Release 2 (9.2)
Part Number A96533-01
 Home |
 Book List |
 Contents |
 Index |
 Master Index |
 Feedback |
| Oracle9i Database Performance Tuning Guide and Reference Release 2 (9.2) Part Number A96533-01 |
|
This chapter explains how to install and configure Statspack.
The chapter contains the following sections:
Statspack is a set of SQL, PL/SQL, and SQL*Plus scripts that allow the collection, automation, storage, and viewing of performance data. Statspack differs from the existing UTLBSTAT/UTLESTAT performance scripts in the following ways:
BSTAT/ESTAT).PERFSTAT store performance statistics. Instead of creating/dropping tables each time, data is inserted into the preexisting tables. This makes it easier to compare historical data.DBMS_JOB or an operating system utility.COMMIT or a ROLLBACK, and so calculates the number of transactions as 'user commits' + 'user rollbacks.' BSTAT/ESTAT considers a transaction to complete with a COMMIT only, and so assumes that transactions = 'user commits.' For this reason, comparing statistics for each transaction between Statspack and BSTAT/ESTAT can result in significantly different ratios.
When you run the Statspack installation script, the user PERFSTAT is created automatically. PERFSTAT owns all objects needed by the Statspack package and is granted limited query-only privileges on the V$ views required for performance tuning.
Statspack users become familiar with the concept of a snapshot, a single collection of performance data. Each snapshot taken is identified by a snapshot ID, which is a unique number generated at the time the snapshot is taken. Each time a new collection is taken, a new SNAP_ID is generated.
The SNAP_ID, along with the database identifier (DBID) and instance number (INSTANCE_NUMBER), comprise the unique key for a snapshot. Use of this unique combination allows storage of multiple instances of an Oracle Real Application Clusters database in the same tables.
After snapshots are taken, you can run the performance report. The report prompts you for start and end snapshot IDs and then calculates activity on the instance between the two snapshots, much like a BSTAT/ESTAT report. To compare, the first SNAP_ID supplied can be considered the equivalent of running BSTAT; the second SNAP_ID specified can be considered the equivalent of ESTAT. Unlike BSTAT/ESTAT, which can by its nature only compare two static data points, the report can compare any two snapshots specified.
The default initial and next extent sizes are 100k, 1MB, or 5MB for all Statspack tables and indexes. Approximately 64MB is required to install Statspack.
The amount of database space required by the Statspack package depends on the frequency of snapshots, the size of the database and instance, and the amount of data collected (which is configurable). It is therefore difficult to provide general storage clauses and space utilization predictions that are accurate at each site.
There are two ways to install Statspack:
Batch mode is useful when you do not want to be prompted for the PERFSTAT user's default and temporary tablespaces.
The first step is to create the PERFSTAT user, which owns all PL/SQL code and database objects created, including the Statspack tables, constraints, and the Statspack package. During installation, you are prompted for the PERFSTAT user's default and temporary tablespaces. The default tablespace is used to create all Statspack objects, such as tables and indexes. The temporary tablespace is used for sort-type activities.
| See Also:
Oracle9i Database Concepts for more information on temporary tablespaces |
To install the package, either change to the ORACLE_HOME rdbms/admin directory, or fully specify the ORACLE_HOME/rdbms/admin directory when calling the installation script, SPCREATE.
To run the installation script, you must use SQL*Plus and connect as a user with SYSDBA privilege. For example, start SQL*Plus, then:
On UNIX:
SQL> CONNECT / AS SYSDBA SQL> @?/rdbms/admin/spcreate
On NT:
SQL> CONNECT / AS SYSDBA SQL> @%ORACLE_HOME%\rdbms\admin\spcreate
The SPCREATE install script runs three other scripts. These scripts are called automatically, so you do not need to run them:
SPCUSR: Creates the user and grants privilegesSPCTAB: Creates the tablesSPCPKG: Creates the packageCheck each of the three output files produced (SPCUSR.LIS, SPCTAB.LIS, SPCPKG.LIS) by the installation to ensure that no errors were encountered.
To install in batch mode, you must assign values to the SQL*Plus variables that specify the default and temporary tablespaces before running SPCREATE. The variables are:
For example, on UNIX:
SQL> CONNECT / AS SYSDBA SQL> define default_tablespace='tools' SQL> define temporary_tablespace='temp' SQL> @?/rdbms/admin/spcreate
SPCREATE no longer prompts for this information.
|
Note: After the setup is complete, you can change the password of the |
The following topics are discussed in this section:
The simplest interactive way to take a snapshot is to login to SQL*Plus as the PERFSTAT user and run the procedure STATSPACK.SNAP. For example:
SQL> CONNECT perfstat/perfstat SQL> EXECUTE statspack.snap;
|
Note: In an Oracle Real Application Clusters environment, you must connect to the instance for which you want to collect data. |
Taking such a snapshot stores the current values for the performance statistics in the Statspack tables. This snapshot can be used as a baseline for comparison with another snapshot taken at a later time.
For better performance analysis, set the initialization parameter TIMED_STATISTICS to true. Statspack will then include important timing information in the data it collects. You can change the TIMED_STATISTICS parameter dynamically by using the ALTER SYSTEM statement. Timing data is important and is usually required by Oracle support to diagnose performance problems.
| See Also:
"Setting the Level of Statistics Collection" for information about |
Typically, to automate the gathering and reporting phases (during a benchmark, for example), you might need to know the snap_id of the snapshot just taken. To take a snapshot and display the snap_id, call the STATSPACK.SNAP function.
snap Function in SQL*PlusUsing an anonymous PL/SQL block,
SQL> variable snap number; SQL> begin :snap := statspack.snap; end; 2 / PL/SQL procedure successfully completed. SQL> print snap SNAP ---------- 12
To make performance comparisons from one day, week, or year to the next, you need multiple snapshots taken over a period of time. The best method to gather snapshots is to automate the collection at regular intervals. You have the following options:
DBMS_JOB procedure to schedule snapshots.The DBMS_JOB package is an Oracle-automated method for collecting statistics. A sample script on how to do this is supplied in SPAUTO.SQL, which schedules a snapshot every hour, on the hour.
You might want to schedule snapshots at regular times each day to reflect your system's OLTP or batch peak loads. For example, you could take snapshots at 9 a.m, 10 a.m, 11 a.m, 12 noon, and 6 p.m. for the OLTP load, and then take a snapshot at 12 midnight and 6 a.m for the batch window.
In order to use DBMS_JOB to schedule snapshots, you must set the JOB_QUEUE_PROCESSES initialization parameter to greater than 0 in the init.ora file, so that the job can be run automatically.
Use the DBMS_JOB.INTERVAL procedure to change the interval of statistics collection. For example, for job number 1:
EXECUTE DBMS_JOB.INTERVAL(1,'SYSDATE+(1/48)');
Where 'SYSDATE+(1/48)' results in the statistics being gathered each 1/48 hours, or every half hour.
To force the job to run immediately:
EXECUTE DBMS_JOB.RUN(<job number>);
To remove the autocollect job:
EXECUTE DBMS_JOB.REMOVE(<job number>);
| See Also:
Oracle9i Supplied PL/SQL Packages and Types Reference for more information on the |
After snapshots are taken, you can generate a performance report. The SQL script that generates the report prompts for the two snapshot IDs to be processed.
The Statspack package includes two reports. Both reports prompt you for a beginning snapshot ID, an ending snapshot ID, and a report name.
SPREPORT.SQL, which is a general instance health report that covers all aspects of instance performance; it calculates and prints ratios and differences for all statistics between the two snapshots, similar to the BSTAT/ESTAT report.Because data gathering is separate from report production, you have flexibility to base a report on any data points you select. For example, as DBA you might want to use the supplied automation script to automate data collection every hour, on the hour. If, at some later point, a performance issue arose that might be better investigated by looking at a three-hour data window, all you have to do is specify the required start point and end point when running the report.
To examine the change in instancewide statistics between two time periods, the SPREPORT.SQL file is run while connected to the PERFSTAT user. The SPREPORT.SQL command file is located in the rdbms/admin directory of the Oracle home.
|
Note: In an Oracle Real Application Clusters environment, you must connect to the instance on which you want to report. |
You are prompted for the following:
SQL> connect perfstat/perfstat SQL> @?/rdbms/admin/spreport
On NT:
SQL> connect perfstat/perfstat SQL> @%ORACLE_HOME%\rdbms\admin\spreport
Sample output:
SQL> connect perfstat/perfstat Connected. SQL> @spreport DB Id DB Name Inst Num Instance ----------- ------------ -------- ------------ 2618106428 PRD1 1 prd1 Completed Snapshots Snap Snap Instance DB Name Id Snap Started Level Comment ------------ ------------ ----- ----------------- ----- ---------------- prd1 PRD1 1 11 May 2000 12:07 5 2 11 May 2000 12:08 5 Specify the Begin and End Snapshot Ids ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ Enter value for begin_snap: 1 Begin Snapshot Id specified: 1 Enter value for end_snap: 2 End Snapshot Id specified: 2 Specify the Report Name ~~~~~~~~~~~~~~~~~~~~~~~ The default report file name is sp_1_2 To use this name, press <return> to continue, otherwise enter an alternative. Enter value for report_name: <press return or enter a new name> Using the report name sp_1_2
The report now scrolls past and is also written to the file specified (for example, sp_1_2.lis).
To run a report without being prompted, assign values to the SQL*Plus variables that specify the begin snap ID, the end snap ID, and the report name before running SPREPORT.
The variables are:
BEGIN_SNAP: specifies the begin snapshot IDEND_SNAP: specifies the end snapshot IDREPORT_NAME: specifies the report output nameSQL> connect perfstat/perfstat SQL> define begin_snap=1 SQL> define end_snap=2 SQL> define report_name=batch_run SQL> @?/rdbms/admin/spreport
SPREPORT no longer prompts for this information.
When you examine the instance report, you often find high-load SQL statements that you want to examine more closely. The SQL report, SPREPORT.SQL, displays statistics, the complete SQL text, and (if a level six snapshot has been taken), information on any SQL plan(s) associated with that statement.
| See Also:
"Snapshot Levels" for information about levels |
The SQL statement to be reported on is identified by a hash value, which is a numerical representation of the statement's SQL text. The hash value for each statement is displayed for each statement in the SQL sections of the instance report.
You run the SPREPORT.SQL file while connected as the PERFSTAT user. The report is located in the rdbms/admin directory of the Oracle home.
|
Note: In an Oracle Real Application Clusters environment, you must connect to the instance on which you want to report. |
The SPREPORT.SQL report prompts you for the following:
SQL> connect perfstat/perfstat Connected. SQL> @spreport DB Id DB Name Inst Num Instance ----------- ------------ -------- --------- 2618106428 PRD1 1 prd1 Completed Snapshots Snap Snap Instance DB Name Id Snap Started Level Comment ------------ ------------ ----- ----------------- ----- ------- prd1 PRD1 37 02 Mar 2001 11:01 6 38 02 Mar 2001 12:01 6 39 08 Mar 2001 09:01 5 40 08 Mar 2001 10:02 5 Specify the Begin and End Snapshot Ids ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ Enter value for begin_snap: 39 Begin Snapshot Id specified: 39 Enter value for end_snap: 40 End Snapshot Id specified: 40 Specify the Hash Value ~~~~~~~~~~~~~~~~~~~~~~ Enter value for hash_value: 1988538571 Hash Value specified is: 1988538571 Specify the Report Name ~~~~~~~~~~~~~~~~~~~~~~~ The default report file name is sp_39_40_1988538571. To use this name, press <return> to continue, otherwise enter an alternative. Enter value for report_name: Using the report name sp_39_40_1988538571
The report scrolls past and is also written to the file you specified,
sp_39_40_1988538571.lis.
Like SPREPORT.SQL, the SQL report can be run in batch mode. To run a report without being prompted, assign values to the SQL*Plus variables that specify the begin snap ID, the end snap ID, and the report name, before running spreport.
The variables are:
BEGIN_SNAP: specifies the begin snapshot IDEND_SNAP: specifies the end snapshot IDHASH_VALUE: specifies the hash valueREPORT_NAME: specifies the report output nameSQL> connect perfstat/perfstat SQL> define begin_snap=39 SQL> define end_snap=40 SQL define hash_value=1988538571 SQL> define report_name=batch_sql_run SQL> @sprepsql
SPREPSQL no longer prompts for this information.
For best performance when running the performance reports, collect optimizer statistics for tables and indexes owned by PERFSTAT. You should do this whenever there is significant change in data volumes in PERFSTAT's tables.
To collect optimizer statistics on the PERFSTAT schema, use DBMS_STATS or DBMS_UTILITY, and specify the PERFSTAT user. For example:
EXECUTE DBMS_STATS.GATHER_SCHEMA_STATS(OWNNAME=>'PERFSTAT',CASCADE=>TRUE);
or
EXECUTE DBMS_UTILITY.ANALYZE_SCHEMA('PERFSTAT','COMPUTE');
Both the snapshot level and the thresholds specified affect the amount of data Statspack captures.
You can change the amount of information gathered by specifying a different snapshot level. The higher the snapshot level, the more data is gathered. The default level set at installation is level 5.
For typical usage, level 5 snapshot is effective on most sites. There are certain situations when using a level 6 snapshot is beneficial. These include the following:
There are other parameters you can configure, in addition to the snapshot level. These parameters are used as thresholds when collecting data on SQL statements; data is captured on any SQL statements that breach the specified thresholds. Snapshot level and threshold information used by the package is stored in the STATS$STATSPACK_PARAMETER table.
You can change the default parameters used for taking snapshots so that they are tailored to the instance's workload. The full list of parameters that can be passed into the MODIFY_STATSPACK_PARAMETER procedure are the same as those for the SNAP procedure.
To temporarily use a snapshot level or threshold that is different from the instance's default snapshot values, you specify the required threshold or snapshot level when taking the snapshot. This value is used only for the immediate snapshot taken; the new value is not saved as the default.
For example, to take a single level 6 snapshot:
SQL> EXECUTE STATSPACK.SNAP(i_snap_level=>6);
You can save the new value as the instance's default in either of two ways:
STATSPACK.SNAP the I_MODIFY_PARAMETER input variable.
SQL> EXECUTE STATSPACK.SNAP - (i_snap_level=>10, i_modify_parameter=>'true');
Setting the I_MODIFY_PARAMETER value to true saves the new thresholds in the STATS$STATSPACK_PARAMETER table. These thresholds are used for all subsequent snapshots.
If the I_MODIFY_PARAMETER was set to false or omitted, then the new parameter values are not saved. Only the snapshot taken at that point uses the specified values. Any subsequent snapshots use the preexisting values in the STATS$STATSPACK_PARAMETER table.
STATSPACK.MODIFY_STATSPACK_PARAMETER procedure. For example, the following statement changes the snapshot level to 10 and modifies the SQL thresholds for BUFFER_GETS and DISK_READS:
SQL> EXECUTE STATSPACK.MODIFY_STATSPACK_PARAMETER - i_snap_level=>10, i_buffer_gets_th=>10000, i_disk_reads_th=>1000);
This procedure changes the values permanently, but does not take a snapshot.
Any level greater than 0 collects general performance statistics, such as wait statistics, system events, system statistics, rollback segment data, row cache, SGA, background events, session events, lock statistics, buffer pool statistics, and parent latch statistics.
This level includes all statistics gathered in the lower level(s), as well as performance data on SQL statements with high resource usage. In a level 5 snapshot (or higher), the time required for the snapshot to complete depends on the SHARED_POOL_SIZE and on the number of SQL statements in the shared pool at the time of the snapshot. The larger the shared pool, the longer it takes to complete the snapshot.
SQL statements gathered by Statspack are those that exceed one of six predefined threshold parameters:
The values of each of these threshold parameters are used when deciding which SQL statements to collect. If a SQL statement's resource usage exceeds any one of these threshold values, then it is captured during the snapshot.
The SQL threshold levels used are either those stored in the table STATS$STATSPACK_PARAMETER or by the thresholds specified when the snapshot is taken.
This level includes all statistics gathered in the lower levels, as well a SQL plans and plan usage data for each of the high-resource SQL statements captured.
A level 6 snapshot gathers valuable information for determining whether the execution plan used for a SQL statement has changed. Therefore, level 6 snapshots should be used whenever a plan might have changed.
To gather the plan for a SQL statement, the statement must be in the shared pool at the time the snapshot is taken, and it must exceed one of the SQL thresholds. To gather plans for all statements in the shared pool, specify the executions threshold to be zero (0) for those snapshots.
| See Also:
"Changing the Default Values for Snapshot Levels and SQL Thresholds" for information on how to do this |
This level includes all statistics gathered in the lower levels, as well as parent and child latch information. Sometimes data gathered at this level can cause the snapshot to take longer to complete. This level can be resource-intensive, and it should only be used when advised by Oracle personnel.
If you want to gather session statistics and wait events for a particular session (in addition to the instance statistics and wait events), specify the session ID in the call to Statspack. The statistics gathered for the session include session statistics, session events, and lock activity. The default behavior is to not gather session level statistics.
For example:
SQL> EXECUTE STATSPACK.SNAP(i_session_id=>3);
Parameters that can be passed to the STATSPACK.SNAP and STATSPACK.MODIFY_STATSPACK_PARAMETER procedures are as follows:
Oracle supports capturing certain performance data with microsecond granularity. Views that include microsecond timing include the following:
V$SYSTEM_EVENT, V$SESSION_EVENT (TIME_WAITED_MICRO column)V$SQL (CPU_TIME, ELAPSED_TIME columns)V$LATCH (WAIT_TIME column)V$SQL_WORKAREA, V$SQL_WORKAREA_ACTIVE (ACTIVE_TIME column)
Because microsecond timing might not be appropriate for rolled-up data, Statspack displays most times in seconds. It displays average times in milliseconds for easier comparison with operating system monitoring utilities, which often report timing in milliseconds.
For clarity, the time units used are specified in the column headings of each timed column in the Statspack report. The following convention are used:
(s) - a second
(cs) - a centisecond (100th of a second)
(ms) - a millisecond (1,000th of a second)
(us) - a microsecond (1,000,000th of a second)
If timings are available, the Statspack report orders wait events by time (in the Top-5 and background and foreground wait events sections).
If TIMED_STATISTICS is false for the instance, but a subset of users or programs set TIMED_STATISTICS to true dynamically, then the Statspack report output can look inconsistent, where some events have timings (those which the individual programs/users waited for) and some do not. The Top-5 section also looks unusual in this situation.
Optimally, TIMED_STATISTICS should be set to true at the instance level, for ease of diagnosing performance problems.
| See Also:
"Setting the Level of Statistics Collection" for information about |
If you want to share data with other sites (for example, if Oracle Support requires the raw statistics), then you can export the PERFSTAT user. An export parameter file (SPUEXP.PAR) is supplied for this purpose. To use this file, supply the export command with the userid parameter, along with the export parameter file name. For example:
exp userid=perfstat/perfstat parfile=spuexp.par
This creates a file called SPUEXP.DMP and the log file SPUEXP.LOG. If you want to load the data into another database, use the import command.
| See Also:
Oracle9i Database Utilities for more information on using export and import |
Purge unnecessary data from the PERFSTAT schema using the SPPURGE.SQL script. This deletes snapshots that fall between the begin and end snapshot IDs you specify.
|
Note: You should export the schema as a backup before running this script, either using your own export parameters or those provided in |
Purging can require the use of a large rollback segment, because all data relating to each snapshot ID to be purged is deleted. You can avoid rollback segment extension errors in one of two ways:
SET TRANSACTION USE ROLLBACK SEGMENT statement before running the SPPURGE.SQL script.
When you run SPPURGE, it displays the instance to which you are connected and the available snapshots. It then prompts you for the low snap ID and high snap ID. All snapshots that fall within this range are purged.
SQL> CONNECT perfstat/perfstat SQL> SET TRANSACTION USE ROLLBACK SEGMENT rbig; SQL> @sppurge Database Instance currently connected to ======================================== Instance DB Id DB Name Inst Num Name ----------- ---------- -------- ---------- 720559826 PERF 1 perf Snapshots for this database instance ==================================== Snap Snap Id Level Snapshot Started Host Comment ---------- ----- --------------------- --------------- ------------------- 1 5 30 Feb 2000 10:00:01 perfhost 2 5 30 Feb 2000 12:00:06 perfhost 3 5 01 Mar 2000 02:00:01 perfhost 4 5 01 Mar 2000 06:00:01 perfhost
Specify the Low Snap ID and High Snap ID range to purge ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ Enter value for losnapid: 1 Using 1 for lower bound. Enter value for hisnapid: 2 Using 2 for upper bound. Deleting snapshots 1 - 2 Purge of specified snapshot range complete. If you want to rollback the purge, it is still possible to do so. Exiting from SQL*Plus automatically commits the purge. SQL> -- end of example output
To purge in batch mode, you must assign values to the SQL*Plus variables that specify the low and high snapshot IDs to purge. The variables are:
SQL> CONNECT perfstat/perfstat SQL> DEFINE losnapid=1 SQL> DEFINE hisnapid=2 SQL> @sppurge
SPPURGE no longer prompts for this information.
To truncate all performance data indiscriminately, use SPTRUNC.SQL. This script truncates all statistics data gathered.
|
Note: Oracle Corporation recommends that you export the schema as a backup before running this script, either using your own export parameters or those provided in |
SQL> CONNECT perfstat/perfstat SQL> @sptrunc ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
|
Note: Running |
If you would like to continue, enter any string, followed by <return>. Enter value for anystring: entered - starting truncate operation Table truncated. <etc> Truncate operation complete.
The unique identifiers for a database instance used by Statspack are the DBID and the INSTANCE_NUMBER. When you use Oracle Real Application Clusters, the INSTANCE_NUMBER could change between startups (either because the INSTANCE_NUMBER parameter is set in the initialization file or because the instances are started in a different order).
Statspack uses the INSTANCE_NUMBER and the DBID to identify the instance's snapshot preferences, so you could have a different set of levels or thresholds being used when taking snapshots of an instance. This happens only under the following set of conditions:
The parameters differ only if the DBA explicitly modified them after installation, either by saving the specified values or by using the MODIFY_STATSPACK_PARAMETER procedure. To check whether any of the Statspack snapshot parameters are different for the instances, query the STATS$STATSPACK_PARAMETER table.
| See Also:
Oracle9i Real Application Clusters Administration for recommendations and issues with setting the |
To deinstall Statspack, connect as a user with SYSDBA privilege and run the following SPDROP script from SQL*Plus. For example:
SQL> CONNECT / AS SYSDBA SQL> @spdrop
This script calls two other scripts:
Check each of two output files produced (SPDTAB.LIS, SPDUSR.LIS) to ensure that the package was completely deinstalled.
The following scripts must be run as a user with SYSDBA privilege:
SPCREATE.SQL: Creates entire Statspack environment (calls SPCUSR.SQL, SPCTAB.SQL, SPCPKG.SQL)SPDROP.SQL: Drops entire Statspack environment (calls SPDTAB.SQL, SPDUSR.SQL)The following scripts are run as a user with SYSDBA privilege by the preceding calling scripts:
The following scripts are run as PERFSTAT by the preceding calling scripts:
SPCUSR.SQL: Creates the Statspack user (PERFSTAT)SPCTAB.SQL: Creates Statspack tablesSPCPKG.SQL: Creates the Statspack packageThe following scripts must be run as PERFSTAT:
SPREPORT.SQL: Generates a Statspack reportSPREPSQL.SQL: Generates a Statspack SQL report for the specific SQL hash value specifiedSPREPINS.SQL: Generates a Statspack report for the database and instance specifiedSPAUTO.SQL: Automates Statspack statistics collection (using DBMS_JOB)The following scripts must be run as SYSDBA:
SPUP90.SQL: Converts data from the 9.0 schema to the 9.2 schema. Back up the existing schema before running the upgrade.SPUP817.SQL must be run first.SPUP816.SQL must be run first, followed by SPUP817.SQL.The following scripts must be run as PERFSTAT:
SPPURGE.SQL: Purges a limited range of Snapshot IDs for a given database instance.SPTRUNC.SQL: Truncates all performance data in Statspack tables
|
Caution: Do not use this script unless you wan to remove all data in the schema you are using. You can choose to export the data as a backup before using this script. |
SPUEXP.PAR: An export parameter file supplied for exporting the whole PERFSTAT user.The SPDOC.TXT file contains instructions and documentation on the Statspack package.
|
 Copyright © 2000, 2002 Oracle Corporation. All Rights Reserved. |
|