Setting up the Oracle Optimizer for Proof of Concept Testing

Copyright © 2018, Oracle and/or its affiliates. All rights reserved. |
Safe Harbor Statement
The following is intended to outline our general product direction. It is intended for
information purposes only, and may not be incorporated into any contract. It is not a
commitment to deliver any material, code, or functionality, and should not be relied upon
in making purchasing decisions. The development, release, and timing of any features or
functionality described for Oracle’s products remains at the sole discretion of Oracle.
2

Setting up the Oracle Optimizer for PoCs
Covering Oracle Database 12c Release 1 and Release 2
and Oracle Database 18c
May 1st 2018
Nigel Bayliss
Optimizer Product Manager
@vldbb
http://blogs.oracle.com/optimizer

Format
• I will use this slide deck to cover the topic in general
• Please interrupt
– Interactive is best
– Use chat to message everyone - if there's any message I will slow down and wait for
you to type full question – Chris will keep an eye out for me
• Anything I can't answer will be noted and I will answer using
@vldbb, http://blogs.oracle.com/optimizer or later Office Hours
• The Oracle Optimizer might be smaller than you think
– If a question is out of scope, I will say so and leave it at that
4

General Principles for PoCs
• Start with a Keep It Simple Stupid (KISS) baseline
• Prioritize stability and consistency
• Exercise control over change
–Make changes for a reason
• Keep moving
5

Copyright © 2018, Oracle and/or its affiliates. All rights reserved. | 6
The Adaptive Optimizer
New in Oracle Database 12c Release 1
12c Optimizer
Adaptive
Statistics
Optimizer Adaptive Features
Change SQL execution plans at runtime
Learn from previous executions and
choose better SQL execution plans
Adaptive
Plans

From Oracle Database 12c
Release 1
Oracle Optimizer Adaptive Features
12c Optimizer
Column Usage
Statistics
SQL Plan
Directives
Dynamic
Statistics
(Adaptive)
G a t h e r S t a t i s t i c s
Column group
statistics created
automatically
G a t h e r S t a t i s t i c s
F O R A L L C O L U M N S
S I Z E A U T O
Histograms
Statistics feedback for
single table cardinality
and joins
Adaptive
Plans
Data
Dictionary

Adaptive Optimizer
• The optimizer adapts over time
– SQL execution plans may change over time
– Performance might not be consistent between runs while it adapts
• Adaptive statistics have an overhead in systems with high hard parse rates
– High hard parse rates are not best practice, but reality gets in the way
– More dynamic sampling queries than in Oracle Database 11g
– Some see library cache contention associated with longer parse times
– Some see 'RC latch' (result cache) contention in Oracle Database 12c Release 1 (fixed
in Oracle Database 12c Release 2)
8
Remember…

The Mechanics of Adaption
•Metadata is created as workloads are executed
•Metadata affects
–SQL execution plans
–Histograms
–Column group statistics
9

Viewing Optimizer Metadata
SELECT DBMS_STATS.REPORT_COL_USAGE('SH,'EMP')
FROM DUAL;
[Also SYS.COL_USAGE$]
SELECT * FROM DBA_SQL_PLAN_DIRECTIVES

Seeing the Affects of Optimizer Metadata
SELECT *
FROM DBA_STAT_EXTENSIONS
SELECT HISTOGRAM
FROM DBA_TAB_COLUMNS

New Histograms and Column Groups
•Column Groups
–When gathering statistics and there are SQL Plan Directives
–Note: this is no longer the default behavior (after 12.1)
•Histograms
–When gathering statistics FOR … COLUMNS SIZE AUTO and
there is column usage
–This is the default behavior
12

Adaptive Optimizer Settings
13

Finer control of adaptive features – new database parameters
From Oracle Database 12c Release 2
12c Optimizer
Adaptive
Statistics
Change plans at runtime Learn from previous executions
OPTIMIZER_ADAPTIVE_FEATURES Obsolete
OPTIMIZER_ADAPTIVE_PLANS OPTIMIZER_ADAPTIVE_STATISTICS
Adaptive
Plans

OPTIMIZER_ADAPTIVE_STATISTICS (FALSE)
15
New default behavior
From Oracle Database 12c Release 2
12c Optimizer
Adaptive
Statistics
Change plans at runtime Learn from previous executions
OPTIMIZER_ADAPTIVE_PLANS (TRUE)
Adaptive
Plans

Oracle Database 12c
Release 2 and 18c
Optimizer Adaptive Features – Recommended Defaults
Optimizer
Column Usage
Statistics
SQL Plan
Directives
Statistics feedback
for single table
cardinality ONLY
G a t h e r S t a t i s t i c sG a t h e r S t a t i s t i c s
S I Z E A U T O
Histograms
Dynamic
Statistics
Adaptive
Plans

The default adaptive feature settings
prioritize consistent performance
17

Recommendations for Oracle Database 12c Release 1
• Install latest Proactive Bundle Patch (>=October 2017)
– e.g. January 2018, Patch 27010930 - Database Proactive Bundle Patch 12.1.0.2.180116
• Recommendations for Adaptive Features in Oracle Database 12c Release 1
(Adaptive Features, Adaptive Statistics and 12c SQL Performance) (Doc ID 2312911.1)
– _fix_control='26664361:7','16732417:1','20243268:1'
• Remove optimizer_adaptive_features parameter from pfile/spfile
• Set the following optimizer parameters (these are the defaults in 12cR2 and 18c):
– optimizer_adaptive_plans=TRUE
– optimizer_adaptive_statistics=FALSE
• On the database, execute the following ('OFF' is the default)
– EXEC DBMS_STATS.SET_GLOBAL_PREFS('AUTO_STAT_EXTENSIONS','OFF')
18
Assuming 12.1.0.2

Recommendations for Oracle Database 12c Release 2
• Do not set the following optimizer parameters - they will take their default
values
– optimizer_adaptive_plans [default TRUE]
– optimizer_adaptive_statistics [default FALSE]
• On the database, execute the following (OFF is the default)
19

Recommendations for Oracle Database 18c
• Do not set the following optimizer parameters - they will take their default
values
– optimizer_adaptive_plans [default TRUE]
– optimizer_adaptive_statistics [default FALSE]
• On the database, execute the following (OFF is the default)
20

Oracle Database 12c
Release 2 and 18c
Optimizer Adaptive Features – Recommended Defaults
Optimizer
Column Usage
Statistics
SQL Plan
Directives
Statistics feedback
for single table
cardinality ONLY
G a t h e r S t a t i s t i c sG a t h e r S t a t i s t i c s
S I Z E A U T O
Histograms
Dynamic
Statistics
Adaptive
Plans
DBMS_STATS preference:
AUTO_STAT_EXTENSIONS
Default is OFF

Using Adaptive Statistics?
• Adaptive statistics can be useful for DSS/BI workloads
– Ad-hoc queries
– Longer-running queries
– Low hard parse rates
• Start with defaults (adaptive statistics disabled) and you may want to test benefit
• Multiple runs may be required to achieve best performance
– SQL plan directives created over time
– SQL plan directives are created even if optimizer_adaptive_statistics=FALSE – they are just not used
– Gathering statistics is required to generate column group stats (and potentially new histograms)
• Setting optimizer_adaptive_statistics = FALSE can be used to restore behavior once test
is complete – and remember you may have new histograms/column group stats
22

Other Recommendations for PoCs
23

Other Recommendations for Oracle Database 12c and 18c
•Gather statistics for you application schemas
•Gather dictionary and fixed object statistics
–DBMS_STATS.GATHER_DICTIONARY_STATS
–DBMS_STATS.GATHER_FIXED_OBJECTS_STATS
24
Optimizer Statistics

• If you do not gather system statistics, then delete them and
restart the database
• If you have your own preferred settings, you may want to use
them, but it is a good opportunity to use defaults instead
• Prior to baseline runs, delete system statistics (as above) – the default settings are
appropriate for most systems
• You might want to include a specific test to evaluate the affect of change
• If you find significant differences with vs without, take the time to
find out why
25
System Statistics

• SQL plan management evolution is automated
– If you are using SQL plan management (SPM), alternative plans may be auto evolved
and accepted
– You might want to disable the auto tuning job or prevent new plans from being
accepted…
• exec DBMS_SPM.SET_EVOLVE_TASK_PARAMETER(
task_name => 'SYS_AUTO_SPM_EVOLVE_TASK'
,parameter => 'ACCEPT_PLANS'
,value => 'FALSE')
• exec DBMS_AUTO_TASK_ADMIN.DISABLE (client_name => 'sql tuning advisor')
26
SQL Plan Management

• STALE stats can be more common than usual in a PoC (e.g. changes in column usage)
• You might acquire new histograms (METHOD_OPT=>'FOR ALL COLUMNS SIZE AUTO')
• Controlling stats gathering
– As a minimum, use the auto stats job to maintain Oracle-owned objects:
DBMS_STATS.SET_GLOBAL_PREFS(pname=>'autostats_target',pvalue=>'oracle')
– If necessary, auto stats can be disabled as follows:
DBMS_AUTO_TASK_ADMIN.DISABLE(client_name=>'auto optimizer stats collection')
– Manually gather stats:
exec dbms_stats.gather_database_stats
– Auto stats can be initiated manually:
exec dbms_stats.gather_database_stats_job_proc
27
Auto statistics gathering

• GTT statistics are session-local by default
– Pre-12c workload may assume and depend on statistics being shared across sessions
– If you need pre-12c behaviour in the PoC then…
– exec dbms_stats.set_global_prefs('GLOBAL_TEMP_TABLE_STATS','SHARED')
28
Global Temporary Tables

Regathering Statistics
• You may need to regather statistics at some point
– E.g. you might want new histograms to match your workload
• If you experience plan regressions, remember that previous stats can be
restored to help isolate issue
– E.g. – DBMS_STATS RESTORE_SCHEMA_STATS, RESTORE_TABLE_STATS etc
• You can test new statistics using pending stats and publish them when
tested
– DBMS_STATS PUBLISH preference
– Parameter: optimizer_use_pending_statistics
– DBMS_STATS.PUBLISH_PENDING_STATS
29

More General Recommendations
30

Think Holistically
• Using different features and settings may improve performance in some
areas and degrade performance elsewhere
– Consider the overall picture
– Strike a balance and keep things simple
• Avoid micro managing individual queries and consider the overall
benefit/cost
– Use Test Case Builder and/or SQLT to capture sub-optimal plans and move on!
• Use the service level agreement to guide what is important and what is not
– OK, OK, an SLA never exists 
– Is a query exhibiting a performance regression key to the business?
31

Think Holistically
• Optimizers are not perfect – they use estimates
– Avoid being fixated by the need to chase apparent 'bugs'
– Work around query regressions
• SQL tuning advisor will help find better plans if you have SQL statements
with performance regressions
• Use plan control and stability features [a later Office Hours topic]
• SQL plan management (SPM) can control SQL execution plans
– Remember you can even test with optimizer_features_enable=<existing/earlier
release> and capture the plans with SPM – then run with OFA=<latest release>
– You can use SPM fix individual queries
32

Think Holistically
• Over-hinted SQL?
– _optimizer_ignore_hints=TRUE
– SQL Patch [see later]
33

Summary
• Start with a KISS baseline
– Latest BP + Defaults
– Gather statistics at the start of the PoC to gain a solid initial baseline
– Remember you can restore statistics and use pending statistics if necessary
• Prioritize stability and consistency
– Use the most recent optimizer adaptive feature defaults
– Remember: the optimizer adapts over time if you use non-defaults
• Exercise control over change
– Be aware of auto jobs and disable where appropriate
• Keep Moving
– Work around problems, save for later
– Plan stability features (e.g. SPM, SQL Patch), SQL Tuning Advisor, TCB, SQLT
34

More
• http://blogs.oracle.com/optimizer
• SPM
https://blogs.oracle.com/optimizer/using-sql-plan-management-to-control-sql-execution-plans
35

More
• Test Case Builder to capture a test case for later diagnosis
https://docs.oracle.com/en/database/oracle/oracle-database/18/tgsql/sql-test-case-builder.html
36

More
• SQL Patch to patch individual SQL statements
https://blogs.oracle.com/optimizer/adding-and-disabling-hints-using-sql-patch
37

More
• SQLT to capture query information for later diagnosis
https://blogs.oracle.com/performancediagnosis/sqlt-version-122171004
38

Safe Harbor Statement
The preceding is intended to outline our general product direction. It is intended for
information purposes only, and may not be incorporated into any contract. It is not a
commitment to deliver any material, code, or functionality, and should not be relied upon
in making purchasing decisions. The development, release, and timing of any features or
functionality described for Oracle’s products remains at the sole discretion of Oracle.
39

Setting up the Oracle Optimizer for Proof of Concept Testing

More Related Content

What's hot (20)

Similar to Setting up the Oracle Optimizer for Proof of Concept Testing (20)

Recently uploaded (20)

Setting up the Oracle Optimizer for Proof of Concept Testing