Performance Tuning Enhancements in Oracle Database 10g
Performance Tuning Enhancements in Oracle Database 10g
Oracle 10g includes many performance tuning enhancements including:
- Automatic Performance Diagnostic and Tuning Features
- Automatic Shared Memory Management
- Wait Model Improvements
- Automatic Optimizer Statistics Collection
- Dynamic Sampling
- CPU Costing
- Optimizer Hints
- Rule Based Optimizer Obsolescence
- Tracing Enhancements
- SAMPLE Clause Enhancements
- Hash Partitioned Global Indexes
Automatic Performance Diagnostic and Tuning Features
Oracle 10g includes several features related to automatic performance diagnostics and tuning.
- Automatic Optimizer Statistics Collection – The name says it all.
- Automatic Workload Repository (AWR) – An extended version of the STATSPACK repository that is the heart of all the new diagnostics and tuning features.
- Automatic Database Diagnostic Monitoring (ADDM) – An automatic diagnostics and tuning tool which uses the information stored in the AWR.
- Automatic SQL Tuning Advisor – A built in SQL tuning feature.
Most of these features are beyond the scope of this article and as such will be dealt with in separate aticles.
Automatic Shared Memory Management
Automatic Shared Memory Management puts Oracle in control of allocating memory within the SGA. The
SGA_TARGET parameter sets the amount of memory available to the SGA. This parameter can be altered dynamically up to a maximum of the
SGA_MAX_SIZE parameter value. Provided the
STATISTICS_LEVEL is set to
ALL and the
SGA_TARGET is set to a value other than “0” Oracle will control the memory pools which would otherwise be controlled by the following parameters.
DB_CACHE_SIZE(default block size)
If these parameters are set to a non-zero value they represent the minimum size for the pool. These minimum values may be necessary if you experience application errors when certain pool sizes drop below a specific threshold.
The following parameters must be set manually and take memory from the quota allocated by the
DB_nK_CACHE_SIZE(non-default block size)
Wait Model Improvements
A number of views have been updated and added to improve the wait model. The updated views include the following.
The new views include the following.
The following are some examples of how these updates can be used.
V$EVENT_NAME view has had three new columns added (
WAIT_CLASS) which indicate the class of the event. This allows easier aggregation of event details.
-- Display time waited for each wait class. SELECT a.wait_class, sum(b.time_waited)/1000000 time_waited FROM v$event_name a JOIN v$system_event b ON a.name = b.event GROUP BY a.wait_class; WAIT_CLASS TIME_WAITED --------------------------- ----------- Application .013388 Commit .003503 Concurrency .009891 Configuration .003489 Idle 232.470445 Network .000432 Other .025698 System I/O .095651 User I/O .109552 9 rows selected.
V$SESSION view has had several columns added that include blocking session and wait information. The wait information means it’s no longer necessary to join to
V$SESSION_WAIT to get wait information for a session.
-- Display blocked session and their blocking session details. SELECT sid, serial#, blocking_session_status, blocking_session FROM v$session WHERE blocking_session IS NOT NULL; no rows selected -- Display the resource or event the session is waiting for. SELECT sid, serial#, event, (seconds_in_wait/1000000) seconds_in_wait FROM v$session ORDER BY sid; SID SERIAL# EVENT SECONDS_IN_WAIT ---------- ---------- ----------------------------------- --------------- 131 20 SQL*Net message from client .000015 133 501 wakeup time manager .000138 134 28448 SQL*Net message to client 0 135 4 queue messages .000003 137 8 SQL*Net message from client .000132 .... 167 1 rdbms ipc message 0 168 1 rdbms ipc message 0 169 1 rdbms ipc message .079485 170 1 pmon timer .092645 29 rows selected.
V$SESSION_WAIT_CLASS view allows you to see the session wait information broken down by wait class for each session.
-- Display session wait information by wait class. SELECT * FROM v$session_wait_class WHERE sid = 134; SID SERIAL# WAIT_CLASS_ID WAIT_CLASS# WAIT_CLASS TOTAL_WAITS TIME_WAITED ---------- ---------- ------------- ----------- ------------------- ----------- ----------- 134 28448 4217450380 1 Application 2 0 134 28448 3875070507 4 Concurrency 1 2 134 28448 2723168908 6 Idle 57 392127 134 28448 2000153315 7 Network 68 5 4 rows selected.
V$SESSION_WAIT_HISTORY view shows historical wait information which allows you to identify issues after the session has ended.
Automatic Optimizer Statistics Collection
By default Oracle 10g automatically gathers optimizer statistics using a scheduled job called
GATHER_STATS_JOB. By default this job runs within a maintenance windows between 10 P.M. to 6 A.M. week nights and all day on weekends. The job calls the
DBMS_STATS.GATHER_DATABASE_STATS_JOB_PROC internal procedure which gathers statistics for tables with either empty or stale statistics, similar to the
DBMS_STATS.GATHER_DATABASE_STATS procedure using the
GATHER AUTO option. The main difference is that the internal job prioritizes the work such that tables most urgently requiring statistics updates are processed first.
In some cases automatically gathering statistics can cause problems. Highly volatile tables and load tables may have their statistics gathered when there is an unrepresentative number of rows present. These situations can be avoided by using one of two methods.
- The current statistics can be deleted and locked to prevent
DBMS_STATSfrom gathering new statistics. If the
OPTIMIZER_DYNAMIC_SAMPLINGparameter is set to 2 (the default) or higher the necessary statistics will be gathered as part of the query optimization stage (See Dynamic Sampling).
BEGIN DBMS_STATS.delete_table_stats('MY_SCHEMA','LOAD_TABLE'); DBMS_STATS.lock_table_stats('MY_SCHEMA','LOAD_TABLE'); END; /
- The statistics can be gathered then locked at a time when the table contains the appropriate data.
BEGIN DBMS_STATS.gather_table_stats('MY_SCHEMA','LOAD_TABLE'); DBMS_STATS.lock_table_stats('MY_SCHEMA','LOAD_TABLE'); END; /
System statistics and statistics for fixed object, such as dynamic performance tables, are not gathered automatically.
Dynamic sampling enables the server to improve performance by:
- Estimate single-table predicate selectivities where available statistics are missing or may lead to bad estimations.
- Estimate statatistics for tables and indexes with missing statistics.
- Estimate statatistics for tables and indexes with out of date statistics.
Dynamic sampling is controled by the
OPTIMIZER_DYNAMIC_SAMPLING parameter which accepts values from “0” (off) to “10” (agressive sampling) with a default value of “2”. At compile-time Oracle determines if dynamic sampling would improve query performance. If so it issues recursive statements to estimate the necessary statistics. Dynamic sampling can be beneficial when:
- The sample time is small compared to the overall query execution time.
- Dynamic sampling results in a better performing query.
- The query may be executed multiple times.
In addition to the
OPTIMIZER_DYNAMIC_SAMPLINGsystem parameter the dynamic sampling level can be set using the
DYNAMIC_SAMPLING optimizer hint for specific queries like the following.
SELECT /*+ dynamic_sampling(emp 10) */ empno, ename, job, sal FROM emp WHERE deptno = 30;
The results of dynamic sampling are repeatable provided no rows are inserted, updated or deleted from the sampled table. The
OPTIMIZER_FEATURES_ENABLE parameter will turns off dynamic sampling if it is set to a version earlier than 9.2.0.
By default the cost model for the optimizer is now CPU+I/O, with the cost unit as time.
SPREAD_MIN_ANALYSIS– Specifies analysis options for spreadsheets.
USE_NL_WITH_INDEX– Specifies a nested loops join.
QB_NAME– Specifies a name for a query block.
NO_QUERY_TRANSFORMATION– Prevents the optimizer performing query transformations.
NO_STAR_TRANSFORMATION– Excludes specific operations from the query plan.
INDEX_SS_DESC– Excludes range scans from the query plan.
- Hints that specify table names have been expanded to accept Global Table Hints. This allows a base table within a view to be specified using the “view-name.table-name” syntax.
- Hints that specify index names have been expanded to accept Complex Index Hints. This allows an index to be specified using the “(table-name.column-name)” syntax instead of the index name.
- Some hints can now optionally accept a query block parameter.
Rule Based Optimizer Obsolescence
The Rule Based Optimizer (RBO) is now obsolete in Oracle 10g. The functionality is still present but no new functionality has been included in it and it is no longer supported by Oracle. It is only present to provide backwards compatibility during the migration to the query optimizer (Cost Based Optimizer). The results of this osolescence are as follows.
RULEoptions for the
OPTIMIZER_MODEparameter still exist but are no longer supported.
- The default value for the
RULEoptimizer hints still exist but are no longer supported.
- Code requiring the RBO must be migrated to use the query optimizer.
The Oracle Trace functionality has been removed from Oracle 10g. Instead the SQL Trace and TKPROF functionality should be used.
In multi-tier environments where statements are passed to different sessions by the application server it can become difficult to trace an individual process from start to finish. To solve this problem Oracle have introduced End to End Application Tracing which allows a client process to be identified via the client identifier rather than the typical session id. Each piece of trace information is linked to the following information.
- Client Identifier – Specifies the “real” end user. Set using the
- Service – Specifies a group of related applications. Created using the
- Module – Specifies a functional area or feature of an application. Set using the
- Action – Specifies the current action (INSERT, UPDATE, DELETE etc.) within the current module. Set using the
End to end tracing can be managed via Enterprise Manager or a set of APIs and views. Here are some examples of how to enable and disable to various types of tracing.
BEGIN -- Enable/Disable Client Identifier Trace. DBMS_MONITOR.client_id_trace_enable (client_id => 'my_id'); DBMS_MONITOR.client_id_trace_disable (client_id => 'my_id'); -- Enable/Disable Service, Module and Action Trace (various overloads). DBMS_MONITOR.serv_mod_act_trace_enable ( service_name => 'my_service'); DBMS_MONITOR.serv_mod_act_trace_enable ( service_name => 'my_service', module_name => 'my_module'); DBMS_MONITOR.serv_mod_act_trace_enable ( service_name => 'my_service', module_name => 'my_module', action_name => 'INSERT'); DBMS_MONITOR.serv_mod_act_trace_disable ( service_name => 'my_service', module_name => 'my_module', action_name => 'INSERT'); -- Enable/Disable Session Trace (various overloads). DBMS_MONITOR.session_trace_enable; DBMS_MONITOR.session_trace_enable ( session_id => 15, serial_num => 1234); DBMS_MONITOR.session_trace_disable ( session_id => 15, serial_num => 1234); END; /
Once the trace files are produced the
trcsess command line utility can be used to filter out the relevant data from multiple files. The utility accepts the following parameters:
OUTPUT– Specifies the name of the consolidated trace file.
SESSION– Consolidates the file based on the specified session id (
CLIENT_ID– Consolidates the file based on the specified client identifier (
SERVICE– Consolidates the file based on the specified service (
MODULE– Consolidates the file based on the specified module (
ACTION– Consolidates the file based on the specified action (
TRACE_FILES– A space separated list of trace files to be searched. If omitted all files in the local directory are searched.
At lease one of the search criteria must be specified. If more than one is specified only trace that matches all the criteria is consolidated. Examples of
trcsess usage are shown below.
# Search all files for this session. trcsess output=session.trc session=144.2274 # Search the specified files for this client identifier. trcsess output=client.trc client_id=my_id db10g_ora_198.trc db10g_ora_206.trc # Search the specified files for this service, module and action combination. trcsess output=client.trc service=my_service module=my_module action=INSERT db10g_ora_198.trc db10g_ora_206.trc
Once the consolidated trace file is produced it can be processed by the
TKPROF utility like any other SQL Trace file.
By default statistics are gathered at the session level. The
DBMS_MONITOR package allows this to be altered to follow the client identifier, service or combinations of the service, module and action.
BEGIN -- Enable/Disable Client Identifier Statistics. DBMS_MONITOR.client_id_stat_enable (client_id => 'my_id'); DBMS_MONITOR.client_id_stat_disable (client_id => 'my_id'); -- Enable/Disable Service, Module and Action Statistics (various overloads). DBMS_MONITOR.serv_mod_act_stat_enable ( service_name => 'my_service'); DBMS_MONITOR.serv_mod_act_stat_enable ( service_name => 'my_service', module_name => 'my_module'); DBMS_MONITOR.serv_mod_act_stat_enable ( service_name => 'my_service', module_name => 'my_module', action_name => 'INSERT'); DBMS_MONITOR.serv_mod_act_stat_disable ( service_name => 'my_service', module_name => 'my_module', action_name => 'INSERT'); END; /
The gathered statistics can be displayed using the following views.
DBA_ENABLED_AGGREGATIONS– Accumulated global statistics.
V$CLIENT_STATS– Accumulated statistics for the specified client identifier.
V$SERVICE_STATS– Accumulated statistics for the specified service.
V$SERV_MOD_ACT_STATS– Accumulated statistics for the specified service, module and action combination.
V$SVCMETRIC– Accumulated statistics for elapsed time of database calls and CPU usage.
SAMPLE Clause Enhancements
SAMPLE clause allows a query to return a limited sample of data by specifying a percentage of rows or blocks to scan. This clause can now be present in complex queries.
-- Query 10% or rows. SELECT e.empno, e.ename, d.dname FROM emp SAMPLE (10) e JOIN dept d ON e.deptno = d.deptno; -- Query 10% of blocks. SELECT e.empno, e.ename, d.dname FROM emp SAMPLE BLOCK (10) e JOIN dept d ON e.deptno = d.deptno;
Hash Partitioned Global Indexes
Support for hash partitioned global indexes has been added in Oracle 10g which can improve performance when a small number of leaf blocks are experiencing high levels of contention. The syntax for creating of a hash paritioned global index is shown below.
CREATE INDEX hgidx ON tab (c1,c2,c3) GLOBAL PARTITION BY HASH (c1,c2) (PARTITION p1 TABLESPACE tbs_1, PARTITION p2 TABLESPACE tbs_2, PARTITION p3 TABLESPACE tbs_3, PARTITION p4 TABLESPACE tbs_4);