DOC1586374_A

An Oracle White Paper September 2013

Best Practices for Gathering Statistics with Oracle E-Business Suite


Disclaimer

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.


Executive Overview ......................................................................................... 1 Introduction ..................................................................................................... 1 FND_STATS Features .................................................................................... 2 History Mode ................................................................................................... 2 Gather Options ................................................................................................ 2 Histograms ...................................................................................................... 3 The AUTO Sampling Option ........................................................................... 4 Custom Scripts ................................................................................................ 4 When to Gather Statistics ............................................................................... 5 Cursor Invalidation .......................................................................................... 6 Locking Statistics ............................................................................................ 6 Database 11g Extended Statistics .................................................................. 7 Gathering Dictionary and Fixed Object Statistics ........................................... 8

Dictionary Statistics .................................................................................... 8 Fixed Statistics ............................................................................................ 8

Global Temporary Tables................................................................................ 9 Temporary Tables ......................................................................................... 10 Incremental Statistics for Partitioned Tables................................................. 10 Improving the Collection Time ...................................................................... 11 Verifying Statistics ......................................................................................... 11 Standalone Patches ...................................................................................... 12 Performance Test Cases & Examples .......................................................... 13

1. Oracle 11g Extended Optimizer Statistics ........................................ 13 2. Incremental Statistics Gathering ....................................................... 17 3. Concurrent Statistics Gathering ........................................................ 20

Appendix A: Related Documentation ............................................................ 21 Whitepapers .............................................................................................. 21 My Oracle Support Knowledge Documents ............................................. 21 Blogs ......................................................................................................... 21

Appendix B: Package Specifications ............................................................ 23 Gather Table Statistics ............................................................................. 23 Gather Schema Statistics ......................................................................... 25


Executive Overview

This paper documents the best practices for gathering statistics. It focuses on new features implemented in the FND_STATS package that can be applied to Oracle E-Business Suite 11i (11.5.10 and later) and Release 12.

Introduction

The Best Practices for Gathering Optimizer Statistics white paper provides an extensive overview of the Oracle database statistics gathering process and is a useful precursor to this paper. It describes several methods for collecting or setting statistics, however, the only supported methods with Oracle E-Business Suite are either the Gather Statistics concurrent program or the FND_STATS package (which is called by the Gather Statistics concurrent program). FND_STATS is a wrapper around the DBMS_STATS package and provides several PL/SQL procedures for gathering statistics at the table, schema, or database level. The FND_STATS package also supports histograms, table exclusions and the new database 11G features such as extended stats, incremental stats gathering for partitioned tables and concurrent stats gathering. Neither the DBMS_STATS package nor the now obsolete ANALYZE command recognize those features and may result in suboptimal execution plans.

1


FND_STATS Features

This section provides a high-level view of how the FND_STATS package interacts with DBMS_STATS. While either the Oracle E-Business Suite or DBMS_STATS will gather statistics for objects with stale or empty (missing) statistics, the FND_STATS package implements the following features in Oracle E-Business Suite:

• Supports collection of statistics for individual schemas, or for the entire Oracle E-Business Suite. To process all E-Business schemas, pass the schema name as 'ALL'. DBMS_STATS does not have this functionality.

• Histograms on columns designated in FND_HISTOGRAM_COLS. This contains a list of histogram columns that Oracle Development have identified as essential.

• The new GATHER_AUTO option combines the collection of statistics for tables with either stale or missing (empty) statistics.

• FND_STATS supports rerun-ability feature. If the Gather Schema Statistics concurrent program is used, FND_STATS maintains the history of each submission. If a run stopped or interrupted, the next submission will resume from the point of failure or termination. Starting with the 10g version of the database, DBMS_STATS 'GATHER AUTO' option with Table Monitoring feature can be used to serve the purpose of rerun-ability.

• FND_STATS_HIST stores historical information about the amount of time it takes the statistics gathering package FND_STATS to gather statistics.

History Mode

The Gather Table Statistics program can optionally backup the existing statistics prior to gathering new statistics. If the value of backup flag parameter is BACKUP, FND_STATS exports the old statistics using DBMS_STATS.EXPORT_TABLE_STATS prior to gathering new statistics and stores them in FND_STATTAB.

The history mode parameter controls the amount of history records that are created:

• Last Run - History records are maintained only for the last gather statistics run. Each subsequent run will overwrite the previous history record for the object. This is the default behavior.

• Full - This mode does not overwrite previous historical information. History records are created for each new run and are identified by the Request ID, which is generated automatically if one is not provided. If this mode is used, the Purge FND_STATS History Records concurrent program should be run periodically to purge the FND_STATS_HIST table.

• None - This mode does not generate any history information and the run cannot be restarted.

If problems ensue post gathering statistics, you can use the RESTORE_SCHEMA_STATS or RESTORE_TABLE_STATS procedure to restore statistics that were previously backed up in the FND_STATTAB table, into the dictionary; statid can be provided to distinguish between different sets of statistics.

Gather Options

Use GATHER_AUTO option to gather stats incrementally. GATHER_AUTO relies on the Oracle Table Monitoring feature to gather information about new or changed objects statistics.

2


In Oracle 9i, it was possible to specify individual tables, but this approach was deprecated in later releases and is now automatically enabled for the entire schemas by default in Oracle 10g and Oracle 11g. To mimic this behavior in Oracle 9i, use the following procedure: exec FND_STATS.ENABLE_SCHEMA_MONITORING (‘ALL’); or to enable monitoring for a single schema: exec FND_STATS.ENABLE_SCHEMA_MONITORING (SCHEMA_NAME); When Gather Options parameter is set to "GATHER AUTO", fnd_stats will only gather statistics for tables that have changed by more than the modification threshold % since the table was last analyzed (statistics gathered). The modifications threshold can be adjusted by the user by passing a value for modpercent, which by default is equal to 10. So the GATHER_AUTO will only gather stats on tables that show up in DBA_TAB_MODIFICATIONS with > 10% changes. FND_STATS.gather_schema_statistics (. . ..,options=>’GATHER AUTO’);

Histograms

A histogram is a special type of column statistic that provides additional information about the frequency and distribution of values within a column. This information is used by the Optimizer when, for example, deciding to use an index or perform a full-table scan, or deciding the join order. Without histograms the Optimizer assumes a uniform distribution of rows across the distinct values in the column. These are created in Oracle E-Business Suite when a value in a column that is referenced in a where clause has a disproportionate number of values that would usually make a full-table scan cheaper than using the index - in other words when a column has non-unique repeating keys and only a few distinct values. This is referred to as data skew. Histograms may also be used for some join predicates, which is beneficial with cardinality estimates. Oracle 11g uses adaptive cursor sharing to ensure that the correct plan is used with bind peeking.

Collecting statistics for histograms adds time to the overall process. Histograms can be problematic and in Oracle 10g there were concerns with bind peeking and how it affects cardinality estimates. The Optimizer would peek at the value of the bind variables on the first hard parse of the SQL statement and then always use that plan for all future executions. The only real solution for Oracle E-Business Suite was to disable bind peeking; dropping the histograms was not an option.

This has been improved in Oracle 11g with the introduction of Adaptive Cursor Sharing. Avoiding histograms will ensure that SQL execution plans will not change depending on the bind value.

Those with custom development can use FND_STATS.CHECK_HISTOGRAM_COLS to determine the viability of proposed histograms. This procedure checks if the leading columns in non-unique indexes have a single value occupancy >=1/75th or more of the sample. The recommendation is that there is a minimum of 3000 rows. The following example shows how to run the procedure:

Set Serveroutput on

EXEC FND_STATS.CHECK_HISTOGRAM_COLS( tablelist => 'owner.table_name1 , owner.table_name2, ....’ )

Once you have checked the design and performance use the LOAD_HISTOGRAM_COLS procedure to seed information in the FND_HISTOGRAM_COLS table, which will ensure that they are updated when Gathering Statistics using the following syntax:

begin

FND_STATS.LOAD_HISTOGRAM_COLS

(action=>'INSERT', appl_id=>&custom_application_id, tabname=>&table_name, colname=>&column_name);

FND_STATS.GATHER_TABLE_STATS( ownname=>&owner_name, tabname=>&table_name);

end;

3


/

The AUTO Sampling Option

Traditionally in Oracle E-Business Suite most statistics were gathered at the default sample size of 10%, which was provided to the ESTIMATE_PERCENT parameter. Experience shows that some of the products subject to data skew benefited from gathering at a higher percentage. This led to a trade-off. A larger sample size would deliver more accurate results, but would take longer to collect; whereas a smaller sample size would collect more quickly, but may not collect statistics that reflect the true distribution of data and might result in sub-optimal runtime performance (especially in the case of data skew).

In Oracle 10g DBMS_STATS, the default value for the ESTIMATE_PERCENT parameter changed to DBMS_STATS.AUTO_SAMPLE_SIZE. Although this allowed Oracle to determine the appropriate sample size, there were some issues with inadequate samples when tables had exceptional data skew. Oracle 11g now uses hash-based sampling that has accuracy close to 100% but executes more quickly than when collecting at 10% using the old method.

The new version of FND_STATS now supports the 'AUTO' option for the ESTIMATE_PERCENT parameter, which in turn uses the DBMS_STATS.AUTO_SAMPLE_SIZE.

Note: Using AUTO is highly recommended as specifying a numeric percentage will use the old statistics gathering algorithm, even if you specify 100%.

Further advantages of using the AUTO sample size instead of a specific percentage include the following:

• It adapts to changes in data volumes, such as system growth or purging, whereas a fixed sampling percentage that was suitable at some point in time will inevitably require regular tests and reviews.

• Historically, a 10% sample was sufficient for most objects, but some products/tables benefited from higher sampling percentage of 30% - 40%, usually due to data skew. Having to determine which products benefitted, or which sample size provided the best benefit in least time, should no longer be necessary.

Note: The AUTO sampling feature with Oracle E-Business Suite only operates as described with Oracle 11g. When used with previous database versions it will use 10% for the estimate_percent parameter.

Note: When invoking the Gather Statistics concurrent program, leave the estimate percentage parameter blank. The program will automatically choose the default value for estimate_percent parameter (depends on the database version). If a value is provided then statistics will be gathered at the specified percentage. If the database version is 11g or higher, the default value for this parameter will be dbms_stats.auto_sample_size, whereas it will be set to 10% for previous releases. Note: AUTO Sampling Statistics Gathering Feature of Oracle Database 11g is implemented with the latest code of FND_STATS. Refer to “Standalone Patches” section, the standalone patches for Oracle 11g are shown in Table 1.

Custom Scripts

As an alternative to using the Gather Statistics concurrent program, custom scripts may be developed that call the FND_STATS package directly. The syntax is as follows:

For schema stats: Exec FND_STATS.gather_schema_statistics('schema_name');

For example:

4


EXEC FND_STATS.gather_schema_statistics('ALL');

EXEC FND_STATS.gather_schema_statistics('XLA');

For table stats: Exec FND_STATS.gather_table_stats(‘owner’,’table_name’);

For example:

EXEC FND_STATS.gather_table_stats('AR','RA_CUSTOMER_TRX_ALL');

When to Gather Statistics

Some Oracle E-Business Suite customers gather statistics too frequently unnecessarily wasting system resources. Having representative statistics, doesn’t mean frequently running Gather Statistics, such as daily. Instead, devise a strategy to ensure that the collection frequency does not cause excessive SQL execution plan instability and that the statistics do not reach a level of staleness that impacts performance. Exceptions include, for example, gathering statistics for a specific set of objects immediately after significant data imports. This is typically automated within Oracle E-Business Suite interface programs.

Oracle E-Business Suite customers are advised to determine the frequency of statistics gathering based on substantive changes to data or data skew, in other words, on volume rather than time. It is highly recommended that statistics are gathered bi-weekly, monthly, or an even longer interval depending on your data.

Consider an example where 1 million records are added per month to a table containing 10M rows. This corresponds to the 10% threshold used when gathering stale statistics and therefore statistics would be collected approximately once each month. Variations in volume that trigger the 10% stale threshold may result in more or less frequent updates to the statistics collection process.

The collection interval may also be determined by monitoring transaction throughput. Create a repeatable baseline of critical business processes and transactions. Review the baseline regularly, such as weekly or monthly, until key transaction performance starts to change. Map the business cycle and substantial data changes including, for example, posting, month end processing, data import, or purging. Use this information to help determine the collection frequency.

Note: A baseline of critical business processes and transactions provide a datum and is exceptionally useful for performance monitoring when patching or updating the system.

When time constrained and the time to gather statistics exceeds the collection window, break the collection process across each time slot by schema, or even individual tables. With the later approach, static and reference tables, which tend to be relatively small, are unlikely to become stale and so it is more important to focus on the more volatile high-transaction tables.

Note: During collection Oracle internally prioritizes the database objects that require statistics.

Administrators who are trying to determine an appropriate strategy may find it useful to monitor the STALE_STATS column in USER_TAB_STATISTICS. However, this information is only updated daily, so you also need business knowledge of the major activities, such as purging or data import or you should additionally refer to information from USER_TAB_MODIFICATIONS, which lists the number of inserts, updates and deletes.

Overall, gathering statistics on a particular day, date, or specific time during the business cycle enables administrators to correlate the gathering statistics event with any unexpected behavior. You should not excessively gather statistics on all schemas such as nightly or weekly.

5


Cursor Invalidation

When an object’s statistics are refreshed, all references to that object in the shared pool or library cache are invalidated and all dependent cursors referencing that object are also invalidated, each will require a new hard parse. Oracle distributes invalidation timestamps for dependent objects needing invalidation, thereby reducing the potentially large number of simultaneous hard parses, but this can still result in an increase in library cache waits especially when collecting statistics across a large number of objects.

Note: When using the Gather Statistics concurrent program, the Invalidate Dependent Cursors parameter is set to “Yes” by default, which is the standard best practice. New execution plans will be generated whenever cursor has been flushed out and reloaded.

While this approach suits most Oracle E-Business Suite installations, this parameter allows you to specify that cursors should not be invalidated. Although this may reduce the CPU utilization, this means that existing plans, which may be sub-optimal, will continue to be used until the dependent cursors are aged out.

Deciding between these two invalidation options is determined by the collection frequency. If you keep your statistics fairly current, you will probably not need to invalidate cursors as the plans from the newly collected statistic are unlikely to be significantly different. In other words, performance should remain consistent and stable and over time. However, if you don’t collect statistics frequently enough, or if you start to experience performance degradation due to stale statistics, then setting this to the default to invalidate the cursors that are dependent on the targeted objects as soon as possible will ensure that the plans are refreshed sooner.

Note: Gather statistics only when there is little system activity or during a defined batch period.

Locking Statistics

Locking prevents object statistics from being overwritten when the collection processes is run. This may be useful when tables are empty when the collection process is run, or when you want to exclude exceptionally large tables while gathering statistics for the remainder of the schema. Consider locking statistics for the following: • Very large static tables that have representative statistics already in place and are rarely updated.

• Volatile tables where data changes dramatically over time. Good examples include intermediate tables used during

batch processing and some interface tables, which are deployed extensively across Oracle E-business Suite. Example application tables include AP_SELECTED_INVOICES and WSH_PR_WORKERS.

• Temporary or interim tables that only contain data at runtime but are typically empty when the collection process runs. In this specific case, not locking statistics results in having zero or un-representative statistics. Note that these do not include Global Temporary tables, which are excluded from FND_STATS Gather Schema Statistics anyway. Note : If a table has zero statistics (num_rows = 0) then the CBO uses these statistics to determine the execution plan and if the table has a lot of rows at time of execution then this execution plan could be very inefficient. If tables have empty statistics (null) then dynamic sampling will be used, which is preferable to using the CBO with incorrect or zero statistics.

There are two distinct methods for protecting table statistics from modification. 1. You can physically lock the statistics using DBMS_STATS.LOCK_TABLE_STATS, which prevents any calls to

FND_STATS or DBMS_STATS from modifying the statistics.

6


2. You can also add the table to a list of excluded tables by using FND_STATS.LOAD_XCLUD_TAB. This prevents the table

from having its statistics modified when using Gather Schema Statistics, but the Gather Table Statistics does not honor the FND exclusion, only the DBMS lock. This is by design; there may be instances when you want to manually collect statitistics.

Oracle E-Business Suite development teams use LOAD_XCLUD_TAB to seed exceptions into the FND_EXCLUDE_TABLE_STATS table (which is a metadata container). Many of the EBS volatile and temporary tables are seeded in this table for reference only. They will not be excluded from FND_STATS Gather Schema Statistics, as they have been seeded with negative application_ids. You should consider these tables as candidates for locking. Note: It is recommended that you use seed tables into FND_EXCLUDE_TABLE_STATS, or use DBMS_STATS.LOCK_TABLE_STATS if certain that you never want the statistics modified. For each volatile or temporary table that you have locked you should initially have empty (null) statistics and rely on dynamic sampling. Delete the statistics if they already contain values. Only if you have performance issues should you consider gathering statistics when the table has a representative load or populating using FND_STATS. SET_TABLE_STATS. Once the statistics are locked using DBMS_STATS.LOCK_TABLE_STATS, the Gather Schema Statistics Concurrent Program will display a message in the request log file explicitly stating that statistics are locked on that table. Use the following to seed a table in FND_EXCLUDE_TABLE_STATS:

FND_STATS.LOAD_XCLUD_TAB(action=>'INSERT', appl_id=>&application_id, tabname=>&table_name) Example

FND_STATS.LOAD_XCLUD_TAB(action=>'INSERT', appl_id=>222,tabname=>'RA_CUSTOMER_TRX_ALL)

Use the following to delete the table entry from the fnd_exclude_table_stats:

FND_STATS.LOAD_XCLUD_TAB(action=>'DELETE', appl_id=>&application_id, tabname=>&table_name) Example

FND_STATS.LOAD_XCLUD_TAB(action=>'DELETE', appl_id=>222, tabname=>'RA_CUSTOMER_TRX_ALL')

Database 11g Extended Statistics

The Oracle Optimizer collects statistics on separate columns but does not have any knowledge about the relationship or correlation between the data stored across separate columns in the same table.

Creating extended statistics on a group of columns increases the accuracy of the combined selectivity of the predicates and hence the cardinality estimate, when the columns are used together in the where clause of a SQL statement. These are typically used when the cardinality estimates on a group of columns is inaccurate.

The column groups are defined using DBMS_STATS.CREATE_EXTENDED_STATS. Once created, Oracle will automatically maintain the statistics on that column group when statistics are gathered on the table.

Note: The new FND_STATS, as provided by Oracle E-Business Suite standalone patches now supports Oracle 11g extended statistics. Refer to “Standalone Patches” section, the standalone patches for Oracle 11g are shown in Table 1.

In Oracle E-Business Suite, FND_STATS.LOAD_EXTNSTATS_COLS is used to seed the multi-column groups. Statistics are automatically maintained by Gather Schema/Table Statistics concurrent programs or when using FND_STATS. The following examples provide the syntax for this procedure:

7


begin

FND_STATS.LOAD_EXTNSTATS_COLS

(action =>'INSERT', appl_id=>&custom_application_id, tabname => &table_name, owner=>&owner, colname1

=>&column_name1, colname2 =>&column_name2, colname3.....);

FND_STATS.GATHER_TABLE_STATS( ownname=>&owner_name, tabname=>&table_name);

end;

/ Note: You need to gather statistics after defining the group of columns.

Gathering Dictionary and Fixed Object Statistics

For the Cost Based Optimizer to work effectively for internal (recursive) SQLs, or application SQLs that use data dictionary views or fixed objects, the statistics need to be gathered for all the dictionary tables (owned by ‘SYS’, ‘SYSTEM’ and ‘SYSAUX’ etc) and the x$ tables used by the dynamic v$ performance views.

FND_STATS does not gather statistics for dictionary or fixed objects and instead you need to use DBMS_STATS in 10g and above. There are no Oracle E-Business Suite specific guidelines on the collection frequency for the dictionary and fixed object statistics.

Dictionary Statistics

Dictionary statistics only need to be gathered when there have been significant structural changes to the database.

e.g.

• After any associated platform or DB upgrade that is part of the overall EBS upgrade.

• After the R12 upgrade.

• After move to OATM

The question is when, and how often do they need to be collected. Typical examples include upgrades, new modules, or the application of large patches where several objects might be created or rebuilt using Data Definition Language (DDL) operations.

Statistics will need to be collected when SQL on the data dictionary objects appears as a high load (usually high elapsed time) in AWR reports or other monitoring tools, or when times in trace/tkprof files are very high for recursive SQL on the data dictionary objects. The following command can be used to gather statistics for all system schemas:

EXECUTE DBMS_STATS.GATHER_DICTIONARY_STATS( estimate_percent => DBMS_STATS.AUTO_SAMPLE_SIZE options

=> 'GATHER AUTO' ); -- In 10G and above.

Fixed Object Statistics

Fixed objects include the x$ tables and their indexes, which typically contain information about the instance or memory structures. Unlike dictionary tables that reside on disk, such as SYS.DBA_DATA_FILES, most of the v$ views, such as V$SQL and V$SQL_PLAN are based on the x$ tables, which are in-memory structures. As such, fixed tables have no IO cost associated with them and therefore when generating an execution plan the CBO only considers the CPU cost.

The automatic statistics gathering job does not gather fixed object statistics and in this case, dynamic sampling is not automatically used for SQL statements involving x$ tables when statistics are missing. Instead, the Optimizer uses predefined default values; however, these defaults may not be representative and could potentially lead to suboptimal execution plans.

8


How and When to Gather Fixed Object Statistics

Use DBMS_STATS.GATHER_FIXED_OBJECT_STATS (available in Oracle 10g and later) to gather fixed object statistics. For optimal performance, and due to the transient nature of the x$ tables, it is important to gather statistics when there is a representative workload on the system. The collection process tends to be resource intensive and if there is insufficient capacity during peak load, choose a time after the database has been running for some time and the v$ views have a significant amount of data. However, gathering fixed objects statistics under heavy load can result in contention issues thereby leading to performance degradation or hangs. Refer to MyOracle Knowledge Document “Fixed Objects Statistics (GATHER_FIXED_OBJECTS_STATS) Considerations (Doc ID 798257.1)” for further information.

The following command can be used to gather fixed object statistics:

EXECUTE DBMS_STATS.GATHER_FIXED_OBJECTS_STATS;

Planning Collection

Fixed object statistics should be gathered when there are significant changes to the database configuration or when accessing the v$ views is of concern. Oracle E-Business Suite examples include the following:

• Major database or application upgrades*

• Platform upgrades (especially Exadata)

• New application modules are deployed

• Performance issues when querying V$ views

• Significant changes in the workload or number of sessions

• After significant database structure changes (e.g. SGA/PGA)

For example, all the x$ tables that contain information about the buffer cache and shared pool, such as x$ tables used in v$buffer_pool or v$shared_pool_advice, may change significantly when the SGA is increased.

Note: In Oracle E-Business Suite it is imperative that you disable the Database 11g automatic DBMS job to gather statistics, but ensure that dictionary statistics are still gathered.

* When upgrading Oracle E-Business Suite, the $APPL_TOP/admin/adstats.sql script disables the Oracle DBMS job that automatically gathers statistics. It then gathers dictionary and fixed object statistics. However, the Database is run in restricted mode during the upgrade. Therefore, it is highly recommended to rerun DBMS_STATS.GATHER_FIXED_OBJECT_STATS post upgrade, when there is a representative load on the system.

Global Temporary Tables

Several Oracle E-Business Suite product teams create and use global temporary tables (GTTs) within a process; they are subsequently dropped when the process completes. These tables are completely dynamic in nature. The standard performance recommendation is not to gather statistics on global temporary tables.

The following approaches can be used to ensure good execution plan:

• Rely on dynamic sampling, which will occur in the absence of statistics. You need to delete any stats using DBMS_STATS.delete_table_stats on Global Temporary Tables as dynamic sampling only works if statistics are empty.

• However, dynamic sampling statistics may not result in an optimal execution plan and often queries will also need to include Optimizer hints to force a particular execution plan.

9


• Note that FND_STATS Gather Schema Statistics excludes Global Temporary Tables anyway, so they do not need to be excluded or locked.

Temporary Tables

Temporary tables in Oracle E-Business Suite are regular tables that should not be confused with GTTs. They tend to be highly volatile and are typically empty for some of the processing cycle and then re-populated. Examples include interface tables, intermediate transient tables, transactions that have been selected by a specific process such as payments or records to be purged.

For these tables you should initially have empty (null) statistics and rely on dynamic sampling. Delete the statistics if they already contain values.

Only if you have performance issues should you consider gathering statistics when the table has a representative load or populating using FND_STATS. SET_TABLE_STATS.

In all cases the statistics should be locked (so that they are not gathered).

Incremental Statistics for Partitioned Tables

Incremental statistics gathering is a new Oracle 11gR2 DBMS_STATS feature and is fully supported by FND_STATS. Oracle derives global statistics by scanning only new or modified partitions rather than the whole table, which is highly beneficial with high volume Oracle E-Business Suite tables.

Incremental global statistics works by collecting statistics and storing a synopsis (statistical metadata) for each partition, which consists of statistical metadata and the columns in each partition. This synopsis is typically only a few KB and is stored in the SYSAUX tablespace. Global statistics are created not by reading the entire table, but by aggregating the synopsis from each partition. Once a synopsis has been created for a new partition, all the synopses will be aggregated to update the table level statistics.

The preconditions that need to be met for this approach include the following:

• The INCREMENTAL preference for the partition table is set to TRUE

• DBMS_STATS.GATHER_*_STATS parameter GRANULARITY includes “ALL” or “AUTO”.

• ESTIMATE_PERCENT is set to AUTO_SAMPLE_SIZE (default in Oracle 11g)

Incremental statistics in conjunction with the new Oracle 11g DBMS_STATS.AUTO_SAMPLE_SIZE yield a significant reduction in the time to collect highly accurate statistics.

Oracle E-Business Suite

Incremental statistics can be applied to any large partitioned tables. Once the table preference has been set, Oracle E-Business Suite Gather Statistics (table or schema) can be used to automatically gather the statistics and create the synopsis.

Incremental Preference

Use the following command to check if the incremental preference is set for the XLA_AE_LINES table (which is part of the Oracle E-Business Suite12 Subledger Accounting):

select dbms_stats.get_prefs('INCREMENTAL', 'XLA','XLA_AE_LINES')

from dual;

10


If it returns TRUE then the preference is set.

The preference can be set using the following command:

exec dbms_stats.set_table_prefs('XLA','XLA_AE_LINES','INCREMENTAL','TRUE');

During testing, it may be useful to know that the preference can be turned off using:

exec dbms_stats.set_table_prefs('XLA', 'XLA_AE_LINES, 'INCREMENTAL', 'FALSE');

Granularity Parameter

When running Gather Statistics at the schema or table level:

• Set the granularity parameter to “ALL” or “AUTO”. Both of these are effectively the same and will invoke all the granularity levels.

• Set the Estimate Percent parameter to “DBMS_STATS.AUTO_SAMPLE_SIZE”, which is the default with Oracle 11g.

If you use custom scripts with FND_STATS, set the granularity to “ALL” or “AUTO”. For example: exec fnd_stats.gather_table_stats ('OWNER', ’TABLE_NAME', granularity =>'ALL');

Improving the Collection Time

There are two ways to speed up the gather statistics collection process.

Degree Of Parallelism

FND_STATS’s DEGREE (Degree of parallelism) helps speed up statistics gathering for large objects. If a degree is not provided, FND_STATS automatically determines the Degree of Parallelism, it defaults to the minimum of parallel_max_servers and cpu_count. This approach has been defined as intra object parallelism.

Concurrent Statistics Gathering

Concurrent statistics gathering is a new DBMS_STATS feature in Oracle 11gR2 that provides inter-object parallelism. When CONCURRENT is set to TRUE (FALSE by default) and Estimate Percent is set to DBMS_STATS.AUTO_SAMPLE_SIZE, Oracle uses the Oracle Job Scheduler ( job_queue_processes should be set to more than 4) and Advanced Queuing to concurrently run multiple gather statistics jobs across multiple objects.

The preference can be set using the following command:

exec DBMS_STATS.SET_GLOBAL_PREFS('CONCURRENT','TRUE');

This needs to be set by logging in as sys user.

When using DEGREE, parallel servers will only work on one table partition at a time. This limitation is overcome when combined with the Oracle 11gR2 concurrent feature.

This combined approach is most beneficial when there is plenty of available CPU. It is aimed at systems with several CPUs that have spare capacity when the gather statistics process is run. It is unlikely to make much difference on small servers (that are less likely to have huge partitions), or large servers that have reach maximum CPU during traditional (non-concurrent) statistics gathering.

Although these features are supported, they may not provide a significant improvement as FND_STATS in 11gR2 already uses the AUTO sample size and parallelism while gathering stats. In addition FND_STATS calls

11


DBMS_STATS.gather_table_stats and not gather_schema_statistics. So this feature can only be used to gather statistics for different partitions of the same table at the same time. So it is only beneficial on partitioned tables, where data is reasonably well distributed across the partitions.

Verifying Statistics

The FND_STATS.VERIFY_STATS procedure can be used to check that statistics are up-to-date. It produces a report and can be run using the following syntax:

set serveroutput on

set long 10000

exec apps.fnd_stats.verify_stats (‘OWNER’, ‘TABLE_NAME’); For example:

exec apps.fnd_stats.verify_stats ('ONT', 'ONT.OE_ORDER_HOLDS_ALL');

Alternatively, you can verify statistics using the scripts from My Oracle Support Knowledge Document 163208.1. bde_last_analyzed.sql - Verifies CBO Statistics.

Standalone Patches

The standalone patches for Oracle 11g are shown in Table 1: Oracle E-Business Suite Standalone Patches for the following new features implemented in FND_STATS code.

1. Database 11g AUTO Sampling Statistics Gathering Option

Those unable to apply the Auto sampling standalone patch should review the automated script provided in My Oracle Knowledge Document 156968.1: coe_stats.sql - Automates CBO Stats Gathering using FND_STATS and Table sizes. This interim solution provides recommendations based on the range, sample size, and collection frequency.

2. Database 11g Extended Statistics Gathering Feature

TABLE 1. ORACLE E-BUSINESS SUITE STANDALONE PATCHES (ORACLE 11G)

E-BUSINESS SUITE VERSION PATCH

12.1.x 16410424:R12.FND.B

12.0.x 16410424:R12.FND.A

11.5.10 14707975

12


Performance Test Cases & Examples

1. Oracle 11g Extended Optimizer Statistics 2. Incremental Statistics Gathering 3. Concurrent Stats Gathering

1. Oracle 11g Extended Optimizer Statistics This simple example demonstrates how you might deploy the Oracle 11g extended (multi-column) optimizer statistics feature in Oracle E-Business Suite. Recall that, individual column statistics provide the selectivity of a specific column, but when the where clause includes multiple columns from the same table, the individual column statistics provide no indication of the relationship between the columns. The extended statistics increase the accuracy of the combined selectivity of the individual predicates, and hence the cardinality estimate, which results in a better execution plan. The FND_LOOKUP_VALUES table provides the basis for this example as it is suitably small, easy to understand, and may be familiar as is used across several modules. It stores Oracle Application Object Library QuickCode values. Essentially each row contains the QuickCode lookup type, the QuickCode itself, its meaning, and additional description (also other columns not directly pertinent to this example). These values are used by “List of Values” across several of the Oracle E-Business Suite forms. Several of the values are only used by specific modules and therefore they are typically both referenced when querying data, which is one of the primary conditions for using extended statistics. As such, this implies that there is a correlation between the columns, and in fact the LOOKUP_TYPE is dependent upon the VIEW_APPLICATION_ID.

1. Statistics without Extended Statistics

Gathering statistics on table without any multi column extension stats for this test case we are using FND_LOOKUP_VALUES table.

- Describe the FND_LOOKUP_VALUES table.

desc FND_LOOKUP_VALUES Datatype Length Mandatory Comments ---------------------- ------------ --------- ------------------------- LOOKUP_TYPE VARCHAR2 (30) Yes QuickCode lookup type LANGUAGE VARCHAR2 (30) Yes Language LOOKUP_CODE VARCHAR2 (30) Yes QuickCode code MEANING VARCHAR2 (80) Yes QuickCode meaning DESCRIPTION VARCHAR2 (240) Description ENABLED_FLAG VARCHAR2 (1) Yes Enabled flag SECURITY_GROUP_ID NUMBER (15) Yes Security group identifier VIEW_APPLICATION_ID NUMBER (15) Yes Identifies which application's view will include the

lookup values TERRITORY_CODE VARCHAR2 (2) Territory code of territory using the language .... .... - Total cardinality of the FND_LOOKUP_VALUES table: select count(1) from FND_LOOKUP_VALUES; COUNT(1) ---------- 532559 - Run gather statistics on FND_LOOKUP_VALUES table: exec fnd_stats.gather_table_stats('APPLSYS','FND_LOOKUP_VALUES'); - Check the updated statistics in dba_tables.

select table_name, num_rows from dba_tables where table_name='FND_LOOKUP_VALUES'; TABLE_NAME NUM_ROWS ------------------------------ ---------- FND_LOOKUP_VALUES 532559

13


- The following histograms are already present in the FND data dictionary: select table_name,column_name from fnd_histogram_cols where table_name='FND_LOOKUP_VALUES'; TABLE_NAME COLUMN_NAME ------------------------------ ------------------------------ FND_LOOKUP_VALUES LANGUAGE FND_LOOKUP_VALUES VIEW_APPLICATION_ID - Review the table data distribution. select VIEW_APPLICATION_ID,lookup_type, count(*) from FND_LOOKUP_VALUES group by VIEW_APPLICATION_ID,lookup_type order by 3 desc; VIEW_APPLICATION_ID LOOKUP_TYPE COUNT(*) ------------------- ------------------------------ ---------- 3 NO_POSTAL_CODE 13698 3 FI_POSTAL_CODE 11496 3 GHR_US_ACADEMIC_INSTITUTION 11043 3 AE_OCCUPATION 8760 3 PL_COMMUNITY 7437 222 NAICS_2002 7035 8901 FV_PSC_TYPE 6906 3 NAIC 6306 222 NAICS_1997 5430 3 GHR_US_ACADEMIC_DISCIPLINE 5127 222 NACE 4599 222 1987 SIC 4512 222 1972 SIC 4338 8405 PE_US_COURSE_SUBJECT 4299 3 DK_POSTCODE_TOWN 4002 8901 FV_NAICS_TYPE 3537 222 1977 SIC 3111 8901 FV_SIC_TYPE 3012 200 NUMBERS 3000 3 GHR_US_OCC_SERIES 2889 3 NO_TAX_MUNICIPALITY 2619 3 GHR_US_LANG_IDENTIFIER 2547 3 GHR_US_FOR_ALL_LOC 2238 3 GHR_US_AGENCY_CODE 2208 0 BUSINESS_ENTITY 2199 222 NAF 2145 3 GHR_US_LEGAL_AUTHORITY 2031 3 GHR_US_HEALTH_PLAN 2001 8405 IMPORT_ERROR_CODE 1890 3 HU_JOB_FEOR_CODES 1884 222 HZ_RELATIONSHIP_ROLE 1521 8901 FV_FSC_TYPE 1365 3 AE_AREA_CODES 1350 - Check the data distribution in fnd_lookup_Values table. select table_name, column_name, num_distinct, density, num_buckets, histogram from user_tab_col_statistics where table_name = 'FND_LOOKUP_VALUES'; TABLE_NAME COLUMN_NAME NUM_DISTINCT DENSITY NUM_BUCKETS HISTOGRAM --------------------------------------------- ------------ ---------- ----------- --------- FND_LOOKUP_VALUES LOOKUP_TYPE 17203 .000058129 1 NONE FND_LOOKUP_VALUES LANGUAGE 3 9.3886E-07 3 FREQUENCY FND_LOOKUP_VALUES LOOKUP_CODE 86164 .000011606 1 NONE FND_LOOKUP_VALUES MEANING 330917 3.0219E-06 1 NONE FND_LOOKUP_VALUES DESCRIPTION 207029 4.8302E-06 1 NONE FND_LOOKUP_VALUES ENABLED_FLAG 4 .25 1 NONE FND_LOOKUP_VALUES SECURITY_GROUP_ID 17 .058823529 1 NONE FND_LOOKUP_VALUES VIEW_APPLICATION_ID 87 9.3886E-07 87 FREQUENCY FND_LOOKUP_VALUES TERRITORY_CODE 10 .1 1 NONE .... .... - Check explain plan of the select for the different values of view_application_id and lookup_type explain plan for select * from FND_LOOKUP_VALUES where VIEW_APPLICATION_ID=201 and lookup_type=:b2; Explained. select plan_table_output from table(dbms_xplan.display('plan_table',null,'serial')); PLAN_TABLE_OUTPUT

14


---------------------------------------------------------------------------------------------------- | Id | Operation | Name | Rows | Bytes | Cost (%CPU)| Time | ---------------------------------------------------------------------------------------------------- | 0 | SELECT STATEMENT | | 1 | 152 | 4 (0)| 00:00:01 | | 1 | TABLE ACCESS BY INDEX ROWID| FND_LOOKUP_VALUES | 1 | 152 | 4 (0)| 00:00:01 | |* 2 | INDEX RANGE SCAN | FND_LOOKUP_VALUES_U1 | 1 | | 3 (0)| 00:00:01 | ---------------------------------------------------------------------------------------------------- Predicate Information: 2 - access("LOOKUP_TYPE"=:B2 AND "VIEW_APPLICATION_ID"=201) explain plan for select * from FND_LOOKUP_VALUES where VIEW_APPLICATION_ID=3 and lookup_type='NO_POSTAL_CODE'; Explained. select plan_table_output from table(dbms_xplan.display('plan_table',null,'serial')); PLAN_TABLE_OUTPUT ---------------------------------------------------------------------------------------------------- | Id | Operation | Name | Rows | Bytes | Cost (%CPU)| Time | ---------------------------------------------------------------------------------------------------- | 0 | SELECT STATEMENT | | 12 | 1680 | 13 (0)| 00:00:01 | | 1 | TABLE ACCESS BY INDEX ROWID| FND_LOOKUP_VALUES | 12 | 1680 | 13 (0)| 00:00:01 | |* 2 | INDEX RANGE SCAN | FND_LOOKUP_VALUES_U2 | 12 | | 3 (0)| 00:00:01 | ---------------------------------------------------------------------------------------------------- Predicate Information: 2 - access("LOOKUP_TYPE"='NO_POSTAL_CODE' AND "VIEW_APPLICATION_ID"=3) explain plan for select * from FND_LOOKUP_VALUES where VIEW_APPLICATION_ID=3 and lookup_type='AE_ARIA_CODE'; Explained. select plan_table_output from table(dbms_xplan.display('plan_table',null,'serial')); PLAN_TABLE_OUTPUT ---------------------------------------------------------------------------------------------------- | Id | Operation | Name | Rows | Bytes | Cost (%CPU)| Time | ---------------------------------------------------------------------------------------------------- | 0 | SELECT STATEMENT | | 12 | 1680 | 13 (0)| 00:00:01 | | 1 | TABLE ACCESS BY INDEX ROWID| FND_LOOKUP_VALUES | 12 | 1680 | 13 (0)| 00:00:01 | |* 2 | INDEX RANGE SCAN | FND_LOOKUP_VALUES_U2 | 12 | | 3 (0)| 00:00:01 | ---------------------------------------------------------------------------------------------------- Predicate Information: 2 - access("LOOKUP_TYPE"='AE_ARIA_CODE' AND "VIEW_APPLICATION_ID"=3)

2. Statistics with Extended Stats The extended statistics approach will create a column statistics entry for the column group in the Data Dictionary. In this example they are gathered manually so that the result can be seen immediately. - Steps to create multi column extended statistics - Query fnd_extnstats_cols table to confirm that extended statistics do not exist. fnd_exnstats_cols is FND table

which holds the columns on which extension stats will be created. select table_name from fnd_extnstats_cols; no rows selected - Delete the current statistics on FND_LOKUP_VALUES table analyze table applsys.fnd_lookup_values delete statistics; Table analyzed. - Create extended statistics for the LOOKUP_TYPE and VIEW_APPLICATION_ID: set serveroutput on - Execute fnd_stats.load_extnstats_cols. Create multi column extension stats using fnd_stats exec fnd_stats.load_extnstats_cols(action =>'INSERT',appl_id=>0,tabname => 'FND_LOOKUP_VALUES',colname1 =>'VIEW_APPLICATION_ID',colname2 =>'LOOKUP_TYPE'); PL/SQL procedure successfully completed. - check if virtual column is created for the multi-columns involved for the extended statistics. select owner,table_name,column_name,histogram from dba_tab_cols where table_name like 'FND_LOOKUP_VALUES' and owner ='APPLSYS'; OWNER TABLE_NAME COLUMN_NAME HISTOGRAM

15


------------------------------ ------------------ ------------------------------ -------------------- APPLSYS FND_LOOKUP_VALUES ZD_EDITION_NAME NONE APPLSYS FND_LOOKUP_VALUES SYS_STU_L5AKW15J1DXUKXDMYW3UJF NONE APPLSYS FND_LOOKUP_VALUES - Above output shows that the virtual column “SYS_STU_L5AKW15J1DXUKXDMYW3UJF“ is created. - Gather statistics on 'FND_LOOKUP_VALUES' to check if histogram is getting created on the virtual column exec fnd_stats.gather_table_stats('APPLSYS','FND_LOOKUP_VALUES'); PL/SQL procedure successfully completed. - check if histogram is created on virtual column select owner,table_name,column_name,histogram from dba_tab_cols where table_name like 'FND_LOOKUP_VALUES' and owner ='APPLSYS'; OWNER TABLE_NAME COLUMN_NAME HISTOGRAM ------------------------------ ------------------------------------------ ------------- APPLSYS FND_LOOKUP_VALUES ZD_EDITION_NAME NONE APPLSYS FND_LOOKUP_VALUES SYS_STU_L5AKW15J1DXUKXDMYW3UJF HEIGHT BALANCED - Above output shows that the height balanced histogram is created on the virtual column - Now check the changes in the explain plan of the select for the different values of view_application_id and

lookup_type explain plan for select * from FND_LOOKUP_VALUES where VIEW_APPLICATION_ID=201 and lookup_type=:b2; Explained. select plan_table_output from table(dbms_xplan.display('plan_table',null,'serial')); PLAN_TABLE_OUTPUT ---------------------------------------------------------------------------------------------------- | Id | Operation | Name | Rows | Bytes | Cost (%CPU)| Time | ---------------------------------------------------------------------------------------------------- | 0 | SELECT STATEMENT | | 30 | 4560 | 27 (0)| 00:00:01 | | 1 | TABLE ACCESS BY INDEX ROWID| FND_LOOKUP_VALUES | 30 | 4560 | 27 (0)| 00:00:01 | |* 2 | INDEX RANGE SCAN | FND_LOOKUP_VALUES_U2 | 30 | | 3 (0)| 00:00:01 | ---------------------------------------------------------------------------------------------------- Predicate Information (identified by operation id): --------------------------------------------------- 2 - access("LOOKUP_TYPE"=:B2 AND "VIEW_APPLICATION_ID"=201) explain plan for select * from FND_LOOKUP_VALUES where VIEW_APPLICATION_ID=3 and lookup_type='NO_POSTAL_CODE' Explained. select plan_table_output from table(dbms_xplan.display('plan_table',null,'serial')); PLAN_TABLE_OUTPUT --------------------------------------------------------------------------------------- | Id | Operation | Name | Rows | Bytes | Cost (%CPU)| Time | --------------------------------------------------------------------------------------- | 0 | SELECT STATEMENT | | 12580 | 1867K| 2729 (2)| 00:00:33 | |* 1 | TABLE ACCESS FULL| FND_LOOKUP_VALUES | 12580 | 1867K| 2729 (2)| 00:00:33 | --------------------------------------------------------------------------------------- Predicate Information (identified by operation id): --------------------------------------------------- 1 - filter("LOOKUP_TYPE"='NO_POSTAL_CODE' AND "VIEW_APPLICATION_ID"=3) explain plan for select * from FND_LOOKUP_VALUES where VIEW_APPLICATION_ID=3 and lookup_type='AE_ARIA_CODE'; Explained. select plan_table_output from table(dbms_xplan.display('plan_table',null,'serial')); PLAN_TABLE_OUTPUT ---------------------------------------------------------------------------------------------------- | Id | Operation | Name | Rows | Bytes | Cost (%CPU)| Time | ---------------------------------------------------------------------------------------------------- | 0 | SELECT STATEMENT | | 24 | 3648 | 22 (0)| 00:00:01 | | 1 | TABLE ACCESS BY INDEX ROWID| FND_LOOKUP_VALUES | 24 | 3648 | 22 (0)| 00:00:01 | |* 2 | INDEX RANGE SCAN | FND_LOOKUP_VALUES_U2 | 24 | | 3 (0)| 00:00:01 | ---------------------------------------------------------------------------------------------------- Predicate Information (identified by operation id):

16


--------------------------------------------------- 2 - access("LOOKUP_TYPE"='AE_ARIA_CODE' AND "VIEW_APPLICATION_ID"=3) Result: Above results shows the cardinality estimation changes after creating multi column extended stats.

2. Incremental Statistics Gathering

Incremental statistics gathering is a new feature of 11G Database. Incremental stats will work only for partitioned tables where global statistics are updated incrementally by scanning only the partitions that have changed from the last run. This simple example demonstrates how you might deploy the Oracle 11g incremental Statistics Gathering feature in Oracle E-Business Suite. - Create a table xla_ae_lines_bkp as a partition table to perform this test case - Insert data from xla_ae_lines table insert into xla.xla_Ae_lines_bkp select * from xla.xla_ae_lines; 1909409 rows created. - Test case without setting incremental preference set time on timing on exec fnd_stats.gather_table_stats('XLA','XLA_AE_LINES_BKP'); PL/SQL procedure successfully completed. Elapsed: 00:00:29.06 - It took 29.06 sec to gather stats for the first time after data insertion. - Check the statistics and global and partition level select table_name,to_Char(last_analyzed,'DD-MON-YY HH24:MI:SS') "last_analyzed" from dba_Tables where table_name like 'XLA_AE_LINES_BKP'; TABLE_NAME last_analyzed ------------------------------ ------------------------ XLA_AE_LINES_BKP 23-SEP-12 07:04:34 select partition_name,to_Char(last_analyzed,'DD-MON-YY HH24:MI:SS') "last_analyzed" from dba_Tab_partitions where table_name like 'XLA_AE_LINES_BKP'; PARTITION_NAME last_analyzed ------------------------------ ------------------------ AP 23-SEP-12 07:04:15 AR 23-SEP-12 07:04:16 CE 23-SEP-12 07:04:16 CST 23-SEP-12 07:04:23 DPP 23-SEP-12 07:04:23 FUN 23-SEP-12 07:04:23 FV 23-SEP-12 07:04:23 GMF 23-SEP-12 07:04:23 IGC 23-SEP-12 07:04:23 IGI 23-SEP-12 07:04:23 LNS 23-SEP-12 07:04:23 OFA 23-SEP-12 07:04:23 OKL 23-SEP-12 07:04:23 OZF 23-SEP-12 07:04:23 PA 23-SEP-12 07:04:24 PAY 23-SEP-12 07:04:24 PN 23-SEP-12 07:04:24 PO 23-SEP-12 07:04:24 PSB 23-SEP-12 07:04:24 - Delete the data from one of the partition to check how statistics are getting calculated delete from xla.xla_ae_lines_bkp where application_id=222; - Gathering stats usign fnd_Stats exec fnd_stats.gather_table_stats('XLA','XLA_AE_LINES_BKP'); PL/SQL procedure successfully completed. Elapsed: 00:00:14.06 - After deleting the data ran gather stats without setting the preference. It took 14.06 sec

17


- Checking global stats and partition stats select table_name,to_Char(last_analyzed,'DD-MON-YY HH24:MI:SS') "last_analyzed" from dba_Tables where table_name like 'XLA_AE_LINES_BKP'; TABLE_NAME last_analyzed ------------------------------ ------------------------ XLA_AE_LINES_BKP 23-SEP-12 07:10:26 select partition_name,to_Char(last_analyzed,'DD-MON-YY HH24:MI:SS') "last_analyzed" from dba_Tab_partitions where table_name like 'XLA_AE_LINES_BKP'; PARTITION_NAME last_analyzed ------------------------------ ------------------------ AP 23-SEP-12 07:10:14 AR 23-SEP-12 07:10:14 CE 23-SEP-12 07:10:14 CST 23-SEP-12 07:10:15 DPP 23-SEP-12 07:10:15 FUN 23-SEP-12 07:10:15 FV 23-SEP-12 07:10:15 GMF 23-SEP-12 07:10:15 IGC 23-SEP-12 07:10:15 IGI 23-SEP-12 07:10:15 LNS 23-SEP-12 07:10:16 OFA 23-SEP-12 07:10:16 OKL 23-SEP-12 07:10:16 OZF 23-SEP-12 07:10:16 PA 23-SEP-12 07:10:17 PAY 23-SEP-12 07:10:17 PN 23-SEP-12 07:10:17 PO 23-SEP-12 07:10:17 PSB 23-SEP-12 07:10:17 19 rows selected. NOTE : statistics are gathered on all the partitions even though only AR parition data is deleted, last_analyzed is updated for all the partitions. - Perform the same test case after setting the preference to true. Following command will set the incremental

preference at the table level so next time when gather statistics is run, it will gather stats incremental on this table

- Drop and Recreate the table xla_ae_lines_bkp as a partition table to perform this test case - Insert data from xla_ae_lines table insert into xla.xla_Ae_lines_bkp select * from xla.xla_ae_lines; 1909409 rows created. exec dbms_stats.set_table_prefs('XLA','XLA_AE_LINES_BKP','INCREMENTAL','TRUE'); - Check if the preference is set select dbms_stats.get_prefs('INCREMENTAL','XLA','XLA_AE_LINES_BKP') from dual; DBMS_STATS.GET_PREFS('INCREMENTAL','XLA','XLA_AE_LINES_BKP') ------------------------------------------------------------ TRUE - Gather statistics after setting the preference.

exec fnd_stats.gather_table_stats('XLA','XLA_AE_LINES_BKP'); PL/SQL procedure successfully completed. Elapsed: 00:00:22.91 - After setting the preference it took 22.91 sec to gather statistics for the first time after data insertion - checking stats timing information select table_name,to_Char(last_analyzed,'DD-MON-YY HH24:MI:SS') "last_analyzed" from dba_Tables where table_name like 'XLA_AE_LINES_BKP'; TABLE_NAME last_analyzed ------------------------------ ------------------------ XLA_AE_LINES_BKP 23-SEP-12 07:17:32 select partition_name,to_Char(last_analyzed,'DD-MON-YY HH24:MI:SS') "last_analyzed" from dba_Tab_partitions where table_name like 'XLA_AE_LINES_BKP';

18


PARTITION_NAME last_analyzed ------------------------------ ------------------------ AP 23-SEP-12 07:17:30 AR 23-SEP-12 07:17:12 CE 23-SEP-12 07:17:10 CST 23-SEP-12 07:17:21 DPP 23-SEP-12 07:17:21 FUN 23-SEP-12 07:17:12 FV 23-SEP-12 07:17:10 GMF 23-SEP-12 07:17:10 IGC 23-SEP-12 07:17:10 IGI 23-SEP-12 07:17:12 LNS 23-SEP-12 07:17:10 OFA 23-SEP-12 07:17:10 OKL 23-SEP-12 07:17:12 OZF 23-SEP-12 07:17:30 PA 23-SEP-12 07:17:12 PAY 23-SEP-12 07:17:12 PN 23-SEP-12 07:17:12 PO 23-SEP-12 07:17:10 PSB 23-SEP-12 07:17:30 19 rows selected. - Deleting the data from one partition to see how incremental statistics gathering will help next time when stats are

gathered. delete from xla_ae_lines_bkp where application_id=222; 100233 rows deleted - Gather statistics after deleting data from application_id 222 partition AR exec fnd_stats.gather_table_stats('XLA','XLA_AE_LINES_BKP'); PL/SQL procedure successfully completed. Elapsed: 00:00:04.11 - After deleting the data for one partition incremental gathering statistics took 4.11 sec - Check global and partition statistics select table_name,to_Char(last_analyzed,'DD-MON-YY HH24:MI:SS') "last_analyzed" from dba_Tables where table_name like 'XLA_AE_LINES_BKP'; TABLE_NAME last_analyzed ------------------------------ ------------------------ XLA_AE_LINES_BKP 23-SEP-12 07:26:20 select partition_name,to_Char(last_analyzed,'DD-MON-YY HH24:MI:SS') "last_analyzed" from dba_Tab_partitions where ta ble_name like 'XLA_AE_LINES_BKP'; PARTITION_NAME last_analyzed ------------------------------ ------------------------ AP 23-SEP-12 07:17:30 AR 23-SEP-12 07:26:18 CE 23-SEP-12 07:17:10 CST 23-SEP-12 07:17:21 DPP 23-SEP-12 07:17:21 FUN 23-SEP-12 07:17:12 FV 23-SEP-12 07:17:10 GMF 23-SEP-12 07:17:10 IGC 23-SEP-12 07:17:10 IGI 23-SEP-12 07:17:12 LNS 23-SEP-12 07:17:10 OFA 23-SEP-12 07:17:10 OKL 23-SEP-12 07:17:12 OZF 23-SEP-12 07:17:30 PA 23-SEP-12 07:17:12 PAY 23-SEP-12 07:17:12 PN 23-SEP-12 07:17:12 PO 23-SEP-12 07:17:10 PSB 23-SEP-12 07:17:30 19 rows selected. Elapsed: 00:00:00.01 Note : statistics are gathered, last_analyzed date is changed for only AR partition and timing has also reduced.

19


- With the above test case we can see improvement in timings when incremental preference is set. - Timing information: Preference set Table_name Operation Timing --------------- ------------ ----------------------------------------------------- --------- NO xla_ae_lines stats gathered after inserting data for first time 29.06 sec Yes xla_ae_lines stats gathered after inserting data for first time 22.91 sec NO xla_ae_lines stats gathered after modifying one partition 14.06 sec Yes xla_ae_lines stats gathered after modifying one partition 04.11 sec --------------- ------------ ----------------------------------------------------- ---------

3. Concurrent Statistics Gathering Concurrent statistics gathering is a new feature introduced in 11gR2. Concurrent statistics gathering is use to run stats on multiple table and partitions concurrently. This simple example demonstrates how you might deploy the Oracle 11g Concurrent Statistics Gathering feature in Oracle E-Business Suite. - Concurrent statistics gathering is controlled by setting global preference exec DBMS_STATS.SET_GLOBAL_PREFS('CONCURRENT','TRUE'); - To run statistics concurrently job_queue_processes should be set to more than 4 - Test case to gather stats concurrently. - Setting job_queue_processes to 10. alter system set job_queue_processes=10; system altered. - Setting concurrent preference to FALSE exec DBMS_STATS.SET_GLOBAL_PREFS('CONCURRENT','FALSE'); - Delete statistics on the table exec dbms_stats.delete_table_stats('XLA','XLA_AE_HEADERS'); - Gather statistics using fnd_stats exec fnd_stats.gather_table_stats('XLA','XLA_AE_HEADERS'); Elapsed: 00:29:30.97 - Set the global preference to TRUE exec DBMS_STATS.SET_GLOBAL_PREFS('CONCURRENT','TRUE'); - Delete the statistics exec dbms_stats.delete_table_stats('XLA','XLA_AE_HEADERS'); PL/SQL procedure successfully completed. - Gathering Statistics exec fnd_stats.gather_table_stats('XLA','XLA_AE_HEADERS'); PL/SQL procedure successfully completed. Elapsed: 00:26:59.54 Results of the test case: Method Level (schema/table) Concurrent processing option Timing ---------- --------------------- ---------------------------- ----------- FND_STATS table(XLA_AE_HEADERS) FALSE 00:29:30.97 FND_STATs table(XLA_AE_HEADERS) TRUE 00:26:59.54 ---------- --------------------- ---------------------------- ----------- From the above test cases, we see fnd_stats (with parallel option, as fnd_stats runs by default with parallel option) with setting global preference to FALSE took 29 mins and 30 secs. fnd_stats ran with setting global preference to TRUE, took 26 mins and 59 secs. We see a net gain of 2 mins 5 secs in the gather statistics timing. And even this small gain could be explained by blocks already being in cache.

20


Appendix A: Related Documentation

This appendix contains important Oracle documents, My Oracle Support Knowledge Documents and Blog references that provide useful information.

Whitepapers

These papers are generic and do not specifically focus on Oracle E-Business Suite.

• Best Practices for Gathering Optimizer Statistics This paper provides an extensive overview of the Oracle Database statistics gathering process and is a useful precursor to this paper. It should be considered as the source for many of the generic statements in this document. This is available on the Oracle Optimizer blog: https://blogs.oracle.com/optimizer/ and Oracle Technet: http://www.oracle.com/technetwork/database/bi-datawarehousing/dbbi-tech-info-optmztn-092214.html

• Understanding Optimizer Statistics This document provides more foundation information about Optimizer statistics and related topics. For example, it includes a clear explanation of the different types of histograms and where they are applicable. This could be considered a precursor to the Best Practices for Gathering Optimizer Statistics paper. This is available on Oracle Technet: http://www.oracle.com/technetwork/database/focus-areas/bi-datawarehousing/twp-optimizer-stats-concepts-110711-1354477.pdf

• Upgrading from Oracle Database 10g to 11g: What to expect from the Optimizer This paper introduces the new Oracle 11g Optimizer features and outlines essential steps before and after the upgrade in order to avoid performance regressions related to plan changes that are likely to occur. This is available on Oracle Technet: http://www.oracle.com/technetwork/database/focus-areas/bi-datawarehousing/twp-upgrading-10g-to-11g-what-to-ex-133707.pdf

My Oracle Support Knowledge Documents

• Fixed Objects Statistics Considerations (Doc ID 798257.1) This generic document summarizes problems including, for example, missing or bad statistics on the x$/fixed tables that can lead to performance degradation or hangs.

• Managing CBO Stats during an upgrade to Oracle 10g or Oracle 11g (Doc ID 465787.1) The core idea presented is to continue gathering Optimizer statistics for application schemas as usual, but create a clean baseline for non-application objects (data dictionary, fixed objects and system performance).

• bde_last_analyzed.sql - Verifies CBO Statistics (Doc ID 163208.1) This script verifies the Optimizer statistics in the data dictionary for all tables, indexes, and partitions. It also validates the statistics on tables and indexes owned by 'SYS'.

• EBPERF FAQ - Collecting Statistics with Oracle EBS 11i and R12 (Doc ID 368252.1)

This FAQ note assists with generic questions about gathering stats for Apps 11i and R12

The following My Oracle Support Knowledge Documents are useful if you are unable to apply the FND_STATS standalone patch:

• coe_stats.sql - Automates CBO Stats Gathering using FND_STATS (Doc ID 156968.1) This interim solution provides recommendations based on the range, sample size, and collection frequency.

Blogs

21

https://blogs.oracle.com/optimizer/

http://www.oracle.com/technetwork/database/bi-datawarehousing/dbbi-tech-info-optmztn-092214.html

http://www.oracle.com/technetwork/database/focus-areas/bi-datawarehousing/twp-optimizer-stats-concepts-110711-1354477.pdf

http://www.oracle.com/technetwork/database/focus-areas/bi-datawarehousing/twp-upgrading-10g-to-11g-what-to-ex-133707.pdf

http://www.oracle.com/technetwork/database/focus-areas/bi-datawarehousing/twp-upgrading-10g-to-11g-what-to-ex-133707.pdf


The Oracle Optimizer blog ( https://blogs.oracle.com/optimizer ) contains excellent generic information on a number of topics from the author of the Best Practices for Gathering Optimizer Statistics paper. Topics include the following • Extended Statistics

• Concurrent Statistics Gathering

• Maintaining statistics on large partitioned tables

• Fixed Objects Statistics and why they are important

In addition there are articles on other tuning related issues such as cardinality feedback.

22

https://blogs.oracle.com/optimizer


Appendix B: Package Specifications

This appendix describes the Gather Table Statistics and Gather Schema Statistics programs. In both cases they attempt to parallelize as much of the work as possible and can optionally backup the existing statistics in the FND_STATTAB table (only if the value of the backup flag parameter is BACKUP) before gathering new statistics, ensuring that they can be restored if necessary.

Gather Table Statistics

The Gather Table Statistics program gathers table, column and index statistics. This program calls the FND_STATS.GATHER_TABLE_STATS package; the parameters are shown in Table B1. It is useful when collecting statistics for very large tables that have been excluded from the main gather statistics collection due to time constraints. It is also useful to update the statistics for a small set of tables when investigating performance of a single SQL statement.

Figure 1 : Oracle E-Business Suite - Gather Table Statistics Concurrent Request Submission window Figure 1 shows an example of submitting Gather Table Statistics. Note that Estimate Percent is blank; a description of the parameters follows.

TABLE B1. FND_STATS.GATHER_TABLE_STATS PARAMETERS

PARAMETER DESCRIPTION

Owner Name The table owner.

Table Name The table name.

Estimate Percent The sampling percentage. If left blank, for 11g defaults to dbms_stats.auto_sample_size and for

23


10g and prior it the default value of 10 is used. The valid range is from 0 to 100.

Degree The degree of parallelism to be used when gathering statistics. If a Degree is not provided, it defaults to the minimum of parallel_max_servers and cpu_count.

Partition Name The partition name.

Backup Flag Specify BACKUP to back up your statistics. (BACKUP/NOBACKUP)

Granularity The granularity of statistics to collect (only relevant for partitioned tables). Valid values are: • DEFAULT - Gather global and partition–level statistics.

• SUBPARTITION - Gather subpartition–level statistics.

• PARTITION - Gather partition–level statistics.

• GLOBAL - Gather global statistics.

• ALL - Gather all (subpartition, partition, and global) statistics.

History Mode This parameter controls the number of history records. The options, which are discussed in the History Mode section, include LASTRUN, FULL and NONE. The default is LASTRUN.

Invalidate Dependent Cursors This indicates whether dependent cursors should be invalidated. This parameter is ignored prior to Oracle 9iR2

24


Gather Schema Statistics

The Gather Schema Statistics program gathers table, column and index statistics for every object in the schema. This program calls the FND_STATS.GATHER_SCHEMA_STATS package; the parameters are shown in Table B2. This is the recommended process to run in order to update Oracle E-Business Suite statistics. Various considerations are explained throughout the main part of this paper. When the collection time exceeds the available time slot, it may be useful to exclude very large tables and schedule these at different times.

If this procedure fails at any time during operation, supplying the request ID, for the request that failed, can restart it. The request ID, which can be queried from the FND_STATS_HIST table, is captured when the program is started from concurrent manager or will be automatically created if the FND_STATS.GATHER_SCHEMA_STATS is used directly or in a custom script.

Figure 2 : Oracle E-Business Suite - Gather Schema Statistics Concurrent Request Submission window Figure 2 shows an example of submitting Gather Schema Statistics. Note that Estimate Percent is blank; a description of the parameters follows.

TABLE B2: FND_STATS.GATHER_SCHEMA_STATS PARAMETERS

PARAMETER DESCRIPTION

Schema Name ALL will collect statistics for every registered Oracle E-Business Suite schema listed in FND_PRODUCT_INSTALLATIONS table.

Estimate percent The sampling percentage. If left blank, for 11g defaults to dbms_stats.auto_sample_size and for 10g and prior it the default value of 10 is used. The valid range is from 0 to 100.

25


Degree The degree of parallelism for gathering statistics. Default value is MIN (DB init parameters: parallel_max_servers, cpu_count).

Backup Flag Specify BACKUP to back up your statistics. (BACKUP/NOBACKUP)

Restart Request ID If the Gather Schema Statistics run failed or was stopped, it can be re-submitted with this parameter and it will pick up where the failed run left off.

History Mode This parameter controls the number of history records. The options, which are discussed in the History Mode section, include LASTRUN, FULL and NONE. The default is LASTRUN.

Gather Options This parameter specifies how objects are selected for statistics gathering. • GATHER (Default) - All schema tables and indexes are selected for statistics gathering.

• GATHER_AUTO - Tables of the schema schemaname for which the percentage of modifications

has exceeded modpercent are selected for statistics gathering. Table monitoring needs to be

enabled before using this option.

• GATHER_EMPTY - Statistics are gathered only for tables and indexes that are missing

statistics.

• LIST_AUTO - It only provides a list of objects for gather statistics if GATHER_AUTO is used.

• LIST_EMPTY - It only provides a list of objects for gather statistics if GATHER_EMPTY is used.

Modification Threshold This option is applicable only to GATHER AUTO and LIST AUTO options. This parameter specifies the percentage of modifications that have to take place on a table before it can be picked up for AUTO statistics gathering.

Invalidate Dependent Cursors This indicates whether dependent cursors should be invalidated. The default is 'Y'.

26

Best Practices for Gathering Statistics with Oracle E-Business Suite Aug 2013

Authors: Deepak Bhatnagar, Mohammed Saleem, Andy Tremayne

Editor: Robert Farrington

Oracle E-Business Suite Performance Group

Oracle Corporation World Headquarters 500 Oracle Parkway Redwood Shores, CA 94065 U.S.A.

Worldwide Inquiries: Phone: +1.650.506.7000 Fax: +1.650.506.7200

oracle.com

Copyright © 2013, Oracle and/or its affiliates. All rights reserved.

This document is provided for information purposes only, and the contents hereof are subject to change without notice. This document is not warranted to be error-free, nor subject to any other warranties or conditions, whether expressed orally or implied in law, including implied warranties and conditions of merchantability or fitness for a particular purpose. We specifically disclaim any liability with respect to this document, and no contractual obligations are formed either directly or indirectly by this document. This document may not be reproduced or transmitted in any form or by any means, electronic or mechanical, for any purpose, without our prior written permission.

Oracle and Java are registered trademarks of Oracle and/or its affiliates. Other names may be trademarks of their respective owners.

Intel and Intel Xeon are trademarks or registered trademarks of Intel Corporation. All SPARC trademarks are used under license and are trademarks or registered trademarks of SPARC International, Inc. AMD, Opteron, the AMD logo, and the AMD Opteron logo are trademarks or registered trademarks of Advanced Micro Devices. UNIX is a registered trademark of The Open Group. 0113

DOC1586374_A

Documents

dictionary statistics

fixed statistics

incremental statistics

locking statistics

g extended statistics

fixed object statistics

sole discretion of oracle

oracle white paper