout limit specifies the amount of time Adaptive Server can spend optimizing
a query as a percentage of the total time spent processing the query.
The timeout is activated only if:
• At least one complete plan has been retained as the best plan, and
• The optimization timeout limit has been exceeded.
Use sp_configure to set optimization timeout limit at the server level. For
example, to limit optimization time to 10 percent of total query processing
time, enter:
sp_configure “optimization timeout limit”, 10
To set optimization timeout limit at the session level, use:
set plan optimeoutlimit n

Performance and Tuning Series: Query Processing and Abstract Plans 255
Controlling parallel optimization

This command overrides the server setting.

The default value is 10 percent; you can specify any value from 1 to 1000.
At the server level, there is a separate configuration parameter, optimization
timeout limit, for the server-level default timeout value within stored procedure
compilations. The default value is 40 percent; you can specify any value from
1 to 4000.
For more information about optimization timeout limit, see “Limiting the time
spent optimizing a query” on page 16.

Controlling parallel optimization

The goal of executing queries in parallel is to get the fastest response time, even
if it involves more total work from the server.
To enable and control parallel processing, Adaptive Server provides these
configuration parameters:
• number of worker processes

• max parallel degree

• max resource granularity

• max repartition degree

With the exception of number of worker processes, each of these parameters can
be set at the server and the session level. To view the current session-level
value of a parameter, use the select command. For example, to view the current
value of max resource granularity, enter:
select @@resource_granularity

Note When set or viewed at the session level, these parameters do not include
“max.”

256 Adaptive Server Enterprise

 CHAPTER 7 Controlling Optimization

number of worker processes

Use number of worker processes to specify the maximum number of worker
processes that Adaptive Server can use at any one time for all simultaneously
running parallel queries.
number of worker processes is a server-wide configuration parameter only; use
sp_configure to set the parameter. For example, to set the maximum number of
worker processes to 200, enter:
sp_configure “number of worker processes”, 200

Specifying the number of worker processes available for parallel

processing
Use max parallel degree to specify the maximum number of worker processes
allowed per query. You can configure max parallel degree at the server or the
session level.
For example, to set max parallel degree to 60 at the server level, enter:
sp_configure “max parallel degree”, 60
To set max parallel degree to 60 at the session level, enter:
set parallel_degree 60
The value of max parallel degree must be equal to or less than the current value
of number of worker processes. Setting max parallel degree to 1 turns off parallel
processing—Adaptive Server scans all tables and indexes serially. To enable
parallel partition scans, set max parallel degree equal to or greater than the
number of partitions in the table you are querying.

max resource granularity

Use max resource granularity to specify the percentage of total memory that
Adaptive Server can allocate to a single query. You can set the parameter at the
server or session level.
For example, to set max resource granularity to 35 percent at the server level,
enter:
sp_configure “max resource granularity”, 35
To set max resource granularity to 35 percent at the session level, enter:

Performance and Tuning Series: Query Processing and Abstract Plans 257
Concurrency optimization for small tables

set resource_granularity 35
The value of this parameter can affect the query optimizer’s choice of operators
for a query. If max resource granularity is set low, many hash- and sort-based
operators cannot be chosen. max resource granularity also affects the scheduling
algorithm.

max repartition degree

Use max repartition degree to suggest a number of worker processes that the
query processor can use to partition a data stream. You can set max repartition
degree at the server or query level.

Note The value of max repartition degree is a suggestion only; the query
processor decides the optimal number.

max repartition degree is most useful when the tables being queried are not
partitioned, but partitioning the resultant data stream may improve
performance by allowing concurrent SQL operations.
For example, to set max repartition degree to 15 at the server level, enter:
sp_configure “max repartition degree”, 15
To set max repartition degree to 15 at the session level, enter:
set repartition_degree 15
The value of max repartition degree cannot exceed the current value of max
parallel degree. Sybase recommends that you set the value of this parameter
equal to or less than the number of CPUs or disk systems that can work in
parallel.

Concurrency optimization for small tables

For data-only-locked tables of 15 pages or fewer, Adaptive Server does not
consider a table scan if there is a useful index on the table. Instead, it always
chooses the cheapest index that matches any search argument that can be
optimized in the query. The locking required for an index scan provides higher
concurrency and reduces the chance of deadlocks, although slightly more I/O
may be required than for a table scan.

258 Adaptive Server Enterprise

 CHAPTER 7 Controlling Optimization

If concurrency on small tables is not an issue, and you want to optimize the I/O
instead, use sp_chgattribute to disable this optimization. For example, to turn
off concurrency optimization for a table:
sp_chgattribute tiny_lookup_table,
“concurrency_opt_threshold”, 0
With concurrency optimization disabled, the query processor can choose table
scans when they require fewer I/Os.
You can also increase the concurrency optimization threshold for a table. This
command sets the concurrency optimization threshold for a table to 30 pages:
sp_chgattribute lookup_table,
“concurrency_opt_threshold”, 30
The maximum value for the concurrency optimization threshold is 32,767.
Setting the value to -1 enforces concurrency optimization for a table of any
size; this setting may be useful when a table scan is chosen over indexed
access, and the resulting locking results in increased contention or deadlocks.
The current setting is stored in systabstats.conopt_thld and is printed as part of
optdiag output.

Changing the locking scheme

Concurrency optimization affects only data-only-locked tables. Table 7-9
shows the effect of changing the locking scheme.
Table 7-9: Effects of alter table on concurrency optimization settings
Changing from Effect on stored value
Allpages to data-only Set to 15, the default
Data-only to allpages Set to 0
One data-only scheme to another Configured value retained

Performance and Tuning Series: Query Processing and Abstract Plans 259
Concurrency optimization for small tables

260 Adaptive Server Enterprise

CH A PTE R 8 Optimization for Cursors

This chapter discusses performance issues related to cursors. Cursors are

a mechanism for accessing the results of a SQL select statement one row
at a time (or several rows, if you use set cursors rows). Since cursors use
a different model from ordinary set-oriented SQL, the way cursors use
memory and hold locks has performance implications for your
applications. In particular, cursor performance issues include locking at
the page and at the table level, network resources, and overhead of
processing instructions.
Topic Page
Definition 261
Resources required at each stage 264
Cursor modes 266
Index use and requirements for cursors 267
Comparing performance with and without cursors 269
Locking with read-only cursors 272
Isolation levels and cursors 273
Partitioned heap tables and cursors 274
Optimizing tips for cursors 274

Definition
A cursor is a symbolic name that is associated with a select statement. It
enables you to access the results of a select statement one row at a time.
Figure 8-1 shows a cursor accessing the authors table.

Performance and Tuning Series: Query Processing and Abstract Plans 261
Definition

Figure 8-1: Cursor example

Result set
Cursor with select * from
authors where state = ’KY’ A978606525 Marcello Duncan KY

A937406538 Carton Nita KY

A1525070956Porczyk Howard KY
Programming can:
- Examine a row A913907285 Bier Lane KY
- Take an action based on row
values

You can think of a cursor as a “handle” on the result set of a select statement.
It enables you to examine and possibly manipulate one row at a time.

Set-oriented versus row-oriented programming

SQL was conceived as a set-oriented language. Adaptive Server is extremely
efficient when it works in set-oriented mode. Cursors are required by ANSI
SQL standards; when they are needed, they are very powerful. However, they
can have a negative effect on performance.
For example, this query performs the identical action on all rows that match the
condition in the where clause:
update titles
set contract = 1
where type = ’business’
The optimizer finds the most efficient way to perform the update. In contrast,
a cursor would examine each row and perform single-row updates if the
conditions were met. The application declares a cursor for a select statement,
opens the cursor, fetches a row, processes it, goes to the next row, and so forth.
The application may perform quite different operations depending on the
values in the current row, and the server’s overall use of resources for the cursor
application may be less efficient than the server’s set level operations.
However, cursors can provide more flexibility than set-oriented programming.
Figure 8-2 shows the steps involved in using cursors. The function of cursors
is to get to the middle box, where the user or application code examines a row
and decides what to do, based on its values.

262 Adaptive Server Enterprise

 CHAPTER 8 Optimization for Cursors

Figure 8-2: Cursor flowchart

Declare cursor

Open cursor

Fetch row

Process row
(Examine/Update/Delete)

Yes Get next

row?

No
Close cursor

Deallocate cursor

Example
Here is a simple example of a cursor with the “Process Rows” step shown
above in pseudocode:
declare biz_book cursor
for select * from titles
where type = ’business’
go
open biz_book
go
fetch biz_book
go
/* Look at each row in turn and perform
** various tasks based on values,

Performance and Tuning Series: Query Processing and Abstract Plans 263
Resources required at each stage

** and repeat fetches, until

** there are no more rows
*/
close biz_book
go
deallocate cursor biz_book
go
Depending on the content of the row, the user might delete the current row:
delete titles where current of biz_book
or update the current row:
update titles set title="The Rich
Executive’s Database Guide"
where current of biz_book

Resources required at each stage

Cursors use memory and require locks on tables, data pages, and index pages.
When you open a cursor, memory is allocated to the cursor and to store the
query plan that is generated. While the cursor is open, Adaptive Server holds
intent table locks and sometimes row or page locks. Figure 8-3 shows the
duration of locks during cursor operations.

264 Adaptive Server Enterprise

 CHAPTER 8 Optimization for Cursors

Figure 8-3: Resource use by cursor statement

Declare cursor

Open cursor

Fetch row

Process row
(Examine/Update/Delet Table
Row locks
or (intent); Memory
Yes Get next page some
row? locks row or
page
No
locks
Close cursor

Deallocate cursor

The memory resource descriptions in Figure 8-3 and Table 8-1 refer to ad hoc
cursors for queries sent by isql or Client-Library™. For other kinds of cursors,
the locks are the same, but the memory allocation and deallocation differ
somewhat depending on the type of cursor being used, as described in
“Memory use and execute cursors” on page 266.
Table 8-1: Locks and memory use for isql and Client-Library client
cursors
Cursor
command Resource use
declare cursor When a cursor is declared, Adaptive Server uses only enough memory to store the query text.

Performance and Tuning Series: Query Processing and Abstract Plans 265
Cursor modes

Cursor
command Resource use
open When a cursor is opened, Adaptive Server allocates memory to the cursor and to store the query
plan that is generated. The server optimizes the query, traverses indexes, and sets up memory
variables. The server does not access rows yet, unless it needs to build worktables. However, it
does set up the required table-level locks (intent locks). Row and page locking behavior depends
on the isolation level, server configuration, and query type.
See the Performance and Tuning Series: Locking and Concurrency Control for more
information.
fetch When a fetch is executed, Adaptive Server gets the row(s) required and reads specified values
into the cursor variables or sends the row to the client. If the cursor needs to hold lock on rows
or pages, the locks are held until a fetch moves the cursor off the row or page or until the cursor
is closed. The lock is either a shared or an update lock, depending on how the cursor is written.
close When a cursor is closed, Adaptive Server releases the locks and some of the memory allocation.
You can open the cursor again, if necessary.
deallocate cursor When a cursor is deallocated, Adaptive Server releases the rest of the memory resources used
by the cursor. To reuse the cursor, declare it again.

Memory use and execute cursors

The descriptions of declare cursor and deallocate cursor in Table 8-1 refer to ad
hoc cursors that are sent by isql or Client-Library. Other kinds of cursors
allocate memory differently:
• For cursors that are declared on stored procedures, only a small amount of
memory is allocated at declare cursor time. Cursors declared on stored
procedures are sent using Client-Library or the precompiler and are known
as execute cursors.
• For cursors declared within a stored procedure, memory is already
available for the stored procedure, and the declare statement does not
require additional memory.

Cursor modes
There are two cursor modes: read-only and update. As the names suggest, read-
only cursors can only display data from a select statement; update cursors can
be used to perform positioned updates and deletes.

266 Adaptive Server Enterprise

 CHAPTER 8 Optimization for Cursors

Read-only mode uses shared page or row locks. If read committed with lock is
set to 0, and the query runs at isolation level 1, it uses instant duration locks,
and does not hold the page or row locks until the next fetch.
Read-only mode is in effect when you specify for read only or when the cursor’s
select statement uses distinct, group by, union, or aggregate functions, and in
some cases, an order by clause.
Update mode uses update page or row locks. It is in effect when:
• You specify for update.
• The select statement does not include distinct, group by, union, a subquery,
aggregate functions, or the at isolation read uncommitted clause.
• You specify shared.
If column_name_list is specified, only those columns are updatable.
See the Performance and Tuning Series: Locking and Concurrency Control for
more information.
Specify the cursor mode when you declare the cursor. If the select statement
includes certain options, the cursor is not updatable even if you declare it for
update.

Index use and requirements for cursors

When a query is used in a cursor, it may require or choose different indexes
than the same query used outside of a cursor.

Allpages-locked tables
For read-only cursors, queries at isolation level 0 (dirty reads) require a unique
index. Read-only cursors at isolation level 1 or 3 should produce the same
query plan as the select statement outside of a cursor.
The index requirements for updatable cursors mean that updatable cursors may
use different query plans than read-only cursors. Updatable cursors have these
indexing requirements:
• If the cursor is not declared for update, a unique index is preferred over a
table scan or a nonunique index.

Performance and Tuning Series: Query Processing and Abstract Plans 267
Index use and requirements for cursors

• If the cursor is declared for update without a for update of list, a unique
index is required on allpages-locked tables. An error is raised if no unique
index exists.
• If the cursor is declared for update with a for update of list, then only a
unique index without any columns from the list can be chosen on an
allpages-locked table. An error is raised if no unique index qualifies.
When cursors are involved, an index that contains an IDENTITY column is
considered unique, even if the index is not declared unique. In some cases,
IDENTITY columns must be added to indexes to make them unique, or the
optimizer might be forced to choose a suboptimal query plan for a cursor query.

Data-only-locked tables
In data-only-locked tables, fixed row IDs are used to position cursor scans, so
unique indexes are not required for dirty reads or updatable cursors. The only
cause for different query plans in updatable cursors is that table scans are used
if columns from only useful indexes are included in the for update of list.

Table scans to avoid the Halloween problem

The Halloween problem is an update anomaly that can occur when a client
using a cursor updates a column of the cursor result-set row, and that column
defines the order in which the rows are returned from the table. For example,
if a cursor was to use an index on last_name, first_name, and update one of
these columns, the row could appear in the result set a second time.
To avoid the Halloween problem on data-only-locked tables, Adaptive Server
chooses a table scan when the columns from an otherwise useful index are
included in the column list of a for update clause.
For implicitly updatable cursors declared without a for update clause, and for
cursors where the column list in the for update clause is empty, cursors that
update a column in the index used by the cursor may encounter the Halloween
problem.

268 Adaptive Server Enterprise

 CHAPTER 8 Optimization for Cursors

Comparing performance with and without cursors

This section examines the performance of a stored procedure written two
different ways:
• Without a cursor – this procedure scans the table three times, changing the
price of each book.
• With a cursor – this procedure makes only one pass through the table.
In both examples, there is a unique index on titles(title_id).

Sample stored procedure without a cursor

This is an example of a stored procedure without cursors:
/* Increase the prices of books in the
** titles table as follows:
**
** If current price is <= $30, increase it by 20%
** If current price is > $30 and <= $60, increase
** it by 10%
** If current price is > $60, increase it by 5%
**
** All price changes must take effect, so this is
** done in a single transaction.
*/

create procedure increase_price

as

/* start the transaction */

begin transaction
/* first update prices > $60 */
update titles
set price = price * 1.05
where price > $60

/* next, prices between $30 and $60 */

update titles
set price = price * 1.10
where price > $30 and price <= $60

/* and finally prices <= $30 */

update titles
set price = price * 1.20

Performance and Tuning Series: Query Processing and Abstract Plans 269
Comparing performance with and without cursors

where price <= $30

/* commit the transaction */

commit transaction

return

Sample stored procedure with a cursor

This procedure performs the same changes to the underlying table as the
procedure written without a cursor, but it uses cursors instead of set-oriented
programming. As each row is fetched, examined, and updated, a lock is held
on the appropriate data page. Also, as the comments indicate, each update
commits as it is made, since there is no explicit transaction.
/* Same as previous example, this time using a
** cursor. Each update commits as it is made.
*/
create procedure increase_price_cursor
as
declare @price money

/* declare a cursor for the select from titles */

declare curs cursor for
select price
from titles
for update of price

/* open the cursor */

open curs

/* fetch the first row */

fetch curs into @price

/* now loop, processing all the rows

** @@sqlstatus = 0 means successful fetch
** @@sqlstatus = 1 means error on previous fetch
** @@sqlstatus = 2 means end of result set reached
*/
while (@@sqlstatus != 2)
begin
/* check for errors */
if (@@sqlstatus = 1)
begin
print "Error in increase_price"

270 Adaptive Server Enterprise

 CHAPTER 8 Optimization for Cursors

return
end

/* next adjust the price according to the

** criteria
*/
if @price > $60
select @price = @price * 1.05
else
if @price > $30 and @price <= $60
select @price = @price * 1.10
else
if @price <= $30
select @price = @price * 1.20

/* now, update the row */

update titles
set price = @price
where current of curs

/* fetch the next row */

fetch curs into @price
end

/* close the cursor and return */

close curs
return
Which procedure do you think will have better performance, one that performs
three table scans or one that performs a single scan via a cursor?

Cursor versus noncursor performance comparison

Table 8-2 shows statistics gathered against a 5000-row table. The cursor code
takes over 4 times longer, even though it scans the table only once.
Table 8-2: Sample execution times against a 5000-row table
Procedure Access method Time
increase_price Uses three table scans 28 seconds
increase_price_cursor Uses cursor, single table scan 125 seconds

Results from tests like these can vary widely. They are most pronounced on
systems that have busy networks, a large number of active database users, and
multiple users accessing the same table.

Performance and Tuning Series: Query Processing and Abstract Plans 271
Locking with read-only cursors

In addition to locking, cursors involve more network activity than set

operations and incur the overhead of processing instructions. The application
program needs to communicate with Adaptive Server regarding every result
row of the query. This is why the cursor code took much longer to complete
than the code that scanned the table three times.
Cursor performance issues include:
• Locking at the page and table level
• Network resources
• Overhead of processing instructions
If there is a set-level programming equivalent, it may be preferable, even if it
involves multiple table scans.

Locking with read-only cursors

Here is a piece of cursor code you can use to display the locks that are set up at
each point in the life of a cursor. The following example uses an allpages-
locked table. Execute the code in Figure 8-4, and pause at the arrows to execute
sp_lock and examine the locks that are in place.

Figure 8-4: Read-only cursors and locking experiment input

declare curs1 cursor for

select au_id, au_lname, au_fname
from authors
where au_id like ’15%’
for read only
go
open curs1
go
fetch curs1
go
fetch curs1
go 100
close curs1
go
deallocate cursor curs1
go

Table 8-3 shows the results.

272 Adaptive Server Enterprise

 CHAPTER 8 Optimization for Cursors

Table 8-3: Locks held on data and index pages by cursors

Event Data page
After declare No cursor-related locks.
After open Shared intent lock on authors.
After first fetch Shared intent lock on authors and shared page lock on a page in authors.
After 100 fetches Shared intent lock on authors and shared page lock on a different page in authors.
After close No cursor-related locks.

If you issue another fetch command after the last row of the result set has been
fetched, the locks on the last page are released, so there will be no cursor-
related locks.
With a data-only-locked table:
• If the cursor query runs at isolation level 1, and read committed with lock is
set to 0, you do not see any page or row locks. The values are copied from
the page or row, and the lock is immediately released.
• If read committed with lock is set to 1 or if the query runs at isolation level
2 or 3, you see either shared page or shared row locks at the point that
Table 8-3 indicates shared page locks. If the table uses datarows locking,
the sp_lock report includes the row ID of the fetched row.

Isolation levels and cursors

The query plan for a cursor is compiled and optimized when the cursor is
opened. You cannot open a cursor and then use set transaction isolation level to
change the isolation level at which the cursor operates.
Since cursors using isolation level 0 are compiled differently from those using
other isolation levels, you cannot open a cursor at isolation level 0 and open or
fetch from it at level 1 or 3. Similarly, you cannot open a cursor at level 1 or 3
and then fetch from it at level 0. Attempts to fetch from a cursor at an
incompatible level result in an error message.
Once the cursor has been opened at a particular isolation level, you must
deallocate the cursor before changing isolation levels. The effects of changing
isolation levels while the cursor is open are as follows:
• Attempting to close and reopen the cursor at another isolation level fails
with an error message.

Performance and Tuning Series: Query Processing and Abstract Plans 273
Partitioned heap tables and cursors

• Attempting to change isolation levels without closing and reopening the

cursor has no effect on the isolation level in use and does not produce an
error message.
You can include an at isolation clause in the cursor to specify an isolation level.
The cursor in the example below can be declared at level 1 and fetched from
level 0 because the query plan is compatible with the isolation level:
declare cprice cursor for
select title_id, price
from titles
where type = "business"
at isolation read uncommitted

Partitioned heap tables and cursors

A cursor scan of an unpartitioned heap table can read all data up to and
including the final insertion made to that table, even if insertions took place
after the cursor scan started.
If a heap table is partitioned, data can be inserted into one of the many page
chains. The physical insertion point may be before or after the current position
of a cursor scan. This means that a cursor scan against a partitioned table is not
guaranteed to scan the final insertions made to that table.

Note If cursor operations require all inserts to be made at the end of a single
page chain, do not partition the table used in the cursor scan.

Optimizing tips for cursors

Here are several optimizing tips for cursors:
• Optimize cursor selects using the cursor, not an ad hoc query.
• Use union or union all instead of or clauses or in lists.
• Declare the cursor’s intent.
• Specify column names in the for update clause.

274 Adaptive Server Enterprise

 CHAPTER 8 Optimization for Cursors

• Fetch more than one row if you are returning rows to the client.
• Keep cursors open across commits and rollbacks.
• Open multiple cursors on a single connection.

Optimizing for cursor selects using a cursor

A standalone select statement may be optimized very differently than the same
select statement in an implicitly or explicitly updatable cursor. When you are
developing applications that use cursors, always check your query plans and
I/O statistics using the cursor, rather than using a standalone select. In
particular, index restrictions of updatable cursors require very different access
methods.

Using union instead of or clauses or in lists

Cursors cannot use the dynamic index of row IDs generated by the OR strategy.
Queries that use the OR strategy in standalone select statements usually
perform table scans using read-only cursors. Updatable cursors may need to
use a unique index and still require access to each data row, in sequence, in
order to evaluate the query clauses.
A read-only cursor using union creates a worktable when the cursor is declared,
and sorts it to remove duplicates. Fetches are performed on the worktable. A
cursor using union all can return duplicates and does not require a worktable.

Declaring the cursor’s intent

Always declare a cursor’s intent: read-only or updatable. This gives you greater
control over concurrency implications. If you do not specify the intent,
Adaptive Server decides for you, and very often it chooses updatable cursors.
Updatable cursors use update locks, thereby preventing other update locks or
exclusive locks. If the update changes an indexed column, the optimizer may
need to choose a table scan for the query, resulting in potentially difficult
concurrency problems. Be sure to examine the query plans for queries that use
updatable cursors.

Performance and Tuning Series: Query Processing and Abstract Plans 275
Optimizing tips for cursors

Specifying column names in the for update clause

Adaptive Server acquires update locks on the pages or rows of all tables that
have columns listed in the for update clause of the cursor select statement. If the
for update clause is not included in the cursor declaration, all tables referenced
in the from clause acquire update locks.
The following query includes the name of the column in the for update clause,
but acquires update locks only on the titles table, since price is mentioned in the
for update clause. The table uses allpages locking. The locks on authors and
titleauthor are shared page locks:

declare curs3 cursor

for
select au_lname, au_fname, price
from titles t, authors a,
titleauthor ta
where advance <= $1000
and t.title_id = ta.title_id
and a.au_id = ta.au_id
for update of price
Table 8-4 shows the effects of:
• Omitting the for update clause entirely—no shared clause
• Omitting the column name from the for update clause
• Including the name of the column to be updated in the for update clause
• Adding shared after the name of the titles table while using for update of
price

In this table, the additional locks, or more restrictive locks for the two versions
of the for update clause are emphasized.

276 Adaptive Server Enterprise

 CHAPTER 8 Optimization for Cursors

Table 8-4: Effects of for update clause and shared on cursor locking
Clause titles authors titleauthor
None sh_page on index
sh_page on data sh_page on data sh_page on data
for update updpage on index updpage on index
updpage on data updpage on data updpage on data
for update of sh_page on index
price updpage on data sh_page on data sh_page on data
for update of sh_page on index
price sh_page on data sh_page on data sh_page on data
+ shared

Using set cursor rows

The SQL standard specifies a one-row fetch for cursors, which wastes network
bandwidth. Using the set cursor rows query option and Open Client’s
transparent buffering of fetches, you can improve performance:
ct_cursor(CT_CURSOR_ROWS)
Be careful when you choose the number of rows returned for frequently
executed applications using cursors—tune them to the network.
See the Performance and Tuning Series: Basics for an explanation of this
process.

Keeping cursors open across commits and rollbacks

ANSI closes cursors at the conclusion of each transaction. Transact-SQL
provides the set option close on endtran for applications that must meet ANSI
behavior. By default, however, this option is turned off. Unless you must meet
ANSI requirements, leave this option off to maintain concurrency and
throughput.
If you must be ANSI-compliant, decide how to handle the effects on Adaptive
Server. Should you perform a lot of updates or deletes in a single transaction?
Or should you keep the transactions short?

Performance and Tuning Series: Query Processing and Abstract Plans 277
Optimizing tips for cursors

If you choose to keep transactions short, closing and opening the cursor can
affect throughput, since Adaptive Server needs to rematerialize the result set
each time the cursor is opened. If you choose to perform more work in each
transaction, this can cause concurrency problems, since the query holds locks.

Opening multiple cursors on a single connection

Some developers simulate cursors by using two or more connections from DB-
Library™. One connection performs a select and the other performs updates or
deletes on the same tables. This has very high potential to create application
deadlocks. For example:
• Connection A holds a shared lock on a page. As long as there are rows
pending from Adaptive Server, a shared lock is kept on the current page.
• Connection B requests an exclusive lock on the same pages and then waits.
• The application waits for Connection B to succeed before invoking
whatever logic is needed to remove the shared lock. But this never
happens.
Since Connection A never requests a lock that is held by Connection B, this is
not a server-side deadlock.

278 Adaptive Server Enterprise

CH A PTE R 9 Query Processing Metrics

Topic Page
Overview 279
Executing QP metrics 280
Accessing metrics 280
Using metrics 282
Clearing metrics 284
Restricting query metrics capture 285
Understanding the UID in sysquerymetrics 286

Overview
Query processing (QP) metrics identify and compare empirical metric
values in query execution. When a query is executed, it is associated with
a set of defined metrics that are the basis for comparison in QP metrics.
Captured metrics include:
• CPU execution time – the time, in milliseconds, it takes to execute the
query.
• Elapsed time – the time, in milliseconds, from after the compile to the
end of the execution.
• Logical I/O – the number of logical I/O reads.
• Physical I/O – the number of physical I/O reads.
• Count – the number of times a query is executed.
• Abort count – the number of times a query is aborted by the resource
governor due to a resource limit being exceeded.
Each metric, except count and abort count, has three values: minimum,
maximum, and average.

Performance and Tuning Series: Query Processing and Abstract Plans 279
Executing QP metrics

Executing QP metrics
You can activate and use QP metrics at the server level or at the session level.
At the server level, use sp_configure with the enable metrics capture option. The
QP metrics for ad hoc statements are captured directly into a system catalog,
while the QP metrics for statements in a stored procedure are saved in a
procedure cache. When the stored procedure or query in the statement cache is
flushed, the respective captured metrics are written to the system catalog.
sp_configure "enable metrics capture", 1
At a session level, use set metrics_capture on/off:
set metrics_capture on/off

Accessing metrics
QP metrics are always captured in the default group, which is group 1 in each
respective database. Use sp_metrics ‘backup’ to move saved QP metrics from
the default running group to a backup group. Access metric information using
a select statement with order by against the sysquerymetrics view. See
“sysquerymetrics view” on page 280 for details.
You can also use a data manipulation language (DML) statement to sort the
metric information and identify the specific queries for evaluation. See Chapter
2, “Understand Component Integration Services,” in the Component
Integration Services Users Guide, which is part of the Adaptive Server
Enterprise documentation set. .

sysquerymetrics view
Field Definition
uid User ID
gid Group ID
id Unique ID
hashkey Hash key over the SQL query text
sequence Sequence number for a row when multiple rows are required for the text of the SQL code
exec_min Minimum execution time

280 Adaptive Server Enterprise

 CHAPTER 9 Query Processing Metrics

Field Definition
exec_max Maximum execution time
exec_avg Average execution time
elap_min Minimum elapsed time
elap_max Maximum elapsed time
elap_avg Average elapsed time
lio_min Minimum logical I/O
lio_max Maximum logical I/O
lio_avg Average logical I/O
pio_min Minimum physical I/O
pio_max Maximum physical I/O
pio_avg Average physical I/O
cnt Number of times the query has been executed
abort_cnt Number of times a query is aborted by the resource governor when a resource limit is exceeded
qtext Query text

Average values in this view are calculated using:

new_avg = (old_avg * old_count + new_value )/ (old_count + 1) = old_avg +
round((new_value - old_avg)/(old_count + 1))
This is an example of the sysquerymetrics view:
select * from sysquerymetrics

uid gid hashkey id sequence exec_min

exec_max exec_avg elap_min elap_max elap_avg lio_min
lio_max lio_avg pio_min pio_max pio_avg cnt abort_cnt
qtext
----------- ----------- ----------- ----------- ----------- -----------
----------- ----------- ----------- ----------- ----------- -----------
----------- ----------- ----------- ----------- ----------- -----------
-------------------------------------------------------------------------
1 1 106588469 480001710 0 0
0 0 16 33 25 4
4 4 0 4 2 2 0
select distinct c1 from t_metrics1 where c2 in (select c2 from t_metrics2)
The above example displays a record for a SQL statement. The query text of
the statement is select distinct c1 from t_metrics1 where c2 in (select c2 from
t_metrics2):

• This statement has been executed twice so far (cnt = 2).

Performance and Tuning Series: Query Processing and Abstract Plans 281
Using metrics

• The minimum elapsed time is 16 milliseconds; the maximum elapsed time

is 33 milliseconds, and the average elapsed time is 25 milliseconds
• All the execution times are 0, and this may be due to the CPU execution
time being less than 1 millisecond.
• The maximum physical I/O is 4, which is consistent with the maximum
logical I/O. However, the minimum physical I/O is 0 because data is
already in cache in the second run. The logical I/O, at 4, should be static
whether or not the data is in memory

Using metrics
Use the information produced by QP metrics to identify:
• Query performance regression
• Most expensive query in a batch of running queries
• Most frequently run queries
When you have information on the queries that may be causing problems, you
can tune the queries to increase efficiency.
For example, identifying and fine-tuning an expensive query may be more
effective than tuning the cheaper ones in the same batch.
You can also identify the queries that are run most frequently, and fine-tune
them to increase efficiency.
Turning on query metrics may involve extra I/O for every query executed, so
there may be performance impact. However, also consider the benefits
mentioned above. You may want to gather statistical information from
monitoring tables instead of turning on metrics.
Both QP metrics and monitoring tables can be used to gather statistical
information. However, you can use QP metrics instead of the monitoring tables
to gather aggregated historical query information in a persistent catalog, rather
than have transient information from the monitor tables.

282 Adaptive Server Enterprise

 CHAPTER 9 Query Processing Metrics

Examples
You can use QP metrics to identify specific queries for tuning and possible
regression on performance.

Identifying the most expensive statement

Typically, to find the most expensive statement as the candidate for tuning,
sysquerymetrics provides CPU execution time, elapsed time, logical IO, and
physical I/O as options for measure. For example, a typical measure is based
on logical I/O. Use the following query to find the statements that incur too
many IOs as the candidates for tuning:
select lio_avg, qtext from sysquerymetrics order by lio_avg
lio_avg qtext
-----------
------------------------------------------------------
2
select c1, c2 from t_metrics1 where c1 = 333
4
select distinct c1 from t_metrics1 where c2 in (select c2 from t_metrics2)
6
select count(t_metrics1.c1) from t_metrics1, t_metrics2,
t_metrics3 where (t_metrics1.c2 = t_metrics2.c2 and
t_metrics2.c2 = t_metrics3.c2 and t_metrics3.c3 = 0)
164
select min(c1) from t_metrics1 where c2 in (select t_metrics2.c2 from
t_metrics2, t_metrics3 where (t_metrics2.c2 = t_metrics3.c2 and t_metrics3.c3
= 1))

(4 rows affected)
The best candidate for tuning can be seen in the last statement of the above
results, which has the biggest value (164) for average logical IO.

Identifying the most frequently used statement for tuning

If a query is used frequently, fine-tuning may improve its performance. Identify
the most frequently used query using the select statement with order by:
select elap_avg, cnt, qtext from sysquerymetrics order by cnt
elap_avg cnt
qtext
----------- -----------
--------------------------------------------------
0 1

Performance and Tuning Series: Query Processing and Abstract Plans 283
Clearing metrics

select c1, c2 from t_metrics1 where c1 = 333

16 2

select distinct c1 from t_metrics1 where c2 in (select c2 from t_metrics2)

24 3

select min(c1) from t_metrics1 where c2 in (select t_metrics2.c2 from

t_metrics2, t_metrics3 where (t_metrics2.c2 = t_metrics3.c2 and t_metrics3.c3
= 1))

78 4

select count(t_metrics1.c1) from t_metrics1, t_metrics2, t_metrics3 where

(t_metrics1.c2 = t_metrics2.c2 and t_metrics2.c2 = t_metrics3.c2 and
t_metrics3.c3 = 0)

(4 rows affected)
The best candidate for tuning can be seen in the last statement of the above
results, which has the biggest value (78).

Identifying possible performance regression

In some cases, when a server is upgraded to a newer version, QP metrics may
be useful for comparing performance. To identify queries that may have some
degradation, after a server-version upgrade:
1 Back up the QP metrics from the old server into a backup group:
sp_metrics ‘backup’, ’@gid’
2 Enable QP metrics on the new server:
sp_configure “enable metrics capture”, 1
3 Compare QP metrics output from the old and new servers to identify any
queries that may have regression problems.

Clearing metrics
Use sp_metrics ‘flush’ to flush all aggregated metrics in memory to the system
catalog.The aggregated metrics for all statements in memory are set to zero.
sp_metrics ‘drop’, ‘@gid’ [, ‘@id'

284 Adaptive Server Enterprise

 CHAPTER 9 Query Processing Metrics

To remove one entry, use:

sp_metrics ‘drop’, ‘<gid>’, ‘<id>’
You can also use filter to remove QP metrics from the system catalog, based on
some metrics conditions:
sp_metrics ‘filter’, ‘@gid’, [, ‘@predicate’]
This example deletes all QP metrics in group 1 where lio_max < 100:
sp_metrics ‘filter’,’1’,’lio_max < 100'

Restricting query metrics capture

These configuration parameters set the query metrics threshold for capture into
the catalog:
• metrics lio max

• metrics pio max

• metrics elap max

• metrics exec max

These parameters let you filter out trivial metrics before writing metrics
information to the catalog.
By default, these configuration parameters are set to 0 (off).
For example, to not capture those query plans for which lio is less than 10, use:
sp_configure ‘metrics lio max’, 10
If you do not set any of these configuration parameters, Adaptive Server
captures the query metrics to the system tables. However, if you set any of
these configuration parameters, Adaptive Server uses only those nonzero
configuration parameters as thresholds for determining whether to capture
query metrics.
For example, if you set metrics elap max to a non-zero value, but no others,
query metrics are captured only if the elapsed time is bigger than the
configured value. Because the other three configuration parameters are set to
0, they do not act as thresholds for capturing metrics.

Performance and Tuning Series: Query Processing and Abstract Plans 285
Understanding the UID in sysquerymetrics

Understanding the UID in sysquerymetrics

The user ID (UID) of sysquerymetrics is 0 when all table names in a query that
are not qualified by user name are owned by the database owner.
Example 1 select * from t1 where cl = 1
t1 is owned by database owner and is shared by different users. 0 is the UID for
the entry into sysquerymetrics no matter which user issues the query.
Example 2 select * from t2 where cl = 1
In this case, t2 is owned by user1. user1’s UID is used for the entry in
sysquerymetrics, since t2 is unqualified and is not owned by the database
owner.
Example 3 select * from u1.t3 where cl = 1
Here, t3 is owned by u1 and is qualified by u1, so UID 0 is used.
This increases the sharing of metrics between user IDs to reduce the number of
entries in sysqueryplans. Aggregation of metrics for identical queries with
different user IDs is done automatically. Turn on trace flag 15361 to use the
UID of the user who issues the query.

Note QP metrics for insert...selec, /update, delete statements are captured when
at least one table is involved. CIS-related queries and insert...values statements
are not included.

286 Adaptive Server Enterprise

CH A PTE R 1 0 Using Statistics to Improve
Performance

Accurate statistics are essential to query optimization. In some cases,

adding statistics for columns that are not leading index keys also improves
query performance. This chapter explains how and when to use the
commands that manage statistics.
Topic Page
Statistics maintained in Adaptive Server 287
Importance of statistics 288
Updating statistics 290
Automatically updating statistics 294
Configuring automatic update statistics 297
Column statistics and statistics maintenance 300
Creating and updating column statistics 302
Choosing step numbers for histograms 306
Scan types, sort requirements, and locking 307
Using the delete statistics command 310
When row counts may be inaccurate 311

Statistics maintained in Adaptive Server

These key optimizer statistics are maintained in Adaptive Server:
• Statistics per partition – table row count; table page count. An
unpartitioned table is considered to have one partition for the
purposes of the systabstats catalog. Statistics per partition can be
found in systabstats.

Performance and Tuning Series: Query Processing and Abstract Plans 287
Importance of statistics

• Statistics per index: index row count – index height; index leaf page count.
A local index has a separate systabstats row for each index partition. A
global index, which is considered a partitioned index with one partition,
has one systabstats row. Statistics per index: index row count can be found
in systabstats.
• Statistics per column: data distribution. Statistics per column be found in
sysstatistics.

• Statistics per group of column – density information. Statistics per group

of column can be found in sysstatistics.
• Statistics per partition –
• Column statistics – data distribution per column; density per group of
columns. Column statistics can be found in sysstatistics.
Throughout this chapter, density is a statistical measurement of the uniqueness
of a given column’s values, and a histogram is a statistical representation of the
distribution of values of a given column of the relation.

Importance of statistics
The Adaptive Server cost-based optimizer uses statistics about the tables,
indexes, partitions, and columns named in a query to estimate query costs. It
chooses the access method that the optimizer determines has the least cost. But
this cost estimate cannot be accurate if statistics are not accurate.
Some statistics, such as the number of pages or rows in a table, are updated
during query processing. Other statistics, such as the histograms on columns,
are updated only when you execute update statistics, or when indexes are
created.
If your query is performing slowly and you seek help from Technical Support
or a Sybase newsgroup on the Internet, one of the first questions you are likely
be asked is “Did you run update statistics?” Use the optdiag command to see
when update statistics was last run for each column on which statistics exist:
Last update of column statistics: Aug 31 2004
4:14:17:180PM

288 Adaptive Server Enterprise

 CHAPTER 10 Using Statistics to Improve Performance

Another command you may need for statistics maintenance is delete statistics.
Dropping an index does not drop the statistics for that index. If the distribution
of keys in the columns changes after the index is dropped, but the statistics are
still used for some queries, the outdated statistics can affect query plans.
Histogram statistics from a global index are more accurate than histogram
statistics generated by a local index. For a local index, statistics are created on
each partition, and are then merged to create a global histogram using
approximations as to how overlapping histogram cells from each partition
should be combined. With a global index, the merge step, with merging
estimates, does not occur. In most cases, there is no issue with update statistics
on a local index. However, if there are significant estimation errors in queries
involving partitioned tables, histogram accuracy can be improved by creating
and dropping a global index on a column rather than updating the statistics on
a local index.

Nonbinary character set histogram interpolation

Adaptive Server versions 15.0.2 and later allow selectivity estimates to have
the same accuracy as the binary character set, without requiring an excessive
number of histogram steps. This benefits queries like the following, which uses
range predicates:
select * from t1 where charcolumn > "LMC0021" and
charcolumn <= "LMC0029"
If ranges specified falls into the same histogram cell, Adaptive Server can
much more accurately estimate this selectivity.
In versions of Adaptive Server earlier than 15.0.2, only the default binary
character set benefited from histogram interpolation, which is used to estimate
the selectivity of range predicates. For all other character sets, Adaptive Server
made a selectivity estimate of 50 percent for a histogram cell. This typically
required Adaptive Server to use a large number of histogram cells for character
column histograms to reduce the error associated with this estimate.

Performance and Tuning Series: Query Processing and Abstract Plans 289
Updating statistics

Updating statistics
The update statistics command updates column-related statistics such as
histograms and densities. Statistics must be updated on those columns where
the distribution of keys in the index changes in ways that affect the use of
indexes for your queries.
Running update statistics requires system resources. Like other maintenance
tasks, Sybase recommends that you schedule it during at times when the load
on the server is light. In particular, update statistics requires table scans or leaf-
level scans of indexes, may increase I/O contention, may use the CPU to
perform sorts, and uses the data and procedure caches. Use of these resources
can adversely affect queries running on the server.
Using the sampling feature can reduce resource requirements and allow more
flexibility when running this task.

Note Sampling does not affect the update statistics table_name index_name
parameters]. Sampling affects only update index statistics and update all
statistics on unindexed columns and update statistics table_name (column_list).

In addition, some update statistics commands require shared locks, which may
block updates. See “Scan types, sort requirements, and locking” on page 307.
You can also configure Adaptive Server to automatically run update statistics
at times that have minimal impact on the system resources. See “Automatically
updating statistics” on page 294.

Adding statistics for unindexed columns

When you create an index, a histogram is generated for the leading column in
the index. Examples in earlier chapters have shown how statistics for other
columns can increase the accuracy of optimizer statistics.
Consider adding statistics for virtually all columns that are frequently used as
search arguments, as long as your maintenance schedule allows time to keep
these statistics up to date.
In particular, adding statistics for minor columns of composite indexes can
greatly improve cost estimates when those columns are used in search
arguments or joins along with the leading index key.

290 Adaptive Server Enterprise

 CHAPTER 10 Using Statistics to Improve Performance

Limitations for updating statistics on proxy tables and views

If you create indexes on proxy tables, you may experience performance
degradation; you cannot run create index, because Adaptive Server cannot
propogate the command to the remote view. However, if you map the proxy
table to a user table, Adaptive Server can optimize the query plan by creating
indexes on the proxy tables’ join column.
If the number of rows returned by the view varies drastically from one
execution to another, performance degrades when you gather statistics on
proxy tables.
If you are using proxy tables or views, Sybase makes these recommendations:
• Do not run update statistics on proxy tables defined on views.
• If you ran update statistics on a proxy table defined on a view, increase
performance by dropping, then re-creating the proxy table. You cannot use
delete statistics to remove the current statistics, because systabstats retains
the original row count.
• Convert the view to a permanent table.
• Try using the command line trace flag 318 (force reformatting).

update statistics commands

The update statistics commands create statistics, if there are none for a
particular column, or replaces existing statistics. Statistics are stored in the
systabstats and sysstatistics system tables:
update statistics table_name
[[ partition data_partition_name ] [ (column_list ) ] |
index_name [ partition index_partition_name ] ]
[ using step values ]
[ with consumers = consumers] [, sampling=percent]

update index statistics

table_name [[ partition data_partition_name ] |
[ index_name [ partition index_partition_name ] ] ]
[ using step values ]
[ with consumers = consumers] [, sampling=percent]

update all statistics table_name

[ partition data_partition_name ]
[ sp_configure histogram tuning factor, <value>

update table statistics

Performance and Tuning Series: Query Processing and Abstract Plans 291
Updating statistics

table_name [partition data_partition_name ]

delete [ shared ] statistics table_name

[ partition data_partition_name ]
[( column_name[, column_name ] ...)]
• For update statistics:
• table_name – generates statistics for the leading column in each index
on the table.
• table_name index_name – generates statistics for all columns of the
index.
• partition_name – generates statistics for only this partition.
• partition_name table_name (column_name) – generates statistics for
this column of this table on this partition.
• table_name (column_name) – generates statistics for only this
column.
• table_name (column_name, column_name...) – generates a histogram
for the leading column in the set, and multicolumn density values for
the prefix subsets.
• using step values – identifies the number of steps used. The default is
20 steps. To change the default number of steps, use sp_configure.
• sampling = percent – the numeric value of the sampling percentage,
such as 05 for 5%, 10 for 10%, and so on. The sampling integer is
between zero (0) and one hundred (100).
• For update index statistics:
• table_name – generates statistics for all columns in all indexes on the
table.
• partition_name table_name – generates statistics for all columns in all
indexes for the table on this partition.
• table_name index_name – generates statistics for all columns in this
index.
• For update all statistics:
• table_name – generates statistics for all columns of a table.
• table_name partition_name – generates statistics for all columns of a
table on a partition.

292 Adaptive Server Enterprise

 CHAPTER 10 Using Statistics to Improve Performance

• using step values – identifies the number of steps used. The default is
20 steps. To change the default number of steps, use sp_configure.
sp_configure syntax includes the histogram tuning factor, which allows
a greater selection of the number of histogram steps. The default value
for histogram tuning factor is 20. See Chapter 5,4“Setting
Configuration Parameters” in System Administration Guide: Volume
1 for information about sp_configure.

Using sampling for update statistics

The optimizer for Adaptive Server uses the statistics on a database to set up and
optimize queries. To generate optimal results, the statistics must be as current
as possible.
Run the update statistics commands against data sets, such as tables, to update
information about the distribution of key values in specified indexes or
columns, for all columns in an index, or for all columns in a table. The
commands revise histograms and density values for column-level statistics.
The results are then used by the optimizer to calculate the best query plan.
Run update statistics using a sampling method, which can reduce the I/O and
time when your maintenance window is small and the data set is large. If you
are updating a large data set or table that is in constant use, being truncated and
repopulated, you may want to do a statistical sampling to reduce the time and
the size of the I/O. Because sampling does not update the density values, run a
full update statistics prior to using sampling for an accurate density value.
Use caution with sampling, since the results are not fully accurate. Balance
changes to histogram values against the savings in I/O.
Sampling does not update density values created by a non-sampling update
statistics command. Since the density changes very slowly, replacing an
accurate density with an approximation calculated by sampling usually does
not improve the estimate. Density values created by a sampling update statistics
command is updated. Sybase recommends that you use that one non-sampling
update statistics command to establish an accurate density, which can be
followed by numerous sampling update statistics commands. To have sampling
update statistics update the density, delete the column statistics before using
update statistics with sampling.

When you are deciding whether or not to use sampling, consider the size of the
data set, the time constraints you are working with, and if the histogram
produced is as accurate as needed.

Performance and Tuning Series: Query Processing and Abstract Plans 293
Automatically updating statistics

The percentage to use when sampling depends on your needs. Test various
percentages until you receive a result that reflects the most accurate
information on a particular data set; for example:
update statistics authors(auth_id) with sampling = 5 percent
Set server-wide sampling percent using:
sp_configure 'sampling percent', 5
This command sets a server-wide sampling of 5% for update statistics that
allows you to execute update statistics without the sampling syntax. The
percentage can be between zero (0) and one hundred (100) percent.

Automatically updating statistics

The Adaptive Server cost-based query processor estimates the query costs
using the statistics for the tables, indexes, and columns named in a query. The
query processor chooses the access method it determines has the least cost.
However, this cost estimate cannot be accurate if the statistics are not accurate.
You can run update statistics to ensure that the statistics are current, however,
running update statistics has an associated cost because it consumes system
resources such as CPU, buffer pools, sort buffers, and procedure cache.
You can set update statistics to run automatically when it best suits your site and
avoid running it at times that hamper your system. Use the datachange function
to determine the best time for you to run update statistics. datachange also helps
to ensure that you do not unnecessarily run update statistics. You can use the
Job Scheduler templates to determine the objects, schedules, priority, and
datachange thresholds that trigger update statistics, which ensures that critical
resources are used only when the query processor generates more efficient
plans.
Because update statistics is a resource-intensive task, base the decision to run
update statistics on a specific set of criteria. Key parameters that can help you
determine a good time to run update statistics include:
• How much data characteristics have changed since you last ran update
statistics. This is known as the datachange parameter.

• Whether there are sufficient resources available to run update statistics.

These include resources such as the number of idle CPU cycles and
making sure that critical online activity does not occur during update
statistics.

294 Adaptive Server Enterprise

 CHAPTER 10 Using Statistics to Improve Performance

Data change is a key metric that helps you measure the amount of altered data
since you last ran update statistics, and is tracked by the datachange function.
Using this metric and the criteria for resource availability, you can automate the
process of running update statistics. Job Scheduler includes a mechanism to
automatically run update statistics. and also includes customizable templates
you can use to determine when to run update statistics. These inputs include all
parameters to update statistics, the datachange threshold values, and the time to
run update statistics. Job Scheduler runs update statistics at a low priority so it
does not affect critical jobs that are running concurrently.

datachange function
The datachange function measures the amount of change in the data
distribution since update statistics last ran. Specifically, it measures the number
of inserts, updates, and deletes that have occurred on the given object, partition,
or column, and helps you determine if running update statistics would benefit
the query plan.
The syntax for datachange is:
select datachange(object_name, partition_name, colname)
Where:
• object_name – is the object name. This object is assumed to be in the
current database. The object_name cannot be null.
• partition_name – is the data partition name. This can be a null value.
• colname – is the column name for which the datachange is requested. This
can be a null value.
These parameters are all required.
datachange is expressed as a percentage of the total number of rows in the table
or partition (if the partition is specified). The percentage value can be greater
than 100 percent because the number of changes to an object can be much
greater than the number of rows in the table, particularly when the number of
deletes and updates to a table is very high.

The following set of examples illustrate the various uses for the datachange
function. The examples use the following:
• Object name is “O.”
• Partition name is “P.”

Performance and Tuning Series: Query Processing and Abstract Plans 295
Automatically updating statistics

• Column name is “C.”

Passing a valid object, The value reported when you include the object, partition, and column name is
partition, and column determined by the datachange value for the specified column in the specified
name
partition divided by the number of rows in the partition. The result is expressed
as a percentage:
datachange = 100 * (data change value for column C/ rowcount (P))
Using null partition If you include a null partition name, the datachange value is determined by the
names sum of the datachange value for the column across all partitions divided by the
number of rows in the table. The result is expressed as a percentage:
datachange = 100 * (Sum(data change value for (O, P(1-N) , C))/rowcount(O)
Where P(1-N) indicates that the value is summed over all partitions.
Using null column If you include null column names, the value reported by datachange is
names determined by the maximum value of the datachange for all columns that have
histograms for the specified partition divided by the number of rows in the
partition. The result is expressed as a percentage:
datachange = 100 * (Max(data change value for (O, P, Ci))/rowcount(P)
Where the value of i varies through the columns with histograms (for example,
formatid 102 in sysstatistics).

Null partition and If you include null partition and column names, the value of datachange is
column names determined by the maximum value of the datachange for all columns that have
histograms summed across all partitions divided by the number of rows in the
table. The result is expressed as a percentage:
datachange = 100 * ( Max(data change value for (O, NULL, Ci))/rowcount(O)
Where i is 1 through the total number of columns with histograms (for
example, formatid 102 in sysstatistics).
This illustrates datachange gathering statistics:
create table matrix(col1 int, col2 int)
go
insert into matrix values (234, 560)
go
update statistics matrix(col1)
go
insert into matrix values(34,56)
go
select datachange ("matrix", NULL, NULL)
go

296 Adaptive Server Enterprise

 CHAPTER 10 Using Statistics to Improve Performance

------
50.000000
The number of rows in matrix is two. The amount of data that has changed since
the last update statistics command is 1, so the datachange percentage is 100 *
1/2 = 50 percent.
datachange counters are all maintained in memory. These counters are
periodically flushed to disk by the housekeeper or when you run sp_flushstats.

Configuring automatic update statistics

Automatically update statistics by:
• Defining update statistics jobs with Job Scheduler
• Defining update statistics jobs as part of the self-managed installation
• Creating user-defined scripts
Creating user-defined scripts is not discussed in this document.

Using Job Scheduler to update statistics

Job Scheduler includes the update statistics template, which you can use to
create a job that runs update statistics on a table, index, column, or partition.
The datachange function determines when the amount of change in a table or
partition has reached the predefined threshold. You determine the value for this
threshold when you configure the template.
Templates:
• Run update statistics on specific tables, partitions, indexes, or columns.
The templates allow you to define the value for datachange at which you
want update statistics to run.
• Run update statistics at the server level, which configures Adaptive Server
to sweep through the available tables in all databases on the server and
update statistics on all the tables, based on the threshold you determined
when creating your job.
To configure Job Scheduler to automate the process of running update statistics
(the chapters listed are from the Job Scheduler Users Guide:

Performance and Tuning Series: Query Processing and Abstract Plans 297
Configuring automatic update statistics

1 Install and set up Job Scheduler (Chapter 2, “Configuring and Running Job
Scheduler”)
2 Install the stored procedures required for the templates (Chapter 4, “Using
Templates to Schedule Jobs”).
3 Install the templates. Job Scheduler provides the templates specifically for
automating update statistics (Chapter 4, “Using Templates to Schedule
Jobs”).
4 Configure the templates. The templates for automating update statistics
are in the Statistics Management folder.
5 Schedule the job. After you have defined the index, column, or partition
you want tracked, you can also create a schedule that determines when
Adaptive Server runs the job, making sure that update statistics is run only
when it does not impact performance.
6 Identify success or failure. The Job Scheduler infrastructure allows you to
identify success or failure for the automated update statistic.
The template allows you to supply values for the various options of the update
statistics command such as sampling percent, number of consumers, steps, and
so on. Optionally, you can also provide threshold values for the datachange
function, page count, and row count. If you include these optional values, they
are used to determine when and if Adaptive Server should run update statistics.
If the current values for any of the tables, columns, indexes, or partitions
exceed the threshold values, Adaptive Server issues update statistics. Adaptive
Server detects that update statistics has been run on a column. Any query
referencing that table in the procedure cache is recompiled before the next
execution.
When does Adaptive There are many forms of the update statistics command (update statistics,
Server run update update index statistics, and so on); use these different forms depending on your
statistics?
needs.
You must specify three thresholds: rowcount, pagecount, and datachange.
Although values of NULL or 0 are ignored, these values do not prevent the
command from running.
Table 10-1 describes the circumstances under which Adaptive Server
automatically runs update statistics, based on the parameter values you provide.
Table 10-1: When does Adaptive Server automatically run update
statistics?
If the user Action taken by Job Scheduler
Specifies a datachange threshold of zero or NULL Runs update statistics at the scheduled time.

298 Adaptive Server Enterprise

 CHAPTER 10 Using Statistics to Improve Performance

If the user Action taken by Job Scheduler

Specifies a datachange threshold greater than zero Gets all the indexes for the table and gets the leading column
for a table only, and does not request the update for each index. If the datachange value for any leading
index statistics form column is greater than or equal to the threshold, run update
statistics.
Specifies threshold values for the table and index but Gets the datachange value for the leading column of the
does not request the update index statistics form index. If the datachange value is greater than or equal to the
threshold, runs update statistics.
Specifies a threshold value for a table only, and Gets all the indexes for the table and gets the leading column
requests the update index statistics form for each index. If the datachange value for any leading
column exceeds the threshold, runs update statistics.
Specifies threshold values for table and index and Gets the datachange value for the leading column of the
requests the update index statistics form index. If the datachange value is greater than or equal to the
threshold, runs update statistics.
Specifies threshold values for a table and one or Gets the datachange value for each column. If the
more columns (ignores any indexes or requests for datachange value for any column is greater than or equal to
the update index statistics form) the threshold, runs update statistics.

The datachange function compiles the number of changes in a table and

displays this as a percentage of the total number of rows in the table. You can
use this compiled information to create rules that determine when Adaptive
Server runs update statistics. The best time for this to happen can be based on
any number of objectives:
• The percentage of change in a table
• Number of CPU cycles available
• During a maintenance window
After update statistics runs, the datachange counter is reset to zero. The count
for datachange is tracked at the partition level (not the object level) for inserts,
and deletes and at the column level for updates.

Examples of updating statistics with datachange

You can write scripts that check for the specified amount of changed data at the
column, table, or partition level. The time at which you decide to run update
statistics can be based on a number of variables collected by the datachange
function; CPU usage, percent change in a table, percent change in a partition,
and so on.

Performance and Tuning Series: Query Processing and Abstract Plans 299
Column statistics and statistics maintenance

In this example, the authors table is partitioned, and the user wants to run
update statistics when the data changes to the city column in the author_ptn2
partition are greater than or equal to 50%:
select @datachange = datachange("authors","author_ptn2", "city")
if @datachange >= 50
begin
update statistics authors partition author_ptn2(city)
end
go
The user can also specify that the script is executed when the system is idle or
any other parameters.
In this example, the user triggers update statistics when the data changes to the
city column of the authors table are greater than or equal to 100% (the table in
this example is not partitioned):
select @datachange = datachange("authors",NULL, "city")
if @datachange > 100
begin
update statistics authors (city)
end
go

Column statistics and statistics maintenance

Histograms are kept on a per-column basis, rather than on a per-index basis.
This has certain implications for managing statistics:
• If a column appears in more than one index, update statistics, update index
statistics, or create index updates the histogram for the column and the
density statistics for all prefix subsets.
update all statistics updates histograms for all columns in a table.

• Dropping an index does not drop the statistics for the index, since the
optimizer can use column-level statistics to estimate costs, even when no
index exists.
To remove the statistics after dropping an index, you must explicitly delete
them using delete statistics.

300 Adaptive Server Enterprise

 CHAPTER 10 Using Statistics to Improve Performance

If the statistics are useful to the query processor, and to keep the statistics
without having an index, use update statistics, specifying the column
name, for indexes where the distribution of key values changes over time.
• Truncating a table does not delete the column-level statistics in
sysstatistics. In many cases, tables are truncated and the same data is
reloaded.
Since truncate table does not delete the column-level statistics, you need
not run update statistics after the table is reloaded, if the data is the same.
If you reload the table with data that has a different distribution of key
values, run update statistics.
• You can drop and re-create indexes without affecting the index statistics,
by specifying 0 for the number of steps in the with statistics clause to create
index. This create index command does not affect the statistics in
sysstatistics:
create index title_id_ix on titles(title_id)
with statistics using 0 values
This allows you to re-create an index without overwriting statistics that
have been edited with optdiag.
• If two users attempt to create an index on the same table, with the same
columns, at the same time, one of the commands may fail due to an attempt
to enter a duplicate key value in sysstatistics.
• Executing update statistics on a column in a partition of a multipartition
table updates the statistics for that partition, but also updates the global
histogram for that column. This is done by merging the histograms for that
column from each partition in a row-weighted fashion to arrive at a global
histogram for the column.
• Updating statistics on a multipartitioned table for a column, without
specifying a partition, updates the statistics for each partition of the table
for that column, and, as a last step, merges the partition histograms for the
column to create a global histogram for the column.
• The optimizer only uses the global histograms for a multipartitioned table
during compilation, and does not read the partition histograms. This
approach avoids the overhead of merging partition histograms at
compilation time, and instead performs any merging work at DDL time.

Performance and Tuning Series: Query Processing and Abstract Plans 301
Creating and updating column statistics

Creating and updating column statistics

Creating statistics on unindexed columns can improve the performance of
many queries. The optimizer can use statistics on any column in a where or
having clause to help estimate the number of rows from a table that match the
complete set of query clauses on that table.
Adding statistics for the minor columns of indexes and for unindexed columns
that are frequently used in search arguments can greatly improve the
optimizer’s estimates.
Maintaining a large number of indexes during data modification can be
expensive. Every index for a table must be updated for each insert and delete to
the table, and updates can affect one or more indexes.
Generating statistics for a column without creating an index gives the optimizer
more information to use for estimating the number of pages to be read by a
query, without the processing expense of index updates during data
modification.
The optimizer can apply statistics for any columns used in a search argument
of a where or having clause, and for any column named in a join clause.
Use these commands to create and maintain column statistics:
• update statistics, when used with the name of a column, generates statistics
for that column without creating an index on it. See “Adding statistics for
a column with update statistics” on page 305 for information about
syntax.
The optimizer can use these column statistics to more precisely estimate
the cost of queries that reference the column.
• update index statistics, when used with an index name, creates or updates
statistics for all columns in an index. See “Adding statistics for minor
columns with update index statistics” on page 305 for information about
syntax.
If used with a table name, update index statistics updates statistics for all
indexed columns.
• update all statistics creates or updates statistics for all columns in a table.
See “Adding statistics for all columns with update all statistics” on page
306 for information about syntax.
Good candidates for column statistics are:
• Columns frequently used as search arguments in where and having clauses

302 Adaptive Server Enterprise

 CHAPTER 10 Using Statistics to Improve Performance

• Columns included in a composite index, and which are not the leading
columns in the index, but which can help estimate the number of data rows
that need to be returned by a query

When additional statistics may be useful

To determine when additional statistics are useful, run queries using set option
commands and set statistics io. If there are significant discrepancies between
the “rows to be returned” and I/O estimates displayed by set commands and the
actual I/O displayed by statistics io, examine these queries for places where
additional statistics can improve the estimates. Look especially for the use of
default density values for search arguments and join columns.
The set option show_missing_stats command prints the names of columns that
could have used histograms, and groups of columns that could have used
multiattribute densities. This is particularly useful in pointing out where
additional statistics can be useful.
Example 1 set option show_missing_stats long
go
dbcc traceon(3604)
go
DBCC execution completed. If DBCC printed error
messages, contact a user with System Administrator (SA)
role.
select * from part, partsupp
where p_partkey = ps_partkey and p_itemtype =
ps_itemtype
go
NO STATS on column part.p_partkey
NO STATS on column part.p_itemtype
NO STATS on column partsupp.pa_itemtype
NO STATS on density set for E={p_partkey, p_itemtype}
NO STATS on density set for F={ps_partkey, ps_itemtype}
- - - - - - - - - - - - - - -
(200 rows affected)
You can get the same information using the show_final_plan_xml option. The
set plan uses the client option and trace flag 3604 to get the output on the client
side. This differs from the way the message option of set plan is used.
Example 2 dbcc traceon(3604)
DBCC execution completed. If DBCC printed error
messages, contact a user with System Administrator (SA)
role.

Performance and Tuning Series: Query Processing and Abstract Plans 303
Creating and updating column statistics

set plan for show_final_plan_xml to client on

go
select * from part, partsupp
where p_partkey = ps_partkey and p_itemtype =
ps_itemtype
go
<?xml version=”1.0” encoding=”UTF-8”?>
<query>
<planVersion> 1.0 </planVersion>
- - - - - - - - - - - - - - - - - -
<optimizerStatistics>
<statInfo>
<objName>part</objName>
<missingHistogram>
<column>p_partkey</column>
<column>p_itemtype</column>
</missingHistogram>
<missingDensity>
<column>p_partkey</column>
<column>p_itemtype</column>
</missingDensity>
</statInfo>
<statInfo>
<objName>partsupp</objName>
<missingHistogram>
<column>ps_partkey</column>
<column>ps_itemtype</column>
</missingHistogram>
<missingDensity>
<column>ps_partkey</column>
<column>ps_itemtype</column>
</missingDensity>
</statInfo>
</optimizerStatistics>
Use update statistics on part and partsupp to create statistics on p_partkey and
p_itemtype, thus creating a histogram on the leading column (p_partkey) and
the density (p_partkey, p_itemtype). Create a histogram on p_itemtype as well.
Use:
update statistics part(p_partkey, p_itemtype)
go
update statistics part(p_itemtype)
go

304 Adaptive Server Enterprise

 CHAPTER 10 Using Statistics to Improve Performance

Since partsupp has a histogram on ps_partkey, you can create a histogram on

ps_itemtype and a density on (ps_itemtype, ps_partkey). The columns used for
density may be unordered.
update statistics partsupp(ps_itemtype, ps_partkey)

If this procedure is successful, you will not see the “NO STATS” messages
shown in Example 1 when you run the query again.

Adding statistics for a column with update statistics

To add statistics for the price column in the titles table, enter:
update statistics titles (price)
To specify the number of histogram steps for a column, use:
update statistics titles (price)
using 50 values
This command adds a histogram for the titles.pub_id column and generates
density values for the prefix subsets pub_id; pub_id, pubdate; and pub_id,
pubdate, title_id:

update statistics titles(pub_id, pubdate, title_id)

However, this command does not create a histogram on pubdate and
title_id, since a separate update statistics command is needed for every
column for which a histogram is desired.

Note Running update statistics with a table name updates histograms and
densities for leading columns for indexes only; it does not update the statistics
for unindexed columns. To maintain these statistics, run update statistics and
specify the column name, or run update all statistics.

Adding statistics for minor columns with update index statistics

To create or update statistics on all columns in an index, use update index
statistics. The syntax is:
update index statistics
table_name [[ partition data_partition_name ] |
[ index_name [ partition index_partition_name ] ] ]

Performance and Tuning Series: Query Processing and Abstract Plans 305
Choosing step numbers for histograms

[ using step values ]

[ with consumers = consumers ] [, sampling = percent]

Adding statistics for all columns with update all statistics

To create or update statistics on all columns in a table, use update all statistics.
The syntax is:
update all statistics table_name
[partition data_partition_name]

Choosing step numbers for histograms

By default, each histogram has 20 steps, which provides good performance and
modeling for columns that have an even distribution of values. A higher
number of steps can increase the accuracy of I/O estimates for columns that:
• Have a large number of highly duplicated values
• Have unequal or skewed distribution of values
• Are queried using leading wildcards in like queries
The histogram tuning factor default of 20 automatically chooses a step value
between the current requested step value (default 20) and the increased steps
due to the factor (20 * 20 = 400) so that Adaptive Server automatically chooses
the optimal steps value to compensate for the above cases. Overriding the step
values should take into account the larger number of steps already introduced
by the histogram tuning factor.

Note If your database was updated from a pre-11.9 version of Adaptive Server,
the number of steps defaults to the number of steps that were used on the
distribution page.

Increasing the number of steps beyond what is needed for good query
optimization can degrade Adaptive Server performance, largely due to the
amount of space that is required to store and use the statistics. Increasing the
number of steps:
• Increases the disk storage space required for sysstatistics

306 Adaptive Server Enterprise

 CHAPTER 10 Using Statistics to Improve Performance

• Increases the cache space needed to read statistics during query

optimization
• Requires more I/O, if the number of steps is very large
During query optimization, histograms use space borrowed from the procedure
cache. This space is released as soon as the query is optimized.

Choosing a step number

If your table has 5000 rows, and one value in the column that has only one
matching row, you may need to request 5000 steps to get a histogram that
includes a frequency cell for every distinct value. The actual number of steps
is not 5000; it is either the number of distinct values plus one (for dense
frequency cells), or twice the number of values plus one (for sparse frequency
cells).
The sp_configure option histogram tuning factor automatically chooses a larger
number of steps, within parameters, when there are a large number of highly
duplicated values.
The default value of the histogram tuning factor is 20 in Adaptive Server version
15.0 and later. If the requested step count is 50, then update statistics can create
up to 20 * 50 = 1000 steps. This larger number of steps is used only if histogram
distribution is skewed with a number of domain values that are highly
duplicated. However, for a unique column, update statistics uses only 50 steps
to represent the histogram. To most efficiently use histograms, specify a
relatively low number of steps and allow the histogram tuning factor to
determine whether more steps would be useful for optimization. For example,
instead of specifying 1000 steps with a default step count of 1000 to be used by
all histograms, it is better to specify 50 default steps and a histogram tuning
factor of 20. This allows Adaptive Server to determine the best step count,
within the range of 50 to 1000 steps, with which to represent the distribution.

Scan types, sort requirements, and locking

Table 10-2 shows the types of scans performed during update statistics, the
types of locks acquired, and when sorts are needed.

Performance and Tuning Series: Query Processing and Abstract Plans 307
Scan types, sort requirements, and locking

Table 10-2: Scans, sorts, and locking during update statistics

update statistics
specifying Scans and sorts performed Locking
Table name
Allpages-locked table Table scan, plus a leaf-level scan of each Level 1; shared intent table lock,
nonclustered index shared lock on current page
Data-only-locked table Table scan, plus a leaf-level scan of each Level 0; dirty reads
nonclustered index and the clustered index, if one
exists
Table name and clustered index name
Allpages-locked table Table scan Level 1; shared intent table lock,
shared lock on current page
Data-only-locked table Leaf level index scan Level 0; dirty reads
Table name and nonclustered index name
Allpages-locked table Leaf level index scan Level 1; shared intent table lock,
shared lock on current page
Data-only-locked table Leaf level index scan Level 0; dirty reads
Table name and column name
Allpages-locked table Table scan; creates a worktable and sorts the Level 1; shared intent table lock,
worktable shared lock on current page
Data-only-locked table Table scan; creates a worktable and sorts the Level 0; dirty reads
worktable

Sorts for unindexed or nonleading columns

For unindexed columns and columns that are not the leading columns in
indexes, Adaptive Server performs a serial table scan, copying the column
values into a worktable. It then sorts the worktable to build the histogram. The
sort is performed in serial, unless the with consumers clause is specified.
See Chapter 5, “Parallel Query Processing,”.

Locking, scans, and sorts during update index statistics

The update index statistics command generates a series of update statistics
operations that use the same locking, scanning, and sorting as the equivalent
index-level and column-level command. For example, if the salesdetail table
has a nonclustered index named sales_det_ix on salesdetail(stor_id, ord_num,
title_id), this command:

308 Adaptive Server Enterprise

 CHAPTER 10 Using Statistics to Improve Performance

update index statistics salesdetail

performs these update statistics operations:
update statistics salesdetail sales_det_ix
update statistics salesdetail (ord_num)
update statistics salesdetail (title_id)

Locking, scans and sorts during update all statistics

The update all statistics commands generate a series of update statistics
operations for each index on the table, followed by a series of update statistics
operations for all unindexed columns.

Using the with consumers clause

The with consumers clause for update statistics is designed for use on
partitioned tables on Redundant Array of Independent Disks (RAID) devices,
which appear to Adaptive Server as a single I/O device, but can produce the
high throughput required for parallel sorting. See Chapter 5, “Parallel Query
Processing,” for information.

Reducing the impact of update statistics on concurrent processes

Since update statistics uses dirty reads (transaction isolation level 0) for
data-only-locked tables, you can execute it while other tasks are active on the
server; it does not block access to tables and indexes. Updating statistics for
leading columns in indexes requires only a leaf-level scan of the index, and
does not require a sort, so updating statistics for these columns does not affect
concurrent performance very much.
However, updating statistics for unindexed and nonleading columns, which
require a table scan, worktable, and sort, can affect concurrent processing.
• Sorts are CPU-intensive. Use a serial sort, or a small number of worker
processes to minimize CPU utilization. Alternatively, you can use
execution classes to set the priority for update statistics.
See the Performance and Tuning Series: Basics.

Performance and Tuning Series: Query Processing and Abstract Plans 309
Using the delete statistics command

• The cache space required for merging sort runs is taken from the data
cache, and some procedure cache space is also required. Setting the
number of sort buffers to a low value reduces the space used in the buffer
cache.
If number of sort buffers is set to a large value, it takes more space from the
data cache, and may also cause stored procedures to be flushed from the
procedure cache, since procedure cache space is used while merging
sorted values. There are approximately 100 bytes of procedure cache
needed for every row that can fit into the sort buffers specified. For
example, if 500 2K sort buffers are specified, and about 200 rows fit into
each 2K buffer, then 200 * 100 * 500 bytes of procedure cache are needed
to support the sort. This example requires about 5000 2K procedure cache
buffers, if the entire 500 data cache buffers are filled by a sort run.
Creating the worktables for sorts also uses space in tempdb.

Using the delete statistics command

In versions of Adaptive Server earlier than 11.9, dropping an index removed
the distribution page for the index. As of version 11.9.2, maintaining
column-level statistics is under explicit user control, and the optimizer can use
column-level statistics even when an index does not exist. The delete statistics
command allows you to drop statistics for specific columns.
If you create an index and then decide to drop it because it is not useful for data
access, or because of the cost of index maintenance during data modifications,
you must determine whether the:
• Statistics on the index are useful to the optimizer.
• Distribution of key values in the columns for this index are subject to
change over time as rows are inserted and deleted.
If the distribution of key values changes, run update statistics periodically
to maintain useful statistics.
This example deletes the statistics for the price column in the titles table:

310 Adaptive Server Enterprise

 CHAPTER 10 Using Statistics to Improve Performance

delete statistics titles(price)

Note delete statistics removes rows only from sysstatistics; it does not remove
rows from systabstats. You cannot delete the rows in systabstats that described
partition row counts, cluster ratios, page counts, and so on. However, if you use
optdiag simulate statistics to add any simulated systabstats rows to sysstatistics,
then those rows are deleted.

When row counts may be inaccurate

Row count values for the number of rows, number of forwarded rows, and
number of deleted rows may be inaccurate, especially if query processing
includes many rollback commands. If workloads are extremely heavy, and the
housekeeper wash task does not run often, these statistics are more likely to be
inaccurate.
Running update statistics corrects counts in systabstats. When the housekeeper
wash task runs, or when you execute sp_flushstats, row count values are saved
in systabstats.

Note You must set the configuration parameter housekeeper free write percent
to 1 or greater to enable housekeeper statistics flushing.

Running dbcc checktable or dbcc checkdb updates these row count values in
memory.

Performance and Tuning Series: Query Processing and Abstract Plans 311
When row counts may be inaccurate

312 Adaptive Server Enterprise

CH A PTE R 11 Introduction to Abstract Plans

Topic Page
Overview 313
Managing abstract plans 314
Relationship between query text and query plans 315
Full versus partial plans 316
Abstract plan groups 318
How abstract plans are associated with queries 318

Overview
Adaptive Server can generate an abstract plan for a query, and save the
text and its associated abstract plan in the sysqueryplans system table.
Using a rapid hashing method, incoming SQL queries can be compared to
saved query text, and if a match is found, the corresponding saved abstract
plan is used to execute the query.
An abstract plan describes the execution plan for a query using a language
created for that purpose. This language contains operators to specify the
choices and actions that can be generated by the optimizer. For example,
to specify an index scan on the titles table, using the index title_id_ix, the
abstract plan says:
(i_scan title_id_ix titles)
To use this abstract plan with a query, you can modify the query text and
add a PLAN clause:
select * from titles where title_id = “On Liberty”
plan
“(i_scan title_id_ix titles)”
This alternative requires a change to the SQL text; however, the method
described in the first paragraph, that is, the sysqueryplans-based way to
give the abstract plan of a query, does not involve changing the query text.

Performance and Tuning Series: Query Processing and Abstract Plans 313
Managing abstract plans

Abstract plans provide a means for system administrators and

performance tuners to protect the overall performance of a server from
changes to query plans. Changes in query plans can arise due to:
• Adaptive Server software upgrades that affect optimizer choices and
query plans
• New Adaptive Server features that change query plans
• Changing tuning options such as the parallel degree, table
partitioning, or indexing
The main purpose of abstract plans is to provide a means to capture query
plans before and after major system changes. You can then compare sets
of before-and-after query plans to determine the effects of changes on your
queries. Other uses include:
• Searching for specific types of plans, such as table scans or
reformatting
• Searching for plans that use particular indexes
• Specifying full or partial plans for poorly performing queries
• Saving plans for queries with long optimization times
Abstract plans provide an alternative to options that must be specified in
the batch or query to influence optimizer decisions. Using abstract plans,
you can influence the optimization of a SQL statement without modifing
the statement syntax. While matching query text to stored text requires
some processing overhead, using a saved plan reduces query optimization
overhead.

Managing abstract plans

A full set of system procedures allows system administrators and database
owners to administer plans and plan groups. Individual users can view,
drop, and copy the plans for the queries that they have run.
See Chapter 14, “Managing Abstract Plans with System Procedures,” for
more information.

314 Adaptive Server Enterprise

 CHAPTER 11 Introduction to Abstract Plans

Relationship between query text and query plans

For most SQL queries, there are many possible query execution plans.
SQL describes the desired result set, but does not describe how that result
set should be obtained from the database. Consider a query that joins three
tables, such as:
select t1.c11, t2.c21
from t1, t2, t3
where t1.c11 = t2.c21
and t1.c11 = t3.c31
There are many different possible join orders, and depending on the
indexes that exist on the tables, many possible access methods, including
table scans, index scans, and the reformatting strategy. Each join may use
either a nested-loop join or a merge join. These choices are determined by
the optimizer’s query costing algorithms, and are not included in or
specified in the query itself.
When you capture the abstract plan, the query is optimized in the usual
way, except that the optimizer also generates an abstract plan, and saves
the query text and abstract plan in sysqueryplans.

Limits of options for influencing query plans

Adaptive Server provides other options for influencing optimizer choices:
• Session-level options such as set forceplan to force join order or set
parallel_degree to specify the maximum number of worker processes
to use for the query
• Options that can be included in the query text to influence the index
choice, cache strategy, and parallel degree
There are some limitations to using set commands or adding hints to the
query text:
• Not all query plan steps can be influenced, for example, subquery
attachment.
• Some query-generating tools do not support the in-query options or
require all queries to be vendor-independent.

Performance and Tuning Series: Query Processing and Abstract Plans 315
Full versus partial plans

Full versus partial plans

Abstract plans can be full plans, describing all query processing steps and
options, or they can be partial plans. A partial plan might specify that an
index is to be used for the scan of a particular table, without specifying
other access methods. For example:
select t1.c11, t2.c21
from t1, t2, t3
where t1.c11 = t2.c21
and t1.c11 = t3.c31
plan
“(i_scan t3_c31_ix t3)”
The full abstract plan includes:
• The join type, nl_join for nested-loop joins, m_join for merge joins, or
h_join for hash joins.

• The join order.

• The type of scan, t_scan for table scan or i_scan for index scan.
• The name of the index chosen for the tables that are accessed via an
index scan.
• The scan properties: the parallel degree, I/O size, and cache strategy
for each table in the query.
The abstract plan for the query above specifies the join order, the access
method for each table in the query, and the scan properties for each table:
select t1.c11, t2.c21
from t1, t2, t3
where t1.c11 = t2.c21
and t1.c11 = t3.c31
plan
“(i_scan t3_c31_ix t3)”

(nl_join ( nl_join
( t_scan t2 )
( i_scan t1_c11_ix t1 )
)
( i_scan t3_c31_ix t3 )
)
( prop t3
( parallel 1 )
( prefetch 16 )
( lru )

316 Adaptive Server Enterprise

 CHAPTER 11 Introduction to Abstract Plans

)
( prop t1
( parallel 1 )
( prefetch 16 )
( lru )
)
( prop t2
( parallel 1 )
( prefetch 16 )
( lru )
)
If the abstract plan dump mode is on, the query text and the abstract plan
pair are saved in sysqueryplans:
select t1.c11, t2.c21
from t1, t2, t3
where t1.c11 = t2.c21
and t1.c11 = t3.c31
plan
“(i_scan t3_c31_ix t3)”

Creating a partial plan

When abstract plans are captured, full abstract plans are generated and
stored. You can write partial plans that affect only a subset of the optimizer
choices. If the query above had not used the index on t3, but all other parts
of the query plan were optimal, you could create a partial plan for the
query using the create plan command. This partial plan specifies only the
index choice for t3:
create plan
"select t1.c11, t2.c21
from t1, t2, t3
where t1.c11 = t2.c21
and t1.c11 = t3.c31"
"( i_scan t3_c31_ix t3 )"
You can also create abstract plans with the plan clause for select, delete,
update, and other commands that can be optimized. If the abstract plan
dump mode is on, the query text and AP pair are saved in sysqueryplans.
See Chapter 12, “Creating and Using Abstract Plans,” for more
information.

Performance and Tuning Series: Query Processing and Abstract Plans 317
Abstract plan groups

Abstract plan groups

When you install Adaptive Server, there are two abstract plan groups:
• ap_stdout, used by default to capture plans

• ap_stdin, used by default for plan association

A system administrator can enable server-wide plan capture to ap_stdout,

so that all query plans for all queries are captured. Server-wide plan
association uses queries and plans from ap_stdin. If some queries require
specially-tuned plans, they can be made available server-wide.
A system administrator or database owner can create additional plan
groups, copy plans from one group to another, and compare plans in two
different groups.
The capture of abstract plans and the association of abstract plans with
queries always happens within the context of the currently active plan
group. Users can use session-level set commands to enable plan capture
and association.
Some of the ways abstract plan groups can be used are:
• A query tuner can create abstract plans in a group created for testing
purposes without affecting plans for other users on the system
• Using plan groups, “before” and “after” sets of plans can be used to
determine the effects of system or upgrade changes on query
optimization.
See Chapter 12, “Creating and Using Abstract Plans,” for more
information on enabling the capture and association of plans.

How abstract plans are associated with queries

When an abstract plan is saved, all white space (tabs, multiple spaces, and
returns, except for returns that terminate a --style comment) in the query
is trimmed to a single space, and a hash-key value is computed for the
white-space trimmed SQL statement. The trimmed SQL statement and the
hash key are stored in sysqueryplans along with the abstract plan, a unique
plan ID, the user’s ID, and the ID of the current abstract plan group.

318 Adaptive Server Enterprise

 CHAPTER 11 Introduction to Abstract Plans

When abstract plan association is enabled, the hash key for incoming SQL
statements is computed, and this value is used to search for the matching
query and abstract plan in the current association group, with the
corresponding user ID. The full association key of an abstract plans
consists of:
• The user ID of the current user
• The group ID of the current association group
• The full query text
Once a matching hash key is found, the full text of the saved query is
compared to the query to be executed, and used if it matches.
The association key combination of user ID, group ID, and query text
means that for a given user, there cannot be two queries in the same
abstract plan group that have the same query text, but different query
plans.

Abstract plans in cached statements

In Adaptive Server version 15.7 and later, you can save abstract plan
information in the statement cache.
In this example, which includes an abstract plan, the hash table saves
select * from t1 plan '(use optgoal allrows_mix)', as
shown in the SQL TEXT line:
1> select * from t1 plan '(use optgoal allrows_mix)'
2> go
1> dbcc prsqlcache
2> go

Start of SSQL Hash Table at 0x0x1474c9050

Memory configured: 1000 2k pages Memory used: 17 2k pages

Bucket# 243 address 0x0x1474c9f80

SSQL_DESC 0x0x1474cd070
ssql_name *ss0626156152_0290084701ss*
ssql_hashkey 0x0x114a575d ssql_id 626156152
ssql_suid 1 ssql_uid 1 ssql_dbid 1 ssql_spid 0
ssql_status 0x0xa0 ssql_parallel_deg 1

Performance and Tuning Series: Query Processing and Abstract Plans 319
How abstract plans are associated with queries

ssql_isolate 1 ssql_tranmode 32
ssql_keep 0 ssql_usecnt 1 ssql_pgcount 6
ssql_optgoal allrows_mix ssql_optlevel ase_default
SQL TEXT: select * from t1 plan '(use optgoal allrows_mix)'

End of SSQL Hash Table

320 Adaptive Server Enterprise

CH A PTE R 1 2 Creating and Using Abstract
Plans

Use the set command to capture abstract plans and to associate incoming
SQL queries with saved plans. Any user can issue session-level
commands to capture and load plans during a session, and a system
administrator can enable server-wide abstract plan capture and
association. This chapter also describes how to specify abstract plans
using SQL.
Topic Page
Using set commands to capture and associate plans 321
set plan exists check option 327
Using other set options with abstract plans 328
Server-wide abstract plan capture and association modes 330
Creating plans using SQL 331

Using set commands to capture and associate plans

At the session level, any user can enable and disable capture and use of
abstract plans using the set plan dump and set plan load commands. set
plan replace determines whether existing plans are overwritten by changed
plans.
Enabling and disabling abstract plan modes takes effect at the end of the
batch in which the command is included (similar to showplan). Therefore,
change the mode in a separate batch before you run your queries:
set plan dump on
go
/*queries to run*/
go

Performance and Tuning Series: Query Processing and Abstract Plans 321
Using set commands to capture and associate plans

Any set plan commands used in a stored procedure do not affect the
procedure (except those statements affected by deferred compilation) in
which they are included, but remain in effect after the procedure
completes.

Enabling plan capture mode with set plan dump

The set plan dump command activates and deactivates the capture of
abstract plans. You can save the plans to the default group, ap_stdout, by
using set plan dump with no group name:
set plan dump on
To start capturing plans in a specific abstract plan group, specify the group
name. This example sets the group dev_plans as the capture group:
set plan dump dev_plans on
The group that you specify must exist before you issue the set command.
The system procedure sp_add_qpgroup creates abstract plan groups; only
the system administrator or database owner can create an abstract plan
group. Once an abstract plan group exists, any user can dump plans to the
group.
See “Creating a group” on page 373 for information on creating a plan
group.
To deactivate the capturing of plans, use:
set plan dump off
You do not need to specify a group name to end capture mode. Only one
abstract plan group can be active for saving or matching abstract plans at
any one time. If you are currently saving plans to a group, turn off the plan
dump mode, and reenable it for the new group, as shown here:

set plan dump on /*save to the default group*/

go
/*some queries to be captured */
go
set plan dump off
go
set plan dump dev_plans on
go
/*additional queries*/
go

322 Adaptive Server Enterprise

 CHAPTER 12 Creating and Using Abstract Plans

The use of the use database command while set plan dump is in effect
disables plan dump mode.

Associating queries with stored plans

The set plan load command activates and deactivates the association of
queries with stored abstract plans.
To start the association mode using the default group, ap_stdin, use:
set plan load on
To enable association mode using another abstract plan group, specify the
group name:
set plan load test_plans on
Only one abstract plan group can be active for plan association at one time.
If plan association is active for a group, deactivate the current group and
activate plan association for the new group, as shown here:
set plan load test_plans on
go
/*some queries*/
go
set plan load off
go
set plan load dev_plans on
go
The use of the use database command while set plan load is in effect
disables plan load mode.

Using replace mode during plan capture

While plan capture mode is active, you can choose whether to have plans
for the same query replace existing plans by enabling or disabling set plan
replace. To activate plan replacement mode, use:

set plan replace on

Do not specify a group name with set plan replace; it affects the current
active capture group.
To disable plan replacement:

Performance and Tuning Series: Query Processing and Abstract Plans 323
Using set commands to capture and associate plans

set plan replace off

The use of the use database command while set plan replace is in effect
disables plan replace mode.

When to use replace mode

When you are capturing plans, and a query has the same query text as an
already-saved plan, the existing plan is not replaced unless replace mode
is enabled. If you have captured abstract plans for specific queries, and you
are making physical changes to the database that affect optimizer choices,
replace existing plans for these changes to be saved.
Some actions that might require plan replacement are:
• Adding or dropping indexes, or changing keys or key ordering in
indexes
• Changing the partitioning on a table
• Adding or removing buffer pools
• Changing configuration parameters that affect query plans
In most cases, do not enable plan load. When plan association is active, any
plan specifications are used as inputs to the optimizer. For example, if a
full query plan includes the prefetch property and an I/O size of 2K, and
you have created a 16K pool and want to replace the prefetch specification
in the plan, do not enable plan load mode.
You may want to check query plans and replace some abstract plans as
data distribution changes in tables, or after rebuilds on indexes, updating
statistics, or changing the locking scheme.

Using dump, load, and replace modes simultaneously

You can have both plan dump and plan load mode active simultaneously,
with or without replace mode active.

Using dump and load to the same group

If you have enabled dump and load to the same group, without replace
mode enabled:

324 Adaptive Server Enterprise

 CHAPTER 12 Creating and Using Abstract Plans

• If a valid plan exists for the query, it is loaded and used to optimize
the query.
• If a plan exists that is not valid (for example, because an index has
been dropped), a new plan is generated and used to optimize the
query, but is not saved.
• If only a partial plan exists, a full plan is generated, but the existing
partial plan is not replaced
• If a plan does not exist for the query, a plan is generated and saved.
With replace mode also enabled:
• If a valid plan exists for the query, it is loaded and used to optimize
the query.
• If the plan is not valid, a new plan is generated and used to optimize
the query, and the old plan is replaced.
• If the plan is a partial plan, a complete plan is generated and used, and
the existing partial plan is replaced. The specifications in the partial
plan are used as input to the optimizer.
• If a plan does not exist for the query, a plan is generated and saved.

Using dump and load to different groups

If you have dump enabled to one group, and load enabled from another
group, without replace mode enabled:
• If a valid plan exists for the query in the load group, it is loaded and
used. The plan is saved in the dump group, unless a plan for the query
already exists in the dump group.
• If the plan in the load group is not valid, a new plan is generated. The
new plan is saved in the dump group, unless a plan for the query
already exists in the dump group.
• If the plan in the load group is a partial plan, a full plan is generated
and saved in the dump group, unless a plan already exists. The
specifications in the partial plan are used as input to the optimizer.
• If there is no plan for the query in the load group, the plan is generated
and saved in the dump group, unless a plan for the query exists in the
dump group.
With replace mode active:

Performance and Tuning Series: Query Processing and Abstract Plans 325
Using set commands to capture and associate plans

• If a valid plan exists for the query in the load group, it is loaded and
used.
• If the plan in the load group is not valid, a new plan is generated and
used to optimize the query. The new plan is saved in the dump group.
• If the plan in the load group is a partial plan, a full plan is generated
and saved in the dump group. The specifications in the partial plan are
used as input to the optimizer.
• If a plan does not exist for the query in the load group, a new plan is
generated. The new plan is saved in the dump group.

Compile-time changes for some set parameters

In Adaptive Server releases earlier than 15.0.2, the set parameters took
effect after the stored procedure was executed or recompiled. Adaptive
Server release 15.0.2 and later allows you to use optimizer set parameters
at compile time to affect the optimizer in stored procedures or batches.

Note This changed behavior may effect the composition of the result set.
Sybase recommends that you review the result set created by the 15.0.2
versions of the set parameters before using them in your production
systems.
You must reset the set parameter before returning from the stored
procedure or the execution of subsequent stored procedures may be
affected. If you intend to propogate this change to subsequent stored
procedures, use export_options parameter.

Adaptive Server changes the compile-time behavior for these parameters:

• distinc_sorted

• distinct_sorting

• distinct_hashing

• group_sorted

• group_hashing

• bushy_space_search

• parallel_query

326 Adaptive Server Enterprise

 CHAPTER 12 Creating and Using Abstract Plans

• order_sorting

• nl_join

• merge_join

• hash_join

• append_union_all

• merge_union_all

• merge_union_distinct

• hash_union_distinct

• store_index

• index_intersection

• index_union

• multi_table_store_ind

• opportunistic_distict_view

• advanced_aggregation

• replicated_partition

• group_inserting

• basic_optimization

• auto_query_tuning

• query_tuning_mem_limit

• query_tuning_time_limit

• set plan optgoal

set plan exists check option

Use the exists check mode during query plan association to speed
performance when users require abstract plans for fewer than 20 queries
from an abstract plan group. If a small number of queries require plans to
improve their optimization, enabling exists check mode speeds execution
of all queries that do not have abstract plans, because they do not check for
plans in sysqueryplans.

Performance and Tuning Series: Query Processing and Abstract Plans 327
Using other set options with abstract plans

When set plan load and set exists check are both enabled, the hash keys for
up to 20 queries in the load group are cached for the user. If the load group
contains more than 20 queries, exists check mode is disabled. Each
incoming query is hashed; if its hash key is not stored in the abstract plan
cache, then there is no plan for the query and no search is made. This
speeds the compilation of all queries that do not have saved plans.
The syntax is:
set plan exists check {on | off}
You must enable load mode before you enable plan hash-key caching.
A system administrator can configure server-wide plan hash-key caching
with the configuration parameter abstract plan cache. To enable server-
wide plan caching, use:
sp_configure "abstract plan cache", 1

Using other set options with abstract plans

You can combine other set tuning options with set plan dump,
show_abstract_plan, and set plan load.

Using show_abstract_plan to view plans

set option show_abstract_plan prints the optimal abstract plan currently
running on the TDS connection. It prints the plan after optimization and
before execution, and is the only set option show_ command that does not
depend on trace flag 3604 or 3605.
Printing the final plan’s abstract plan is similar to viewing showplan
output: it provides information to the user, but the abstract plan is not
saved in sysqueryplans, and is not used if you use the apstract plan load
mode.
This example shows the optimal abstract plan currently running on the
TDS connection:
1> set option show_abstract_plan on
2> go
1> select r1, sum(s1)
2> from r, s

328 Adaptive Server Enterprise

 CHAPTER 12 Creating and Using Abstract Plans

3> where r2=s2

4> group by r1
The Abstract Plan (AP) of the final query execution plan:
( group_sorted ( nl_join ( i_scan ir12 r ) ( i_scan is21 s ) ) ) ( prop r
(parallel 1 ) ( prefetch 2 ) ( lru ) ) ( prop s ( parallel 1 ) ( prefetch 2 )
(lru ) )
To experiment with the optimizer behavior, this AP can be modified and then
passed to the optimizer using the PLAN clause:
SELECT/INSERT/DELETE/UPDATE ...
PLAN '( ... )'.
r1
----------- -----------
1 2
2 4

(2 rows affected)

Using showplan
When showplan is turned on, and abstract plan association mode has been
enabled with set plan load, showplan prints the plan ID of the matching
abstract plan at the beginning of the showplan output for the statement:
QUERY PLAN FOR STATEMENT 1 (at line 1).
Optimized using an Abstract Plan (ID : 832005995).
If you run queries using the plan clause added to a SQL statement,
showplan displays:

Optimized using the Abstract Plan in the PLAN clause.

Using noexec
You can use noexec mode to capture abstract plans without actually
executing the queries. If noexec mode is in effect, queries are optimized
and abstract plans are saved, but no query results are returned.
To use noexec mode while capturing abstract plans, execute any needed
procedures (such as sp_add_qpgroup) and other set options (such as set
plan dump) before enabling noexec mode. The following example shows a
typical set of steps:
sp_add_qpgroup pubs_dev
go

Performance and Tuning Series: Query Processing and Abstract Plans 329
Server-wide abstract plan capture and association modes

set plan dump pubs_dev on

go
set noexec on
go
select type, sum(price) from titles group by type
go

Using fmtonly
A similar behavior can be obtained for capturing plans in stored
procedures without actually executing the stored procedures, using fmtonly
set.

sp_add_qpgroup pubs_dev
go
set plan dump pubs_dev on
go
set fmtonly on
go
exec stored_proc(...)
go

Using forceplan
If set forceplan on is in effect, and query association is also enabled for the
session, forceplan is ignored if a full abstract plan is used to optimize the
query. If a partial plan does not completely specify the join order:
• First, the tables in the abstract plan are ordered, as specified.
• The remaining tables are ordered as specified in the from clause.
• The two lists of tables are merged.

Server-wide abstract plan capture and association

modes
A system administrator can enable server-wide plan capture, association,
and replacement modes with these configuration parameters:

330 Adaptive Server Enterprise

 CHAPTER 12 Creating and Using Abstract Plans

• abstract plan dump – enables dumping to the default abstract plans

capture group, ap_stdout.
• abstract plan load – enables loading from the default abstract plans
loading group, ap_stdin.
• abstract plan replace – when plan dump mode is also enabled, enables
plan replacement.
• abstract plan cache – enables caching of abstract plan hash IDs;
abstract plan load must also be enabled. See “set plan exists check
option” on page 327.
By default, these configuration parameters are set to 0, which means that
capture and association modes are off. To enable a mode, set the
configuration value to 1:
sp_configure "abstract plan dump", 1
Enabling any of the server-wide abstract plan modes is dynamic; you need
not restart the server.
Server-wide capture and association allows the system administrator to
capture all plans for all users on a server. You cannot override server-wide
modes at the session level.

Creating plans using SQL

You can directly specify the abstract plan for a query by:
• Using the create plan command
• Adding the plan clause to select, insert...select, update, delete and
return commands, and to if and while clauses

For information on writing plans, see Chapter 13, “Abstract Query Plan
Guide.”

Using create plan

The create plan command specifies the text of a query, and the abstract
plan to save for the query.
This example creates an abstract plan:

Performance and Tuning Series: Query Processing and Abstract Plans 331
Creating plans using SQL

create plan
“select avg(price) from titles”
“(scalar_agg
(i_scan type_price_ix titles)
)”
The plan is saved in the current active plan group. You can also specify the
group name:
create plan
“select avg(price) from titles”
“(scalar_agg
(i_scan type_price_ix titles)
)”
into dev_plans
If a plan already exists for the specified query in the current plan group, or
the plan group that you specify, you must first enable replace mode in
order to overwrite the existing plan.
To see the plan ID that is used for a plan you create, create plan can return
the ID as a variable. You must declare the variable first. This example
returns the plan ID:
create plan
“select avg(price) from titles”
“(scalar_agg
(i_scan type_price_ix titles)
)”
into dev_plans
and set @id

select @id
When you use create plan, the query in the plan is not executed. This
means that:
• The text of the query is not parsed, so the query is not checked for
valid SQL syntax.
• The plans are not checked for valid abstract plan syntax.
• The plans are not checked to determine whether they are compatible
with the SQL text.
To guard against errors and problems, immediately execute the specified
query with showplan enabled.

332 Adaptive Server Enterprise

 CHAPTER 12 Creating and Using Abstract Plans

Using the plan clause

You can use the plan clause with the following SQL statements to specify
the plan to use to execute the query:
• select

• insert...select

• delete

• update

• if

• while

• return

This example specifies the plan to use to execute the query:

select avg(price) from titles
plan
“(scalar_agg
(i_scan type_price_ix titles
)”
When you specify an abstract plan for a query, the query is executed using
the specified plan. If you have showplan enabled, this message is printed:
Optimized using the Abstract Plan in the PLAN clause.
When you use the plan clause with a query, any errors in the SQL text, the
plan syntax, and any mismatches between the plan and the SQL text are
reported as errors. For example, this plan uses the wrong abstract plan
operator for the query:
/* wrong operator! */
select * from t1,t2
where c11 = c21
plan
“(union
(t_scan t1)
(t_scan t2)
)"
This plan returns the following message:
Abstract Plan (AP) Warning: An error occurred while applying the AP:
(union (t_scan t1) (t_scan2))
to the SQL query:
select * from t1, t2

Performance and Tuning Series: Query Processing and Abstract Plans 333
Creating plans using SQL

where c11 = c21

Failed to apply the top operator ‘union’ of the following AP fragment:
(union (t_scan t1) (t_scan t2))
The query contains no union that matches the ‘union’ AP operator at this point.
The following template can be used as a basis for a valid AP:
(also_enforce (join (also_enforce (scan t1)) (also_enforce (scan t2)))
)
The optimizer will complete the compilation of this query; the query will be
executed normally.
Plans specified with the plan clause are saved in sysqueryplans only if plan
capture is enabled. If a plan for the query already exists in the current
capture group, enable replace mode to replace an existing plan.

334 Adaptive Server Enterprise

CH A PTE R 1 3 Abstract Query Plan Guide

This chapter covers some guidelines you can use in writing Abstract
Plans.
Topic Page
Overview 335
Tips on writing abstract plans 363
Using abstract plans at the query level 363
Comparing plans before and after 366
Abstract plans for stored procedures 368
Ad hoc queries and abstract plans 370

Overview
Abstract plans allow you to specify the desired execution plan of a query.
Abstract plans provide an alternative to the session-level and query-level
options that force a join order, or specify the index, I/O size, or other query
execution options. The session-level and query-level options are
described in Chapter 12, “Creating and Using Abstract Plans.”
There are several optimization decisions that you cannot specify with set
commands or clauses in the query text, for example:
• Algorithms that implement a given relational operator; for example,
NLJ versus MJ versus HJ or GroupSorted versus GroupHashing versus
GroupInserting

• Subquery attachment
• The join order for flattened subqueries
• Reformatting
In many cases when issuing T-SQL commands, you cannot include set
commands or change the query text. Abstract plans provide an alternative,
more complete method of influencing optimizer decisions.

Performance and Tuning Series: Query Processing and Abstract Plans 335
Overview

Abstract plans are relational algebra expressions that are not included in
the query text. They are stored in a system catalog and associated with
incoming queries based on the text of these queries.

Abstract plan language

The abstract plan language is a relational algebra that uses these operators:
• distinct – a logical operator describing duplicates elimination.

• distinct_sorted – a physical operator describing available

ordering-based duplicates elimination.
• distinct_sorting – a physical operator describing sorting-based
duplicates elimination.
• distinct_hashing – a physical operator describing hashing-based
duplicates elimination.
• group – a logical operator, describing vector aggregation.

• group_sorted – a physical operator describing the available

ordering-based vector aggregation.
• group_hashing – a physical operator describing hashing-based
vector aggregation.
• group_inserting – a physical operator describing clustered index
insertion-based vector aggregation.
• join – the generic join and a high-level logical join operator that
describes inner, outer and existence joins, using nested-loop joins,
merge joins, or hash joins.
• nl_join – specifying a nested-loop join, including all inner, outer,
and existence joins.
• m_join – specifying a merge join, including inner and outer joins.

• h_join – specifying a hash join, including all inner, outer, and

existence joins.
• union – a logical union operator. It describes both the union and the
union all SQL constructs.

• append_union_all – a physical operator implementing union all. It

appends the child result sets, one after the other.

336 Adaptive Server Enterprise

 CHAPTER 13 Abstract Query Plan Guide

• merge_union_all – a physical operator implementing union all. It

merges the child result sets on the subset of the projection that is
ordered in each child, and preserves that ordering.
• merge_union_distinct – a physical operator implementing union
[distinct]. A merge-based duplicates removal algorithm.

• hash_union_distinct – a physical operator implementing union

[distinct]. A merge-based duplicates removal algorithm.

• scalar_agg – a logical operator, describing scalar aggregation.

• scan – a logical operator that transforms a stored table in a flow of

rows, an abstract plan derived table. It allows partial plans that do not
restrict the access method.
• i_scan – a physical operator implementing scan. It directs the
optimizer to use an index scan on the specified table.
• t_scan – a physical operator implementing scan. It directs the
optimizer to use a full table scan on the specified table.
• m_scan – a physical operator implementing scan. It directs the
optimizer to use a multiindex table scan on the specified table,
either index union, index intersection, or both.
• store – a physical operator describing the materialization of an
abstract plan derived table in a stored worktable.
• store_index – a physical operator describing the materialization of an
abstract plan derived table in a clustered index stored worktable; the
optimizer chooses the useful key columns.
• sort – a physical operator describing the sorting of an abstract plan
derived table; the optimizer chooses the useful key columns.
• nested – a filter describing the placement and structure of nested
subqueries.
• xchg – a physical operator describing the on-the-fly repartitioning of
an abstract plan derived table. The abstract plan gives the target
degree, but the optimizer chooses the useful target partitioning.
These additional abstract plan keywords are used for grouping and
identification:
• sequence – groups the elements when a sequence requires multiple
steps.
• hints – groups a set of hints for a partial plan.

Performance and Tuning Series: Query Processing and Abstract Plans 337
Overview

• prop – introduces a set of scan properties for a table: prefetch, lru|mru

and parallel.
• table – identifies a table when correlation names are used, and in
subqueries or views.
• work_t – identifies a worktable.

• in – used with table to identify tables named in a subquery (subq) or

view (view).
• subq – used under the nested operator to indicate the attachment point
for a nested subquery, and to introduce the subqueries’ abstract plan.
All legacy abstract plan operators, such as g_join, are still accepted for
their new counterparts.

Queries, access methods, and abstract plans

For any specific table, there can be several access methods for a specific
query; index scans using different indexes, table scans, the OR strategy,
and reformatting.
This simple query has several choices of access methods:
select * from t1
where c11 > 1000 and c12 < 0
The following abstract plans specify three different access methods:
• Use the index i_c11:
(i_scan i_c11 t1)
• Use the index i_c12:
(i_scan i_c12 t1)
• Do a full table scan:
(t_scan t1)
• Do a multi-scan; that is, the union or intersection of several indexes
of the table, according to the complex clause (hence the more
complex query used in this example):
select * from t1
where (c11 > 1000 or c12 < 0) and (c12 > 1000 or c112 < 0)
plan
“(m_scan t1)”

338 Adaptive Server Enterprise

 CHAPTER 13 Abstract Query Plan Guide

Abstract plans can be full plans, specifying all optimizer choices for a
query, or can specify a subset of the choices, such as the index to use for a
single table in the query, but not the join order for the tables. For example,
using a partial abstract plan, you can specify that the query above should
use some index and let the optimizer choose between i_c11 and i_c12, but
not do a full table scan. The empty parentheses are used in place of the
index name:
(i_scan () t1)
In addition, the query could use either 2K or 16K I/O, or be performed in
serial or parallel.

Derived tables
A derived table is defined by the evaluation of a query expression and
differs from a regular table in that it is neither described in system catalogs
nor stored on disk. In Adaptive Server, a derived table may be a SQL
derived table or an abstract plan derived table.
• A SQL derived table – defined by one or more tables through the
evaluation of a query expression. A SQL derived table is used in the
query expression in which it is defined and exists only for the duration
of the query. See the Transact-SQL User’s Guide.
• An abstract plan derived table – a derived table used in query
processing, the optimization and execution of queries. An abstract
plan derived table differs from a SQL derived table in that it exists as
part of an abstract plan and is invisible to the end user.

Identifying tables
Abstract plans must name all of a query’s tables in a nonambiguous way,
such that a table named in the abstract can be linked to its occurrence in
the SQL query. In most cases, the table name is all that is needed. If the
query qualifies the table name with the database and owner name, these
are also needed to fully identify a table in the abstract plan. For example,
this example uses the unqualified table name:
select * from t1
The abstract plan also uses the unqualified name, (t_scan t1). If a database
name or owner name are provided in the query:
select * from pubs2.dbo.t1

Performance and Tuning Series: Query Processing and Abstract Plans 339
Overview

The abstract plan must use qualifications, (t_scan pubs2.dbo.t1).

However, the same table may occur several times in the same query, as in
this example:
select * from t1 a, t1 b
Correlation names, a and b in the example above, identify the two tables
in SQL. In an abstract plan, the table operator associates each correlation
name with the occurrence of the table:
(join
(t_scan (table (a t1)))
(t_scan (table (b t1)))
)
You can also use a briefer abstract plan, which uses only the correlation
names:
(join
(t_scan a)
(t_scan b)
)
Table names can also be ambiguous in views and subqueries, so the table
operator is used for tables in views and subqueries.
For subqueries, the in and subq operators qualify the name of the table with
its syntactical containment by the subquery. The same table is used in the
outer query and the subquery in this example:
select *
from t1
where c11 in (select c12 from t1 where c11 > 100)
The abstract plan identifies the tables unambiguously:
(join
(t_scan t1)
(i_scan i_c11_c12 (table t1 (in (subq 1))))
)
For views, the in and view operators provide the identification. The query
in this example references a table used in the view:
create view v1
as
select * from t1 where c12 > 100
select t1.c11 from t1, v1
where t1.c12 = v1.c11
Here is the abstract plan:

340 Adaptive Server Enterprise

 CHAPTER 13 Abstract Query Plan Guide

(join
(t_scan t1)
(i_scan i_c12 (table t1 (in (view v1))))
)
In abstract plans generated by Adaptive Server, the view or subquery-
qualified table names are generated only for the tables where they are
needed to remove name ambiguity. For other tables, only the name is
generated.
In abstract plans created by the user, view or subquery-qualified tables
names are required in case of ambiguity; both syntaxes are accepted
otherwise.

Identifying indexes
The i_scan operator requires two operands, the index name and the table
name, as shown here:
(i_scan i_c12 t1)
To specify that some index should be used, without specifying the index,
substitute empty parenthesis for the index name:
(i_scan () t1)

Specifying join order

Adaptive Server performs joins of three or more tables by joining two of
the tables, and joining the abstract plan derived table from that join to the
next table in the join order. This abstract plan derived table is a flow of
rows, as from an earlier nested-loop join in the query execution.
This query joins three tables:
select *
from t1, t2, t3
where c11 = c21
and c12 = c31
and c22 = 0
and c32 = 100
This example shows the binary nature of the join algorithm, using join
operators. The plan specifies the join order t2, t1, t3:

Performance and Tuning Series: Query Processing and Abstract Plans 341
Overview

(join
(join
(scan t2)
(scan t1)
)
(scan t3)
)
The results of the t2-t1 join are then joined to t3. The scan operator in this
example leaves the choice of table scan or index scan up to the optimizer.

Shorthand notation for joins

In general, an N-way left deep nested loops join, with the order t1, t2, t3...,
tN-1, tN is described by:

(join
(join
...
(join
(join
(scan t1)
(scan t2)
)
(scan t3)
)
...
(scan tN-1)
)
(scan tN)
)
This notation can be used as shorthand for the nl_join operator:
(nl_join
(scan t1)
(scan t2)
(scan t3)
...
(scan tN-1)
(scan tN)
)

342 Adaptive Server Enterprise

 CHAPTER 13 Abstract Query Plan Guide

Join order examples

The optimizer could select among several plans for this three-way join
query:
select *
from t1, t2, t3
where c11 = c21
and c12 = c31
and c22 = 0
and c32 = 100
Here are a few examples:
• Use c22 as a search argument on t2, join with t1 on c11, then with t3
on c31:
(nl_join
(i_scan i_c22 t2)
(i_scan i_c11 t1)
(i_scan i_c31 t3)
)
• Use the search argument on t3, and the join order t3, t1, t2:
(nl_join
(i_scan i_c32 t3)
(i_scan i_c12 t1)
(i_scan i_c21 t2)
)
• Do a full table scan of t2, if it is small and fits in cache, still using the
join order t3, t1, t2:
(nl_join
(i_scan i_c32 t3)
(i_scan i_c12 t1)
(t_scan t2)
)
• If t1 is very large, and t2 and t3 individually qualify a large part of t1,
but together a very small part, this plan specifies a star join:
(nl_join
(i_scan i_c22 t2)
(i_scan i_c32 t3)
(i_scan i_c11_c12 t1)
)

Performance and Tuning Series: Query Processing and Abstract Plans 343
Overview

The join operators are generic in that they implement any of the outer
joins, inner joins, and existence joins; the optimizer chooses the correct
join semantics according to the query semantics.

Match between execution methods and abstract plans

There are some limits to join orders and join types, depending on the type
of query. One example is outer joins, such as:
select *
from t1 left join t2
on c11 = c21
Adaptive Server requires the outer member of the outer join to be the outer
table during join processing. Therefore, this abstract plan is illegal:
(join
(scan t2)
(scan t1)
)
Attempting to use this plan results in an error message, the AP application
fails, and the optimizer makes the best attempt to finish compiling the
query.

Specifying join order for queries using views

You can use abstract plans to enforce the join order for merged views. This
example creates a view that performs a join of t2 and t3:
create view v2
as
select *
from t2, t3
where c22 = c32
This query performs a join with the t2 in the view:
select * from t1, v2
where c11 = c21
and c22 = 0
This abstract plan specifies the join order t2, t1, t3:
(nl_join
(scan t2)
(scan t1)
(scan t3)

344 Adaptive Server Enterprise

 CHAPTER 13 Abstract Query Plan Guide

)
Since the table names are not ambiguous, the view qualification is not
needed. However, the following abstract plan is also legal and has the
same meaning:
(nl_join
(scan (table t2(in(view v2))))
(scan t1)
(scan (table t3 (in (view v2))))
)
This example joins with t3 in the view:
select * from t1, v2
where c11 = c31
and c32 = 100
This plan uses the join order t3, t1, t2:
(join
(scan t3)
(scan t1)
(scan t2)
)
This is an example where abstract plans can be used, if needed, to affect
the join order for a query, when set forceplan cannot.

Specifying the join type

Adaptive Server can perform nested-loop, merge, or hash joins. The join
operator leaves the optimizer free to choose the best join algorithm, based
on costing. To specify a nested-loop join, use the nl_join operator; for a
merge join, use the m_join operator, and for a hash join, use the h_join
operator. Abstract plans captured by Adaptive Server always include the
operator that specifies the algorithm, and not the join operator.
This query specifies a join between t1 and t2:
select * from t1, t2
where c12 = c21 and c11 = 0
This abstract plan specifies a nested-loop join:
(nl_join
(i_scan i_c11 t1)
(i_scan i_c21 t2)

Performance and Tuning Series: Query Processing and Abstract Plans 345
Overview

)
The nested-loop plan uses the index i_c11to limit the scan using the search
clause, and then performs the join with t2, using the index on the join
column.
This merge-join plan uses different indexes:
(m_join
(i_scan i_c12 t1)
(i_scan i_c21 t2)
)
The merge join uses the indexes on the join columns, i_c12 and i_c21, for
the merge keys. This query performs a full-merge join and no sort is
needed.
A merge join could also use the index on i_c11 to select only the matching
rows, but then a sort is needed to provide the needed ordering.
(m_join
(sort
(i_scan i_c11 t1)
)
(i_scan i_c21 t2)
)
Finally, this plan does a hash join and a full table scan on the inner side:
(h_join
(i_scan i_c11 t1)
(t_scan t2)
)

Specifying partial plans and hints

Sometimes a full plan is not needed, for example, if the only problem with
a query plan is that the optimizer chooses a table scan instead of using a
nonclustered index, the abstract plan can specify only the index choice,
and leave the other decisions to the optimizer.
The optimizer could choose a table scan of t3 rather than using i_c31 for
this query:
select *
from t1, t2, t3
where c11 = c21

346 Adaptive Server Enterprise

 CHAPTER 13 Abstract Query Plan Guide

and c12 < c31

and c22 = 0
and c32 = 100
The following plan, as generated by the optimizer, specifies join order t2,
t1, t3. However, the plan specifies a table scan of t3:
(nl_join
(i_scan i_c22 t2)
(i_scan i_c11 t1)
(t_scan t3)
)
This full plan could be modified to specify the use of i_c31 instead:
(nl_join
(i_scan i_c22 t2)
(i_scan i_c11 t1)
(i_scan i_c31 t3)
)
However, specifying only a partial abstract plan is a more flexible
solution. As data in the other tables of that query evolves, the optimal join
order can change. The partial plan can specify just one partial plan item.
For the index scan of t3, the partial plan is simply:
(i_scan i_c31 t3)
The optimizer chooses the join order and the access methods for t1 and t2.
Abstract plans are partial by using logical operators instead of physical
operators. For example, the following abstract plan is partial, although it
covers the entire query, as it lets the optimizer choose the join algorithms
and the access methods:
(join
(scan t1)
(scan t2)
(scan t3)
)
Partial plans may also be incomplete at the top, in that the root of the
abstract plan may cover only a part of the query. If this is the case, the
optimizer completes the plan:
(nl_join
(t_scan t1)
(t_scan t2)
)

Performance and Tuning Series: Query Processing and Abstract Plans 347
Overview

However, the plan fragment given in an abstract plan must be complete

down to the leafs. For example, the following abstract plan, which reads
“hash join t1 outer to something” is illegal.
(h_join
(t_scan t1)
()
)

Grouping multiple hints

Sometimes more than one plan fragment is needed. For example, you
might want to specify that some index should be used for each table in the
query, but leave the join order up to the optimizer. When multiple hints are
needed, you can group them with the hints operator:
(hints
(i_scan () t1)
(i_scan () t2)
(i_scan () t3)
)
In this case, the role of the hints operator is purely syntactic; it does not
affect the ordering of the scans.
There are no limits on what may be given as a hint. Partial join orders may
be mixed with partial access methods. This hint specifies that t2 is outer to
t1 in the join order, and that the scan of t3 should use an index, but the
optimizer can choose the index for t3, the access methods for t1 and t2, and
the placement of t3 in the join order:
(hints
(join
(scan t2)
(scan t1)
)
(i_scan () t3)
)

Inconsistent and illegal plans using hints

It is possible to describe inconsistent plans using hints, such as this plan
that specifies contradictory join orders:
(hints
(join
(scan t2)

348 Adaptive Server Enterprise

 CHAPTER 13 Abstract Query Plan Guide

(scan t1)
)
(join
(scan t1)
(scan t2)
)
)
When the query associated with the plan is executed, the query cannot be
compiled, and an error is raised.
Other inconsistent hints do not raise an exception, but may use any of the
specified access methods. This plan specifies both an index scan and a
table scan for the same table:
(hints
(t_scan t3)
(i_scan () t3)
)
In this case, either method may be chosen, and the behavior is
indeterminate.

Creating abstract plans for subqueries

Subqueries are resolved in several ways in Adaptive Server, and the
abstract plans reflect the query execution steps:
• Materialization – the subquery is executed and results are stored in a
worktable or internal variable. See “Materialized subqueries” on page
350.
• Flattening – the query is flattened into a join with the tables in the
main query. See “Flattened subqueries” on page 350.
• Nesting – the subquery is executed once for each outer query row. See
“Nested subqueries” on page 352.
Abstract plans do not allow the choice of the basic subquery resolution
method. This is a rule-based decision and cannot be changed during query
optimization. Abstract plans, however, can be used to influence the plans
for the outer and inner queries. In nested subqueries, abstract plans can
also be used to choose where the subquery is nested in the outer query.

Performance and Tuning Series: Query Processing and Abstract Plans 349
Overview

Materialized subqueries
This query includes a noncorrelated subquery that can be materialized:
select *
from t1
where c11 = (select count(*) from t2)
The first step in the abstract plan materializes the scalar aggregate in the
subquery. The second step uses the result to scan t1:

( sequence
(scalar_agg
(i_scan i_c21 t2)
)
(i_scan i_c11 t1)
)

Flattened subqueries
Some subqueries can be flattened into joins. The join, nl_join, m_join, and
h_join operators leave it to the optimizer to detect when an existence join
is needed. For example, this query includes a subquery introduced with
exists:

select * from t1
where c12 > 0
and exists (select * from t2
where t1.c11 = c21 and c22 < 100)
The semantics of the query require an existence join between t1 and t2. The
join order t1, t2 is interpreted by the optimizer as a semijoin, with the scan
of t2 stopping on the first matching row of t2 for each qualifying row in t1:
(join
(scan t1)
(scan t2)
)
The join order t2, t1 requires other means to guarantee the duplicate
elimination:
(join
(distinct
(scan t2)
)
(scan t1)
)

350 Adaptive Server Enterprise

 CHAPTER 13 Abstract Query Plan Guide

Using this abstract plan, the optimizer can decide to use:

• A unique index on t2.c21, if one exists, with a regular join.
• The unique reformatting strategy, if no unique index exists. In this
case, the query will probably use the index on c22 to select the rows
into a worktable.
• The duplicate elimination sort optimization strategy, performing a
regular join and selecting the results into the worktable, then sorting
the worktable.
The abstract plan does not need to specify the creation and scanning of the
worktables needed for the last two options.

Example of changing the join order in a flattened subquery

The query can be flattened to an existence join:
select *
from t1, t2
where c11 = c21
and c21 > 100
and exists (select * from t3 where c31 != t1.c11)
The “!=” correlation can make the scan of t3 rather expensive. If the join
order is t1, t2, the best place for t3 in the join order depends on whether the
join of t1 and t2 increases or decreases the number of rows, and therefore,
the number of times that the expensive table scan needs to be performed.
If the optimizer fails to find the right join order for t3, the following
abstract plan can be used when the join reduces the number of times that
t3 must be scanned:

(nl_join
(scan t1)
(scan t2)
(scan t3)
)
If the join increases the number of times that t3 needs to be scanned, this
abstract plan performs the scans of t3 before the join:
(nl_join
(scan t1)
(scan t3)
(scan t2)
)

Performance and Tuning Series: Query Processing and Abstract Plans 351
Overview

Nested subqueries
Nested subqueries can be explicitly described in abstract plans if:
• The abstract plan for the subquery is provided.
• The location at which the subquery attaches to the main query is
specified.
Abstract plans allow you to affect the query plan for the subquery, and to
change the attachment point for the subquery in the outer query.
The nested operator specifies the position of the subquery in the outer
query. Subqueries are “nested over” a specific abstract plan derived table.
The optimizer chooses a spot where all the correlation columns for the
outer query are available, and where it estimates that the subquery needs
to be executed the least number of times.
The following SQL statement contains a correlated expression subquery:
select *
from t1, t2
where c11 = c21
and c21 > 100
and c12 = (select c31 from t3
where c32 = t1.c11)
The abstract plan shows the subquery nested over the scan of t1:
(nl_join
(nested
(i_scan i_c12 t1)
(subq
(scalar_agg
(scan t3)
)
)
)
(i_scan i_c21 t2)
)
Aggregation is described in Chapter 2, “Using showplan.” The scalar_agg
abstract plan operator is necessary because all abstract plans, even partial
ones, must be complete down to the leafs.

352 Adaptive Server Enterprise

 CHAPTER 13 Abstract Query Plan Guide

Subquery identification and attachment

Subqueries in the SQL query are matched against abstract plan subqueries
using their underlying tables. As tables are unambiguously identified, so
are the subqueries. For example:
select
(select c11 from t1 where c12 = t3.c32), c31
from t3
where
c32 > (select c22 from t2 where c21 = t3.c31)
plan
“(nested
(nested
(t_scan t3)
(subq
(i_scan i_c11_c12 t1)
)
)
(subq
(i_scan i_c21 t2)
)
)”
However, when table names are ambiguous, the identity of the subquery is
needed to solve the table name ambiguity.
Subqueries are identified with numbers, in the order of their leading
opened parenthesis “(“.
This example has two subqueries; both refer to table t1:
select 1
from t1
where
c11 not in (select c12 from t1)
and c11 not in (select c13 from t1)
In the abstract plan, the subquery which projects out of c12 is named “1”
and the subquery which projects out of c13 is named “2”.
(nested
(nested
(t_scan t1)
(subq
(scalar_agg
(i_scan i_c11_c12 (table t1 (in (subq 1))))
)
)

Performance and Tuning Series: Query Processing and Abstract Plans 353
Overview

)
(subq
(scalar_agg
(i_scan i_c13 (table t1 (in (subq 2))))
)
)
)
In this query, the second subquery is nested in the first:
select * from t1
where c11 not in
(select c12 from t1
where c11 not in
(select c13 from t1)
In this case, the subquery that projects out of c12 is also named “1” and the
subquery that projects out of c13 is also named “2”.
(nested
(t_scan t1
(subq
(scalar_agg
(nested
(i_scan i_c12 (table t1 (in (subq 1))))
(subq
(scalar_agg
(i_scan i_c21 (table t1 (in (subq 2))))
)
)
)
)
)
)

More subquery examples: reading ordering and attachment

The nested operator has the abstract plan derived table as the first operand
and the nested subquery as the second operand. This allows an easy
vertical reading of the join order and subquery placement:
select *
from t1, t2, t3
where c12 = 0
and c11 = c21
and c22 = c32
and 0 < (select c21 from t2 where c22 = t1.c11)

354 Adaptive Server Enterprise

 CHAPTER 13 Abstract Query Plan Guide

In the plan, the join order is t1, t2, t3, with the subquery nested over the
scan of t1:
(nl_join
(nested
(i_scan i_c11 t1)
(subq
(t_scan (table t2 (in (subq 1)))
)
)
(i_scan i_c21 t2)
(i_scan i_c32 t3)
)

Modifying subquery nesting

If you modify the attachment point for a subquery, you must choose a point
at which all of the correlation columns are available.This query is
correlated to two of the tables in the outer query:
select *
from t1, t2, t3
where c12 = 0
and c11 = c21
and c22 = c32
and 0 < (select c31 from t3 where c31 = t1.c11
and c32 = t2.c22)
This plan uses the join order t1, t2, t3, with the subquery nested over the
t1-t2 join:

(nl_join
(nested
(nl_join
(i_scan i_c11_c12 t1)
(i_scan i_c22 t2)
)
(subq
(t_scan (table t3 (in (subq 1))))
)
)
(i_scan i_c32 t3)
)
Since the subquery requires columns from both outer tables, it would be
incorrect to nest it over the scan of t1 or the scan of t2; such errors are
silently corrected during optimization.

Performance and Tuning Series: Query Processing and Abstract Plans 355
Overview

However, the following abstract plan makes the legal request to nest the
subquery over the three-table join:
(nested
(nl_join
(i_scan i_c11_c12 t1)
(i_scan i_c22 t2)
(i_scan i_c32 t3)
)
(subq
(t_scan (table t3 (in (subq 1))))
)
)

Abstract plans for materialized processing of views

In most cases, view processing merges the view definition in the main
query. There are, however, cases when a view needs to be materialized, as
in the case of a self-join:
create view v3(cc31, sum_c32)
as
select c31, sum(c32)
from t3
group by c31

select *
from v3 a, v3 b
where a.c31 = b.c31
In such a case, the abstract plan exposes the worktable and the store
operator that materializes it. The two scans of the worktable are identified
through their correlation names:
(sequence
(store
(group_sorted
(i_scan i_c31 t3)
)
)
(m_join
(sort
(t_scan (work_t (a Worktable)))
( sort
(t_scan (work_t (b Worktable)))
)

356 Adaptive Server Enterprise

 CHAPTER 13 Abstract Query Plan Guide

)
)
The handling of vector aggregation in an abstract plan is described in the
next section.

Abstract plans for queries containing aggregates

This query returns a scalar aggregate:
select max(c11) from t1
There is a physical operator that implements scalar aggregation, therefore,
the optimizer has no choice. However, choosing an index on c11 allows the
max() optimization:

(scalar_agg
(i_scan ic11 t1)
)
Since the scalar aggregate is the top abstract plan operator, removing it and
using the following partial plan has the same outcome:
(i_scan ic11 t1)
The scalar_agg abstract plan is typically needed when it is part of a
subquery and the abstract plan must cover the parent query as well.
Vector aggregation is different, in that there are several physical operators
to implement the group logical operator, which means that the optimizer
has a choice to make. Thus, the abstract plan can force it.
select max(c11)
from t1
group by c12
The following abstract plan examples force each of the three vector
aggregation algorithms:

Note group_sorted requires an ordering on the grouping column, so it

needs to use an index.

(group_sorted
(i_scan i_c12 t1)
)
(group_hashing
(t_scan t1)

Performance and Tuning Series: Query Processing and Abstract Plans 357
Overview

(group_inserting
(t_scan t1)
)

Abstract plans for queries containing unions

The union abstract plan operator describes plans for SQL queries that
contain unions:
select*
from
t1,
(select * from t2
union
select * from t3
) u(u1, u2)
where c11=u1
plan
“(nl_join
(union
(t_scan t2)
(t_scan t3)
)
(i_scan i_c11 t1)
)”
There are two types of union in SQL: union distinct and union [all]. union [all]
is the default.
The m_union_distinct and h_union_distinct abstract plan operators force the
removal of merge or hash-based UNION DISTINCT duplicates. It is illegal
to use these operators with a UNION ALL. The merge-based algorithm
needs, from each of the union children, an ordering covering all union
projection columns.
In the following example, the needed ordering is provided, for the first
child, by the (c11, c12) composite index and, for the second child, by
the sort.
select c11, c12 from t1
union distinct
select c21, c22 from t2
plan
“(m_union distinct

358 Adaptive Server Enterprise

 CHAPTER 13 Abstract Query Plan Guide

(i_scan i_c11_c12 t1)

(sort
(t_scan t2)
)
)”
The union_all and m_union_all abstract plan operators force the append- or
merge-based UNION ALL. It is illegal to use these operators with a UNION
DISTINCT. The merge algorithm needs no ordering for itself; it makes any
useful ordering from the children available to the parent.
In the following example, the ordering provided by the two i_scan
operators is made available, by their m_union_all parent, to the m_join
above.
select *
from
t1,
(select c21, c22 from t2
union
select c31, c32 from t3
) u(u1, u2)
where c11=u1
plan
“(m_join
(m_union_all
(i_scan i_c21 t2)
(i_scan i_c31 t3)
)
(i_scan i_c11 t1)
)”

Using abstract plans when queries need ordering

An ordering is needed either explicitly, in an ORDER BY query, or
implicitly by merge-based operators such as m_join, m_union_distinct, and
group_sorted.

An ordering is produced either explicitly, by the sort abstract plan operator

(the optimizer build the sort key on all columns known to need an
ordering), or implicitly by an i_scan on the indexed columns.
All merge-based operators that require ordering preserve it in their results
for a parent that also requires it.

Performance and Tuning Series: Query Processing and Abstract Plans 359
Overview

In the following example, the i_scan of t1 provides the ordering needed by

the m_join. The i_scan of t2, and the sort over t3’s scan, provides the
ordering needed by m_union_distinct. This ordering also provides the
ordering needed by the m_join. Finally, no top sort is required as the
ordering needed by ORDER BY is provided by the m_join.
select *
from
t1,
(select c21, c22 from t2
union distinct
select c31, c32 from t3
) u(u1, u2)
where c11=u1
order by c11, u2
plan
“(m_join
(m_union_distinct
(i_scan i_c21_c22 t2)
(sort
(t_scan t3)
)
)
(i_scan i_c11 t1)
)”

Specifying the reformatting strategy

In this query, t2 is very large, and has no index:
select *
from t1, t2
where c11 > 0
and c12 = c21
and c22 = 0
The abstract plan that specifies the reformatting strategy on t2 is:
(nl_join
(t_scan t1)
(store_index
(t_scan t2)
)
)

360 Adaptive Server Enterprise

 CHAPTER 13 Abstract Query Plan Guide

The store_index abstract plan operator must be placed on the inner side of
an nl_join. It can be placed over any abstract plan; there is no longer a
single table scan limitation. The legacy (scan (store... )) syntax is
still accepted.

Specifying the OR strategy

An OR strategy uses a set of index scans to limit the scan with each of the
OR terms, then passes the resulting row IDs through a UnionDistinct
operator to get, with a RidJoin from the table, the tuples corresponding to
the unique row IDs.
The m_scan (multiscan) abstract plan operator forces index union, hence
the OR strategy:
select * from t1
where c11 > 10 or c12 > 100
plan
“(m_scan t1)”

When the store operator is not specified

Storing the stream of tuples into a worktable to meet the intraoperator
needs of an algorithm (Sort, GroupInserting, and so on), is treated as a
implementation detail of the algorithm and thus is not exposed in the
abstract plan.
Abstract plans expose only the worktables created for interoperator
reasons, such as the self-joined materialized view. In such a case, none of
the operators needs a work table. The cause is, rather, the global nature of
the plan, of computing an intermediate derived table once and using it
twice.

Abstract plans for parallel processing

Partitioned tables scanned in parallel produce partitioned streams of
tuples. Different operators have specific needs for parallel processing. For
instance, in all joins, either both children must be equipartitioned, or one
child must be replicated.

Performance and Tuning Series: Query Processing and Abstract Plans 361
Overview

The abstract plan xchg operator forces the optimizer to repartition, on-the-
fly, in n ways, its child-derived table. The abstract plan only gives the
degree. The optimizer chooses the most useful partitioning columns and
style (hash, range, list, or round-robin).
In the following example, assume that t1 and t2 are hash partitioned two
ways and three ways on the join columns, and i_c21 is a local index:
select *
from t1, t2
where c11=c21
The following abstract plan repartitions t1 three ways, does a three-way
parallel nl_join, serializes the results, and returns a single data stream to the
client:
(xchg 1
(nl_join
(xchg 3
(t_scan t1)
)
(i_scan i_c21 t2)
)
)
It is not necessary to specify t2’s parallel scan. It is hash-partitioned three
ways, and, as it is joined with an xchg-3, no other plan is legal.
The following abstract plan scans and sorts t1 and t2 in parallel, as each is
partitioned, then serializes them for the m_join:
(m_join
(xchg 1
(sort
(t_scan t1)
)
)
(xchg 1
(sort
(t_scan t2)
)
)
)
(prop t1 (parallel 2))
(prop t2 (parallel 3))
The parallel abstract plan construct is used to make sure that the optimizer
chooses the parallel scan with the native degree.

362 Adaptive Server Enterprise

 CHAPTER 13 Abstract Query Plan Guide

Tips on writing abstract plans

Here are some additional tips for writing and using abstract plans:
• Look at the current plan for the query and at plans that use the same
query execution steps as the plan you need to write. It is often easier
to modify an existing plan than to write a full plan from scratch.
• Capture the plan for the query.
• Use sp_help_qplan to display the SQL text and plan.
• Edit this output to generate a create plan command, or attach an
edited plan to the SQL query using the plan clause.
• It is often best to specify partial plans for query tuning in cases where
most optimizer decisions are appropriate, but only an index choice,
for example, needs improvement.
By using partial plans, the optimizer can choose other paths for other
tables as the data in other tables changes.
• Once saved, abstract plans are static. Data volumes and distributions
may change so that saved abstract plans are no longer optimal.
Subsequent tuning changes made by adding indexes, partitioning a
table, or adding buffer pools may mean that some saved plans are not
performing as well as possible under current conditions. Most of the
time, operate with a small number of abstract plans that solve specific
problems.
Perform periodic plan checks to verify that the saved plans are still
better than the plan that the optimizer would choose.

Using abstract plans at the query level

You can use abstract plans to force the query plan the query processor
choses to allow several query-level settings. See Chapter 7, “Controlling
Optimization,” for more information about using abstract plans at the
query level.
The optimization criteria are handled at the session level by the following
set statements:

Performance and Tuning Series: Query Processing and Abstract Plans 363
Using abstract plans at the query level

set
nl_join|merge_join|hash_join|...
on | off
The use ... abstract plan syntax accepts any number of use forms before
the abstract plan derived table. In versions of Adaptive Server earlir than
15.0, optgoal and opttimeout could not be in the same abstract plan with a
derived table. For example, this statement would need to be separate from
a optgoal statement in a query:
select ...
plan
"(use opttimeoutlimit 10) (i_scan r)"
However, you can include several statements in the same abstract plan by:
• Using several use statements. For example:
select ...
plan
"(use optgoal allrows_dss) (use nl_join off)
(...)"
• Placing several items within one use form. For example:
select ...
plan
"(use (optgoal allrows_dss) (nl_join off))
(...)"
At the query level, use the optimization goal (opt_goal) or timeout
(opttimeout) setting with the use ... abstract plan syntax. At the session
level, use these settings with the set plan ... syntax:
• Optimization goal.
• Optimization timeout
For example, join r outer to s and enable the hash_join without an
optimization goal (opt_goal):
select ...
>
> plan
>
"(use hash_join on)
>
> (join (scan r) (scan s))"
This example uses the opt_goal and allrows_oltp statements, but with
hash_join enabled:

364 Adaptive Server Enterprise

 CHAPTER 13 Abstract Query Plan Guide

select ...
>
> plan
>
> "(use opt_goal allrows_oltp)(use hash_join
on)"
When setting the optimization goal and the optimization criteria at the
query level, the order of the use statements does not affect the outcome.
• The abstract plan optimization goal is set first, and sets the
optimization goal defaults for the optimization criteria.
• You can set abstract plan optimization, which supersedes
optimization goal defaults criteria, you set the optimization goal.

Operator name alignment for abstract plan and optimizer criteria

The names of algorithms differ in how you use them in abstract plans and
how you use them in the set command. For example, a hash join is called
h_join in abstract plans, but is called hash_join in the set command.
Adaptive Server accepts both keywords in the extended abstract plan
syntax. For example:
select ...

plan

"(h_join (t_scan r) (t_scan s))"

is equivalent to:
select ...

plan

"(hash_join (t_scan r) (t_scan s))"

and:
select ...
plan
"(use h_join on)"
and:
select ...
plan

Performance and Tuning Series: Query Processing and Abstract Plans 365
Comparing plans before and after

"(use hash_join on)"

When a table abstract plan is present, it takes precedence:
select ..
from r, s, t
...
plan
"(use hash_join off)
(h_join (t_scan r) (t_scan s))"
The query uses the hash_join for r and s scans; but for the join with t it does
not use hash_join as specified by the use abstract plan form, since it was
not specified in the table abstract plan.

Extending the optimizer criteria set syntax

The set opt criteria statement accepts on/off/default, where default indicates
that you are using the current optimization goal setting for this
optimization criteria (for the complete set syntax, see Reference Manual:
Commands).

Comparing plans before and after

Use abstract query plans to assess the impact of an Adaptive Server
software upgrade or system tuning changes on your query plans. You must
save plans before the changes are made, perform the upgrade or tuning
changes, and then save plans again and compare the plans. The basic set
of steps is:
1 Enable server-wide capture mode by setting the configuration
parameter abstract plan dump to 1. All plans are then captured in the
default group, ap_stdout.
2 Allow enough time for the captured plans to represent most of the
queries run on the system. You can check whether additional plans are
being generated by checking whether the count of rows in the
ap_stdout group in sysqueryplans is stable:

select count(*) from sysqueryplans where gid = 2

366 Adaptive Server Enterprise

 CHAPTER 13 Abstract Query Plan Guide

3 Copy all plans from ap_stdout to ap_stdin (or some other group, if you
do not want to use server-wide plan load mode), using
sp_copy_all_qplans.

4 Drop all query plans from ap_stdout, using sp_drop_all_qplans.

5 Perform the upgrade or tuning changes.
6 Allow sufficient time for plans to be captured to ap_stdout.
7 Compare plans in ap_stdout and ap_stdin, using the diff mode
parameter of sp_cmp_all_qplans. For example, this query compares
all plans in ap_stdout and ap_stdin:
sp_cmp_all_qplans ap_stdout, ap_stdin, diff
This displays only information about the plans that are different in the
two groups.

Effects of enabling server-wide capture mode

When server-wide capture mode is enabled, plans for all queries that can
be optimized are saved in all databases on the server. Some possible
system administration impacts are:
• When plans are captured, the plan is saved in sysqueryplans and log
records are generated. The amount of space required for the plans and
log records depends on the size and complexity of the SQL statements
and query plans. Check space in each database where users will be
active.
You may need to perform more frequent transaction log dumps,
especially in the early stages of server-wide capture when many new
plans are being generated.
• If users execute system procedures from the master database, and
installmaster was loaded with server-wide plan capture enabled, then
plans for the statements that can be optimized in system procedures
are saved in master.sysqueryplans.
This is also true for any user-defined procedures created while plan
capture was enabled. You may want to provide a default database at
login for all users, including system administrators, if space in master
is limited.

Performance and Tuning Series: Query Processing and Abstract Plans 367
Abstract plans for stored procedures

• The sysqueryplans table uses datarows locking to reduce lock

contention. However, especially when a large number of new plans
are being saved, there may be a slight impact on performance.
• While server-wide capture mode is enabled, using bcp saves query
plans in the master database. If you perform bcp using a large number
of tables or views, check sysqueryplans and the transaction log in
master.

Time and space to copy plans

If you have a large number of query plans in ap_stdout, be sure there is
sufficient space to copy them on the system segment before starting the
copy. Use sp_spaceused to check the size of sysqueryplans, and
sp_helpsegment to check the size of the system segment.

Copying plans also requires space in the transaction log.

sp_copy_all_qplans calls sp_copy_qplan for each plan in the group to be
copied. If sp_copy_all_qplans fails at any time due to lack of space or other
problems, any plans that were successfully copied remain in the target
query plan group.

Abstract plans for stored procedures

For abstract plans to be captured for the SQL statements that can be
optimized in stored procedures:
• The procedures must be created while plan capture or plan association
mode is enabled. (This saves the text of the procedure in
sysprocedures.)

• The procedure must be executed with plan capture mode enabled, and
the procedure must be read from disk, not from the procedure cache.
This sequence of steps captures the query text and abstract plans for all
statements in the procedure that can be optimized:
set plan dump dev_plans on
go
create procedure myproc as ...
go

368 Adaptive Server Enterprise

 CHAPTER 13 Abstract Query Plan Guide

exec myproc
go
If the procedure is in cache, and the plans for the procedure are not being
captured, execute the procedure with recompile. Similarly, once a stored
procedure has been executed using an abstract query plan, the plan in the
procedure cache is used so that query plan association does not take place
unless the procedure is read from disk.
You can use set fmtonly on to capture plans for a stored procedure without
actually executing the statements in a stored procedure.

Procedures and plan ownership

When plan capture mode is enabled, abstract plans for the statements in a
stored procedure that can be optimized are saved with the user ID of the
owner of the procedure.
During plan association mode, association for stored procedures is based
on the user ID of the owner of the procedure, not the user who executes the
procedure. This means that once an abstract query plan is created for a
procedure, all users who have permission to execute the procedure use the
same abstract plan.

Procedures with variable execution paths and optimization

Executing a stored procedure saves abstract plans for each statement that
can be optimized, even if the stored procedure contains control-of-flow
statements that can cause different statements to be run, depending on
parameters to the procedure or other conditions.
Adaptive Server loads and saves the abstract plans when the stored
procedures are compiled, not when they are executed.
When Adaptive Server compiles a stored procedure (usually when it is
first run), it saves an abstract plan for each optimized statement. Adaptive
Server does not influence the abstract plan capture, or whether the stored
procedure contains control-of-flow statements that cause different
statements to be executed, depending on the procedure’s parameters.

Performance and Tuning Series: Query Processing and Abstract Plans 369
Ad hoc queries and abstract plans

If you run the query a second time (without recompilation) with different
parameters that use a different code path, because Adaptive Server already
optimized and saved the plans for all statements from the earlier
compilation, both the plans and the abstract plans for the statements in this
different code path are available, and are based on the prior stored
procedure’s run parameter values, whether or not these statement were
executed.
However, abstract plans for procedures do not solve the problem caused
by procedures with statements that are optimized differently depending on
conditions or parameters. For example is a procedure where users provide
the low and high values for a between clause, with a query such as:
select title_id
from titles
where price between @lo and @hi
Depending on the parameters, the best plan could either be an index access
or a table scan. The abstract plan may specify either access method,
depending on the parameters used for the initial execution of the
procedure. Abstract plans that are saved while executing queries or stored
procedures in tempdb are lost if the server is restarted.

Ad hoc queries and abstract plans

Abstract plan capture saves the full text of the SQL query and abstract plan
association is based on the full text of the SQL query. If users submit ad
hoc SQL statements, rather than using stored procedures or Embedded
SQL, abstract plans are saved for each different combination of query
clauses. This can result in a very large number of abstract plans.
For exampe, if users check the price of a specific title_id using select
statements, an abstract plan is saved for each statement. The following two
queries each generate an abstract plan:
select price from titles where title_id = "T19245"
select price from titles where title_id = "T40007"
In addition, there is one plan for each user, that is, if several users check
for the title_id “T40007,” a plan is save for each user ID.
If such queries are included in stored procedures, there are two benefits:
• Only only one abstract plan is saved, for example, for the query:

370 Adaptive Server Enterprise

 CHAPTER 13 Abstract Query Plan Guide

select price from titles where title_id =

@title_id
• The plan is saved with the user ID of the user who owns the stored
procedure, and abstract plan association is made based on the
procedure owner’s ID.
Using Embedded SQL, the only abstract plan is saved with the host
variable:
select price from titles
where title_id = :host_var_id

Performance and Tuning Series: Query Processing and Abstract Plans 371
Ad hoc queries and abstract plans

372 Adaptive Server Enterprise

CH A PTE R 1 4 Managing Abstract Plans with
System Procedures

This chapter provides an introduction to the basic functionality and use of

the system procedures for working with abstract plans. For detailed
information on each procedure, see the Reference Manual: Procedures.
Topic Page
Managing an abstract plan group 373
Finding abstract plans 377
Managing individual abstract plans 377
Managing all plans in a group 381
Importing and exporting groups of plans 385

Managing an abstract plan group

You can use system procedures to create, drop, rename, and provide
information about an abstract plan group.

Creating a group
Use sp_add_qpgroup to create and names an abstract plan group. Unless
you are using the default capture group, ap_stdout, you must create a plan
group before you can begin capturing plans. For example, to start saving
plans in a group called dev_plans, you must create the group, then issue
the set plan dump command, specifying the group name:
sp_add_qpgroup dev_plans
set plan dump dev_plans on
/*SQL queries to capture*/
Only a system administrator or database owner can add abstract plan
groups. Once a group is created, any user can dump or load plans from the
group.

Performance and Tuning Series: Query Processing and Abstract Plans 373
Managing an abstract plan group

Dropping a group
Use sp_drop_qpgroup to drop an abstract plan group.
The following restrictions apply to sp_drop_qpgroup:
• Only a system administrator or database owner can drop abstract plan
groups.
• You cannot drop a group that contains plans. To remove all plans from
a group, use sp_drop_all_qplans, specifying the group name.
• You cannot drop the default abstract plan groups ap_stdin and
ap_stdout.

This command drops the dev_plans plan group:

sp_drop_qpgroup dev_plans

Getting information about a group

sp_help_qpgroup prints information about an abstract plan group, or about
all abstract plan groups in a database.
When you use sp_help_qpgroup without a group name, it prints the names
of all abstract plan groups, the group IDs, and the number of plans in each
group:
sp_help_qpgroup
Query plan groups in database ‘pubtune’
Group GID Plans
------------------------------ ----------- -----------
ap_stdin 1 0
ap_stdout 2 2
p_prod 4 0
priv_test 8 1
ptest 3 51
ptest2 7 189

When you use sp_help_qpgroup with a group name, the report provides
statistics about plans in the specified group. This example reports on the
group ptest2:
sp_help_qpgroup ptest2
Query plans group 'ptest2', GID 7

Total Rows Total QueryPlans

374 Adaptive Server Enterprise

 CHAPTER 14 Managing Abstract Plans with System Procedures

----------- ----------------
452 189
sysqueryplans rows consumption, number of query
plans per row count
Rows Plans
----------- -----------
5 2
3 68
2 119
Query plans that use the most sysqueryplans rows
Rows Plan
----------- -----------
5 1932533918
5 1964534032
Hashkeys
-----------
123
There is no hash key collision in this group.
When reporting on an individual group, sp_help_qpgroup reports:
• The total number of abstract plans, and the total number of rows in the
sysqueryplans table.

• The number of plans that have multiple rows in sysqueryplans. The

plans are listed in descending order, starting with the plans with the
largest number of rows.
• Information about the number of hash keys and hash-key collisions.
Abstract plans are associated with queries by a hashing algorithm
over the entire query.
When a system administrator or the database owner executes
sp_help_qpgroup, the procedure reports on all of the plans in the database
or in the specified group. When any other user executes sp_help_qpgroup,
it reports only on plans that he or she owns.
sp_help_qpgroup provides several report modes. The report modes are:

Mode Information returned

full The number of rows and number of plans in the group, the number of plans that use two or more
rows, the number of rows and plan IDs for the longest plans, and number of hash keys, and has-
key collision information. This is the default report mode.
stats All of the information from the full report, except hash-key information.
hash The number of rows and number of abstract plans in the group, the number of hash keys, and hash-
key collision information.

Performance and Tuning Series: Query Processing and Abstract Plans 375
Managing an abstract plan group

Mode Information returned

list The number of rows and number of abstract plans in the group, and the following information for
each query/plan pair: hash key, plan ID, first few characters of the query, and the first few
characters of the plan.
queries The number of rows and number of abstract plans in the group, and the following information for
each query: hash key, plan ID, first few characters of the query.
plans The number of rows and number of abstract plans in the group, and the following information for
each plan: hash key, plan ID, first few characters of the plan.
counts The number of rows and number of abstract plans in the group, and the following information for
each plan: number of rows, number of characters, hash key, plan ID, first few characters of the
query.

This example shows the output for the counts mode:

sp_help_qpgroup ptest1, counts
Query plans group 'ptest1', GID 3

Total Rows Total QueryPlans

----------- ----------------
48 19

Query plans in this group

Rows Chars hashkey id query

----- --------- ----------- ----------- ----------------------------
3 623 1801454852 876530156 select title from titles ...
3 576 476063777 700529529 select au_lname, au_fname...
3 513 444226348 652529358 select au1.au_lname, au1....
3 470 792078608 716529586 select au_lname, au_fname...
3 430 789259291 684529472 select au1.au_lname, au1....
3 425 1929666826 668529415 select au_lname, au_fname...
3 421 169283426 860530099 select title from titles ...
3 382 571605257 524528902 select pub_name from publ...
3 355 845230887 764529757 delete salesdetail where ...
3 347 846937663 796529871 delete salesdetail where ...
2 379 1400470361 732529643 update titles set price =...

Renaming a group
A system administrator or database owner can rename an abstract plan
group with sp_rename_qpgroup. This example changes the name of the
group from dev_plans to prod_plans:

376 Adaptive Server Enterprise

 CHAPTER 14 Managing Abstract Plans with System Procedures

sp_rename_qpgroup dev_plans, prod_plans

The new group name cannot be the name of an existing group.

Finding abstract plans

Use sp_find_qplan to search both the query text and the plan text to find
plans that match a given pattern.
This example finds all plans where the query includes the string “from
titles”:
sp_find_qplan "%from titles%"
This example searches for all abstract plans that perform a table scan:
sp_find_qplan "%t_scan%"
When a system administrator or database owner executes sp_find_qplan,
the procedure examines and reports on plans owned by all users. When
other users execute the procedure, sp_find_qplan searches and reports on
only plans that they own.
To search just one abstract plan group, specify the group name. This
example searches only the test_plans group, finding all plans that use a
particular index:
sp_find_qplan "%i_scan title_id_ix%", test_plans
For each matching plan, sp_find_qplan prints the group ID, plan ID, query
text, and abstract plan text.

Managing individual abstract plans

You can use system procedures to print the query and text of individual
plans, to copy, drop, or compare individual plans, or to change the plan
associated with a particular query.

Performance and Tuning Series: Query Processing and Abstract Plans 377
Managing individual abstract plans

Viewing a plan
Use sp_help_qplan to report on individual abstract plans. It provides three
types of reports that you can specify: brief, full, and list. The brief report
prints only the first 78 characters of the query and plan; use full to see the
entire query and plan, or list to display only the first 20 characters of the
query and plan.
This example prints the default brief report:
sp_help_qplan 588529130
gid hashkey id
----------- ----------- -----------
8 1460604254 588529130
query
---------------------------------------------------------------
select min(price) from titles
plan
---------------------------------------------------------------
(plan
(i_scan type_price titles)
()
)
(prop titles
(parallel ...
A system administrator or database owner can use sp_help_qplan to report
on any plan in the database. Other users can view only the plans that they
own.
sp_help_qpgroup reports on all plans in a group. See “Getting information
about a group” on page 374.

Copying a plan to another group

Use sp_copy_qplan to copy an abstract plan from one group to another
existing group. This example copies the plan with plan ID 316528161
from its current group to the prod_plans group:
sp_copy_qplan 316528161, prod_plans
sp_copy_qplan verifies that the query does not already exist in the
destination group. If a possible conflict exists, sp_copy_qplan runs
sp_cmp_qplans to check plans in the destination group. In addition to the
message printed by sp_cmp_qplans, sp_copy_qplan prints messages when:

378 Adaptive Server Enterprise

 CHAPTER 14 Managing Abstract Plans with System Procedures

• The query and plan you are trying to copy already exists in the
destination group
• Another plan in the group has the same user ID and hash key
• Another plan in the group has the same hash key, but the queries are
different
If there is a hash-key collision, the plan is copied. If the plan already exists
in the destination group or if it would give an association key collision, the
plan is not copied. The messages printed by sp_copy_qplan contain the
plan ID of the plan in the destination group, so you can use sp_help_qplan
to check the query and plan.
A system administrator or the database owner can copy any abstract plan.
Other users can copy only plans that they own. The original plan and group
are not affected by sp_copy_qplan. The copied plan is assigned a new plan
ID, the ID of the destination group, and the user ID of the user who ran the
query that generated the plan.

Dropping an individual abstract plan

Use sp_drop_qplan to drop individual abstract plans. This example drops
the specified plan:
sp_drop_qplan 588529130
A system administrator or database owner can drop any abstract plan in the
database. Other users can drop only plans that they own.
To find abstract plan IDs, use sp_find_qplan to search for plans using a
pattern from the query or plan, or use sp_help_qpgroup to list the plans in
a group.

Comparing two abstract plans

Given two plan IDs, sp_cmp_qplans compares two abstract plans and the
associated queries. For example:
sp_cmp_qplans 588529130, 1932533918
sp_cmp_qplans prints one message reporting the comparison of the query,
and a second message about the plan, as follows:
• For the two queries, one of:

Performance and Tuning Series: Query Processing and Abstract Plans 379
Managing individual abstract plans

• The queries are the same.

• The queries are different.
• The queries are different but have the same hash key.
• For the plans:
• The query plans are the same.
• The query plans are different.
This example compares two plans where the queries and plans both match:
sp_cmp_qplans 411252620, 1383780087
The queries are the same.
The query plans are the same.
This example compares two plans where the queries match, but the plans
are different:
sp_cmp_qplans 2091258605, 647777465
The queries are the same.
The query plans are different.
sp_cmp_qplans returns a status value showing the results of the
comparison.
Table 14-1: Return status values for sp_cmp_qplans
Return value Meaning
0 The query text and abstract plans are the same.
+1 The queries and hash keys are different.
+2 The queries are different, but the hash keys are the same.
+10 The abstract plans are different.
100 One or both of the plan IDs does not exist.

A system administrator or database owner can compare any two abstract

plans in the database. Other users can compare only plans that they own.

Changing an existing plan

Use sp_set_qplan to change the abstract plan for an existing plan ID
without changing the ID or the query text. You can use sp_set_qplan only
when the plan text is 255 or fewer characters.
sp_set_qplan 588529130, "(i_scan title_ix titles)"

380 Adaptive Server Enterprise

 CHAPTER 14 Managing Abstract Plans with System Procedures

A system administrator or database owner can change the abstract plan for
any saved query. Other users can modify only plans that they own.
When you execute sp_set_qplan, the abstract plan is not checked against
the query text to determine whether the new plan is valid for the query, or
whether the tables and indexes exist. To test the validity of the plan,
execute the associated query.
You can also use create plan and the plan clause to specify the abstract plan
for a query. See “Creating plans using SQL” on page 331.

Managing all plans in a group

You can use system procedures to copy all plans in one abstract plan group
to another group, compare all abstract plans in two groups and reports, and
drop all abstract plans in a group.

Copying all plans in a group

Use sp_copy_all_qplans to copy all of the plans in one abstract plan group
to another group. This example copies all of the plans from the test_plans
group to the helpful_plans group:
sp_copy_all_qplans test_plans, helpful_plans
The helpful_plans group must exist before you execute sp_copy_all_qplans.
It can contain other plans.
sp_copy_all_qplans copies each plan in the group by executing
sp_copy_qplan, so copying a plan may fail for the same reasons that
sp_copy_qplan might fail. See “Comparing two abstract plans” on page
379.
Each plan is copied as a separate transaction, and failure to copy any single
plan does not cause sp_copy_all_qplans to fail. If sp_copy_all_qplans fails
for any reason, and has to be restarted, you see a set of messages for the
plans that have already been successfully copied, telling you that they exist
in the destination group.

Performance and Tuning Series: Query Processing and Abstract Plans 381
Managing all plans in a group

A new plan ID is assigned to each copied plan. The copied plans have the
original user’s ID. To copy abstract plans and assign new user IDs, you
must use sp_export_qpgroup and sp_import_qpgroup. See “Importing and
exporting groups of plans” on page 385.
A system administrator or database owner can copy all plans in the
database. Other users can copy only plans that they own.

Comparing all plans in a group

Use sp_cmp_all_qplans to compare all abstract plans in two groups and
reports:
• The number of plans that are the same in both groups
• The number of plans that have the same association key, but different
abstract plans
• The number of plans that are present in one group, but not the other
This example compares the plans in ap_stdout and ap_stdin:
sp_cmp_all_qplans ap_stdout, ap_stdin
If the two query plans groups are large, this might take some
time.
Query plans that are the same
count
-----------
338
Different query plans that have the same association key

count
-----------
25
Query plans present only in group ’ap_stdout’ :

count
-----------
0
Query plans present only in group ’ap_stdin’ :

count
-----------
1

382 Adaptive Server Enterprise

 CHAPTER 14 Managing Abstract Plans with System Procedures

With the additional specification of a report-mode parameter,

sp_cmp_all_qplans provides detailed information, including the IDs,
queries, and abstract plans of the queries in the groups. The mode
parameter lets you get the detailed information for all plans, or just those
with specific types of differences.
Table 14-2: Report modes for sp_cmp_all_qplans
Mode Reported information
counts The counts of plans that are the same, plans that have the same association key, but different groups,
and plans that exist in one group, but not the other. This is the default report mode.
brief The information provided by counts, plus the IDs of the abstract plans in each group where the plans
are different, but the association key is the same, and the IDs of plans that are in one group, but not
in the other.
same All counts, plus the IDs, queries, and plans for all abstract plans where the queries and plans match.
diff All counts, plus the IDs, queries, and plans for all abstract plans where the queries and plans are
different.
first All counts, plus the IDs, queries, and plans for all abstract plans that are in the first plan group, but
not in the second plan group.
second All counts, plus the IDs, queries, and plans for all abstract plans that are in the second plan group,
but not in the first plan group.
offending All counts, plus the IDs, queries, and plans for all abstract plans that have different association keys
or that do not exist in both groups. This is the combination of the diff, first, and second modes.
full All counts, plus the IDs, queries, and plans for all abstract plans. This is the combination of same
and offending modes.

This example shows the brief report mode:

sp_cmp_all_qplans ptest1, ptest2, brief
If the two query plans groups are large, this might take some time.
Query plans that are the same

count
-----------
39
Different query plans that have the same association key

count
-----------
4

ptest1 ptest2

id1 id2
----------- -----------

Performance and Tuning Series: Query Processing and Abstract Plans 383
Managing all plans in a group

764529757 1580532664
780529814 1596532721
796529871 1612532778
908530270 1724533177
Query plans present only in group ’ptest1’ :

count
-----------
3

id
-----------
524528902
1292531638
1308531695

Query plans present only in group ’ptest2’ :

count
-----------
1

id
-----------
2108534545

Dropping all abstract plans in a group

Use sp_drop_all_qplans to drop all abstract plans in a group. This example
drops all abstract plans in the dev_plans group:
sp_drop_all_qplans dev_plans
When a system administrator or the database owner executes
sp_drop_all_qplans, all plans belonging to all users are dropped from the
specified group. When another user executes this procedure, it affects only
the plans owned by that user.

384 Adaptive Server Enterprise

 CHAPTER 14 Managing Abstract Plans with System Procedures

Importing and exporting groups of plans

Use sp_export_qpgroup and sp_import_qpgroup to copy groups of plans
between sysqueryplans and a user table. This allows a system
administrator or database owner to:
• Copy abstract plans from one database to another on the same server
• Create a table that can be copied out of the current server with bcp,
and copied into another server
• Assign different user IDs to existing plans in the same database

Exporting plans to a user table

Use sp_export_qpgroup to copy all plans for a specific user from an
abstract plan group to a user table. This example copies plans owned by
the database owner (dbo) from the fast_plans group, creating a table called
transfer:
sp_export_qpgroup dbo, fast_plans, transfer
sp_export_qpgroup uses select...into to create a table with the same
columns and datatypes as sysqueryplans. If you do not have the
select into/bulkcopy/pllsort option enabled in the database, you can specify
the name of another database. This command creates the export table in
tempdb:

sp_export_qpgroup mary, ap_stdout, "tempdb..mplans"

The table can be copied out using bcp, and copied into a table on another
server. The plans can also be imported to sysqueryplans in another
database on the same server, or the plans can be imported into
sysqueryplans in the same database, with a different group name or user
ID.

Importing plans from a user table

Use sp_import_qpgroup to copy plans from tables created by
sp_export_qpgroup into a group in sysqueryplans. This example copies the
plans from the table tempdb.mplans into ap_stdin, assigning the user ID for
the database owner:
sp_import_qpgroup "tempdb..mplans", dbo, ap_stdin

Performance and Tuning Series: Query Processing and Abstract Plans 385
Importing and exporting groups of plans

You cannot copy plans into a group that already contains plans for the
specified user.

386 Adaptive Server Enterprise

Index

A cursors and 277

index specification 235
abstract plan cache configuration parameter 331
associating queries with plans
abstract plan derived tables 339
plan groups and 318
abstract plan dump configuration parameter 331
session-level 323
abstract plan groups
association key
adding 373
defined 319
creating 373
plan association and 319
dropping 374
sp_cmp_all_qplans and 382
exporting 385
sp_copy_qplan and 379
importing 385
attribute-insensitive operation
information about 374
parallelism 155
overview of use 318
attribute-sensitive operation
plan association and 318
parallelism 169
plan capture and 318
automatically
procedures for managing 373–385
update statistics 297
abstract plan load configuration parameter 331
abstract plan replace configuration parameter 331
abstract plans
comparing 379 B
copying 378
buffers unavailable 238
finding 377
information about 378
pattern matching 377
viewing with sp_help_qplan 378 C
accessing
query processing metrics 280 capturing plans
adding session-level 322
abstract plan groups 373 cheap direct updates 28
statistics for unindexed columns 290 clearing query processing metrics 284
close command
adding statistics 290
adjustment memory and 266
close on endtran option, set 277
managing runtime 200
recognizing runtime 200 clustered indexes
reducing runtime 201 prefetch and 237
runtime 199 column-level statistics 300
ALS generating the update statistics 305
truncate table and 301
user log cache 243
update statistics and 300
ALS. see Asynchronous Log Service 241
application design comparing abstract plans 379
composite indexes

Performance and Tuning Series: Query Processing and Abstract Plans 387
Index

update index statistics and 305 deallocate cursor command

compute by processing 88 memory and 266
concurrency optimization debugging aids
for small tables 259 set forceplan on 232
concurrency optimization threshold declare cursor command
deadlocks and 259 memory and 265
connections default settings
cursors and 278 number of tables optimized 234
controlling parallelism at session level 139 deferred
controlling parallelism for a query 141 index updates 31
converted search arguments 6 updates 30
copying degree
abstract plans 378 setting max parallel 136
plan groups 381 delete operations
plans 378, 381 joins and update mode 30
covered queries update mode in joins 30
specifying cache strategy for 239 delete operator 62, 195
creating delete statistics command
abstract plan groups 373 managing statistics and 310
column statistics 302 deleting plans 379, 384
search arguments 18 density join 12
cursor rows option, set 277 derived tables
cursors abstract plan derived tables 339
execute 266 SQL 19
Halloween problem 268 SQL derived tables 339
indexes and 267 differing parallel query results 144
isolation levels and 273 direct updates 27
locking and 264 cheap 28
modes 266 expensive 29
multiple 278 in-place 27
read-only 266 joins and 30
stored procedures and 266 discontinued trace commands
updatable 266 XML 119
drop index command
statistics and 310
dropping
D abstract plan groups 374
data modification update modes 27 indexes specified with index 235
data pages prefetching 237 plans 379, 384
datachange function duplication
statistics 295 update performance effect of 30
datatypes dynamic parameters, analyzing 122
join 13
deadlocks
concurrency optimization threshold settings 259
table scans and 259

388 Adaptive Server Enterprise

 Index

E H
elimination partition 197 Halloween problem
emit operator 52 cursors and 268
enable parallelism 135 hash join
engine query execution 20 operator 74
equi-join, transitive closure 7 hash union
exceptions, optimization goals 16 operator 91
exchange hash vector aggregate
operator 148 operator 86
worker process mode 151 hash-based table scan 157
exchange, pipemanagement 149 HashDistinctOp operator 83
execute cursors histograms
memory use of 266 join 12
executing steps, number of 306
query processing metrics 280
execution
preventing with set noexec on 37
exists check mode 327 I
expensive direct updates 29 I/O
exporting plan groups 385 direct updates and 27
expressions, join 14 prefetch keyword 236
range queries and 236
specifying size in queries 236
update operations and 29
F IDENTITY columns
factors, analyzed for optimization 5 cursors and 268
fetching cursors, memory and 266 importing abstract plan groups 385
finding abstract plans 377 index scan 159
fixed-length columns clustered 163
indexes and update modes 36 clustered, partitioned table 163
for update option, declare cursor covered using nonclustered global 162
optimizing and 276 global nonclustered 159
forceplan option, set 232 nonclustered, partitioned table 163
from table 54 noncovered, global nonclustered 159
function indexes
datachange , statistics 295 cursors using 267
large I/O for 236
search arguments 10
specifying for queries 234
G update index statistics on 305
goals, optimization 15 update modes and 35
goals, optimization exceptions 16 update operations and 28, 29
group sorted agg update statistics on 305
operator 85 in-place updates 27
group sorted agg operator 85 insert
GroupSorted (Distinct) operator 81 operator 62, 195

Performance and Tuning Series: Query Processing and Abstract Plans 389
Index

isolation levels query engine 21

cursors 273 query execution 25
query plans 21
lightweight procedures and dynamic SQL statements
122
J locking
Job Scheduler statistics 307
update statistics 297 log scan 59
join 12 LRU replacement strategy
both tables with useless partitioning 173 specifying 240
number of tables considered by optimizer 233
operator 69
outer 180
parallelism 169 M
parallelism, one table with useful partitioning 171 maintenance tasks
parallelism, replicated 175 forced indexes 235
parallelism, tables with same useful partitioning 169 forceplan checking 232
semi 180 maintenance, statistics 300
serial 178 max repartition degree
table order in 232 setting 137
update mode and 30 max resource granularity
updates using 28, 29, 30 setting 136
join memory
density 12 cursors and 264
expressions 14 merge join operator 71
histograms 12 merge union operator 91
mixed data types 13 messages, dropped index 235
or predicates 14 minor columns
ordering 14 update index statistics and 305
joins modifying abstract plans 380
updates using 29 MRU replacement strategy, disabling 241
jtc option, set 246

N
K names
keys, index index clause and 235
update operations on 28 index prefetch and 237
nary nested loop join operator 76
nested loop join 70
networks
L cursor activity of 272
large I/O nonequality, operators 11
index leaf pages 236 nonleading columns sort statistics 308
Lava null columns
operators 23 optimizing updates on 35

390 Adaptive Server Enterprise

 Index

number (quantity of) vector aggregation 84

cursor rows 277 operators, nonequality 11
tables considered by optimizer 233 optimization
additional paths 9
cursors 266
example search arguments 11
O limit time optimizing query 16
object sizes operators 4
tuning 19 predicate transformation 9
open command query transformation 6
memory and 266 techniques 4
operations optimization goals, exceptions 16
insert, delete, update 195 optimization problems 17
operator optimization, factors analyzed 5
delete operator 62 optimization, goals 15
exchange 148 optimizer 34–36
group sorted agg 85 overriding 221
group sorted agg 85 query 3
hash join 74 updates and 34
hash union 91 optimizer diagnostic utility 229–230
hash vector aggregate 86 configuring Adaptive Server 230
merge union 91 run sp_opt_querystats 230
nary nested loop join 76 sp_opt_querystats 229–??
remote scan 99 sp_opt_querystats , information provided 229
restrict 94 option
RID join 100 set rowcount 145
scan 52 or list 53
scroll 99 or predicates
sequencer 97 join 14
sort 94 OR strategy, cursors and 275
sqfilter 102 order, tables in a join 232
store 95 ordering, join 14
text delete 63 output
union all 90 statement 44
update operator 62 XML diagnostic 112
operator ,insert operator 62 overhead
operator, emit 52 cursors 272
operator, merge join 71 deferred updates 30
operators
GroupSorted (Distinct) 81
HashDistinctOp 83
Lava 23 P
optimization 4 pages, data
query plans 52 prefetch and 237
ScalarAggOp 93 parallel
SortOp (Distinct) 82 query execution model 148

Performance and Tuning Series: Query Processing and Abstract Plans 391
Index

query plans 146 pipe management, exchange 149

query processing 133 plan dump option, set 321
setting max degree 136 plan groups
setting max resource granularity 136 adding 373
table scan 156 copying 381
union all 167 copying to a table 385
parallel degree, setting max scan 138 creating 373
parallel processing dropping 374
query 134 dropping all plans in 384
parallelism 17 exporting 385
attribute-insensitive operation 155 information about 374
attribute-sensitive operation 169 overview of use 318
controlling at session level 139 plan association and 318
controlling for a query 141 plan capture and 318
distinct vector aggregation 184 reports 374
in-partitioned vector aggregation 180 plan load option, set 323
join 169 plan replace option, set 323
join, both tables with useless partitioning 173 plans
join, one table with useful partitioning 171 changing 380
join, replicated 175 comparing 379
join, tables with same useful partitioning 169 copying 378, 381
outer joins 180 deleting 384
query with IN list 184 dropping 379, 384
query with OR clause 186 finding 377
query with order by clause 188 modifying 380
reformatting 176 searching for 377
repartitioned vector aggregation 181 predicate transformation 9
semijoins 180 prefetch
serial join 178 data pages 237
serial vector aggregation 183 disabling 239
setting number of worker processes 135 enabling 239
SQL operatoions 154 queries 236
table scan 155 sp_cachestrategy 241
two-phased vector aggregation 182 prefetch keyword, I/O size and 236
vector aggregation 180 problems optimizing queries 17
parallelism, enable 135 process_limit_action 200
partial plans
specifying with create plan 317
partition
elimination 197 Q
skew 198 QP metrics. See query processing metrics
table scan 157 queries
performance execution settings 37
number of tables considered by optimizer 234 specifying I/O size 236
permissions specifying index for 234
XML 122 queries, problems optimizing 17

392 Adaptive Server Enterprise

 Index

query remote scan operator 99

Lava execution 25 replication, update operations and 28
limit optimizing time 16 reports
not run in parallel 199 cache strategy 241
optimizer 3 plan groups 374
OR clause 186 restrict operator 94
parallel execution model 148 results, differing parallel query 144
parallel processing 134 RID join operator 100
select-into clause 192 RID scan 57
set local variables 145 row counts statistics, inaccurate 311
with IN list 184 row ID (RID) update operations and 28
with order by clause 188 runtime
query analysis 34–36 adjustment 199
dynamic parameters 122 managing adjustment 200
showplan and 37 recognizing adjustment 200
sp_cachestrategy 241 reducing adjustments 201
query optimization 111
query plans 47
Lava 21
operators 52 S
parallel 146 sampling
suboptimal 235 statistics 293
updatable cursors and 275 use for updating statistics 293
query processing scalar aggregation
parallel 133 serial 166
query processing metrics two phased 165
accessing 280 ScalarAggOp operator 93
clearing 284 scan
executing 280 clustered index 163
sysquerymetrics view 282 clustered index on partitioned tables 163
using 282 index 159
query, execution engine 20 index global nonclustered 159
index noncovered of global nonclustered 159
index, covered use nonclustered global 162
local indexes 163
R nonclustered, partitioned table 163
range queries, large I/O for 236 operator 52
read-only cursors 266 scan types statistics 307
indexes and 267 scroll operator 99
locking and 272 search arguments
reduce update statistics impact 309 creating 18
referential integrity example of optimization 11
constraints 65 indexes 10
update operations and 28 transitive closure 7
updates using 30 search arguments, converted 6
reformatting parallelism 176 searching for abstract plans 377

Performance and Tuning Series: Query Processing and Abstract Plans 393
Index

select command SortOp (Distinct) operator 82

specifying index 234 sp_add_qpgroup system procedure 373
select-into query 192 sp_cachestrategy system procedure 241
sequencer sp_chgattribute system procedure
operator 97 concurrency_opt_threshold 259
serial sp_cmp_qplans system procedure 379
scalar aggregation 166 sp_copy_all_qplans system procedure 381
union all 168 sp_copy_qplan system procedure 378
serial table scan 155 sp_drop_all_qplans system procedure 384
set sp_drop_qpgroup system procedure 374
local variables 145 sp_drop_qplan system procedure 379
XML command 112 sp_export_qpgroup system procedure 385
set command sp_find_qplan system procedure 377
examples 140 sp_help_qpgroup system procedure 374
forceplan 232 sp_help_qplan system procedure 378
jtc 246 sp_import_qpgroup system procedure 385
plan dump 321 sp_opt_querystats
plan exists 327 configuring Adative Server 230
plan load 323 running 230
plan replace 323 system procedure 229
set plan dump command 322 truncating output 230
set plan exists check 327 sp_set_qplan system procedure 380
set plan load command 323 speed (server)
set plan replace command 323 cheap direct updates 28
set rowcount deferred index deletes 33
option 145 deferred updates 30
set theory operations direct updates 27
compared to row-oriented programming 262 expensive direct updates 29
setting in-place updates 27
mac parallel degree 136 updates 27
max repartition degree 137 sqfilter, operator 102
max resource granularity 136 SQL
max scan parallel degree 138 parallelism 154
number of worker processes 135 SQL derived tables 339
shared keyword, cursors and 267 SQL standards
shared locks cursors and 262
read-only cursors 267 SQL tables
showplan derived 19
statement level output 44 statement level output 44
using 37, 201 statistics
skew, partition 198 adding for unindexed columns 290
sort column-level 300, 302, 305
operator 94 creating column statistics 302
statistics, unindexed columns 308 datachange function 295
sort requirements deleting table and column with delete statistics
statistics 307 310

394 Adaptive Server Enterprise

 Index

drop index and 300 transitive closure

getting additional 303 equi-join 7
locking 307 search arguments 7
sampling 293 triggers
scan types 307 update mode and 34
sort requirements 307 update operations and 28
sorts for unindexed columns 308 truncate table command
truncate table and 301 column-level statistics and 301
update statistics 291 tuning
update statistics automatically 297 according to object size 19
updating 290, 302 advanced techniques for 221–259
using 287 range queries 235
using Job Scheduler 297 two phased scalar aggregation 165
statistics clause, create index command 301
statisticsmaintenance 300
statisticssorts, nonleading columns 308
steps U
deferred updates 30 unindexed columns 290
direct updates 27 union all
store operator 90
operator 95 parallel 167
stored procedures serial 168
cursors within 270 union operator
subqueries 189 cursors and 275
sysquerymetrics view unique indexes
query processing metrics 282 update modes and 35
update 195
update all statistics 302
update all statistics command 306
T update cursors 266
table count option, set 233 update index statistics 302, 305, 308
table scan update locks
forcing 234 cursors and 267
hash-based 157 update modes
parallel 156 cheap direct 28
parallelism 155 deferred 30
partition-based 157 deferred index 31
serial 155 direct 30
techniques, optimization 4 expensive direct 29
testing, index forcing 235 indexing and 35
text delete, operator 63 in-place 27
transaction logs, update operation and 27 joins and 30
transformation optimizing for 34
predicate 9 triggers and 34
transformations update operations 27
query optimization 6 update operator 62

Performance and Tuning Series: Query Processing and Abstract Plans 395
Index

update statistics 291

column-level 305
column-level statistics 305
managing statistics and 300
with consumers clause 309
updating statistics 290, 293, 302
use sampling 293
user IDs
changing with sp_import_qpgroup 385
user log cache, in ALS 243

V
variables, setting local 145
vector aggregation 180
distinct 184
in-partitioned 180
repartitioned 181
serial 183
two-phased 182
vector aggregation operators 84
view
sysquerymetrics, query processing metrics 282

W
with statistics clause, create index command 301
worker process mode
exchange 151
worker processes
setting number 135

X
XML
diagnostic output 112
discontinued trace commands 119
permissions 122
set 112

396 Adaptive Server Enterprise