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.
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 TYPICAL or 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)SHARED_POOL_SIZELARGE_POOL_SIZEJAVA_POOL_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
SGA_TARGET parameter.DB_KEEP_CACHE_SIZEDB_RECYCLE_CACHE_SIZEDB_nK_CACHE_SIZE(non-default block size)STREAMS_POOL_SIZELOG_BUFFER
Wait Model Improvements
A number of views have been updated and added to improve the wait model. The updated views include the following.
V$EVENT_NAMEV$SESSIONV$SESSION_WAIT
The new views include the following.
V$ACTIVE_SESSION_HISTORYV$SESSION_WAIT_HISTORYV$SESS_TIME_MODELV$SYS_TIME_MODELV$SYSTEM_WAIT_CLASSV$SESSION_WAIT_CLASSV$EVENT_HISTOGRAMV$FILE_HISTOGRAMV$TEMP_HISTOGRAM
The following are some examples of how these updates can be used.
The
V$EVENT_NAME view has had three new columns added (WAIT_CLASS_ID, WAIT_CLASS# and WAIT_CLASS) which indicate the class of the event. This allows easier aggregation of event details.-- Display time waited for each wait class.
SELECT a.wait_class, sum(b.time_waited)/1000000 time_waited
FROM v$event_name a
JOIN v$system_event b ON a.name = b.event
GROUP BY a.wait_class;
WAIT_CLASS TIME_WAITED
--------------------------- -----------
Application .013388
Commit .003503
Concurrency .009891
Configuration .003489
Idle 232.470445
Network .000432
Other .025698
System I/O .095651
User I/O .109552
9 rows selected.The
V$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$session
ORDER BY sid;
SID SERIAL# EVENT SECONDS_IN_WAIT
---------- ---------- ----------------------------------- ---------------
131 20 SQL*Net message from client .000015
133 501 wakeup time manager .000138
134 28448 SQL*Net message to client 0
135 4 queue messages .000003
137 8 SQL*Net message from client .000132
....
167 1 rdbms ipc message 0
168 1 rdbms ipc message 0
169 1 rdbms ipc message .079485
170 1 pmon timer .092645
29 rows selected.The
V$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$session_wait_class
WHERE sid = 134;
SID SERIAL# WAIT_CLASS_ID WAIT_CLASS# WAIT_CLASS TOTAL_WAITS TIME_WAITED
---------- ---------- ------------- ----------- ------------------- ----------- -----------
134 28448 4217450380 1 Application 2 0
134 28448 3875070507 4 Concurrency 1 2
134 28448 2723168908 6 Idle 57 392127
134 28448 2000153315 7 Network 68 5
4 rows selected.The
V$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 theOPTIMIZER_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
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.
OPTIMIZER_DYNAMIC_SAMPLING system 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.CPU Costing
By default the cost model for the optimizer is now CPU+I/O, with the cost unit as time.
Optimizer Hints
New hints:
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_USE_NL,NO_USE_MERGE,NO_USE_HASH,NO_INDEX_FFS,NO_INDEX_SSandNO_STAR_TRANSFORMATION- Excludes specific operations from the query plan.INDEX_SS,INDEX_SS_ASC,INDEX_SS_DESC- Excludes range scans from the query plan.
Updated hints:
- 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.
Renamed hints:
NO_PARALLEL- FormallyNOPARALLEL.NO_PARALLEL_INDEX- FormallyNOPARALLEL_INDEX.NO_REWRITE- FormallyNOREWRITE.
Deprecated hints:
AND_EQUALHASH_AJMERGE_AJNL_AJHASH_SJNL_SJEXPAND_GSET_TO_UNIONORDERED_PREDICATESROWIDSTAR
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.
- The
CHOOSEandRULEoptions for theOPTIMIZER_MODEparameter still exist but are no longer supported. - The default value for the
OPTIMIZER_MODEparameter isALL_ROWS. - The
CHOOSEandRULEoptimizer hints still exist but are no longer supported. - Code requiring the RBO must be migrated to use the query optimizer.
Tracing Enhancements
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
DBMS_SESSION.SET_IDENTIFIERprocedure. - Service - Specifies a group of related applications. Created using the
DBMS_SERVICE.CREATE_SERVICEprocedure. - Module - Specifies a functional area or feature of an application. Set using the
DBMS_APPLICATION_INFO.SET_MODULEprocedure. - Action - Specifies the current action (INSERT, UPDATE, DELETE etc.) within the current module. Set using the
DBMS_APPLICATION_INFO.SET_ACTIONprocedure.
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 (SID.SERIAL#columns fromV$SESSION).CLIENT_ID- Consolidates the file based on the specified client identifier (CLIENT_IDENTIFIERcolumn fromV$SESSION).SERVICE- Consolidates the file based on the specified service (SERVICE_NAMEcolumn fromV$SESSION).MODULE- Consolidates the file based on the specified module (MODULEcolumn fromV$SESSION).ACTION- Consolidates the file based on the specified action (ACTIONcolumn fromV$SESSION).TRACE_FILES- A space separated list of trace files to be searched. If omitted all files in the local directory are searched.
At lease one of the search criteria must be specified. If more than one is specified only trace that matches all the criteria is consolidated. Examples of
trcsess usage are shown below.# Search all files for this session. trcsess output=session.trc session=144.2274 # Search the specified files for this client identifier. trcsess output=client.trc client_id=my_id db10g_ora_198.trc db10g_ora_206.trc # Search the specified files for this service, module and action combination. trcsess output=client.trc service=my_service module=my_module action=INSERT db10g_ora_198.trc db10g_ora_206.trc
Once the consolidated trace file is produced it can be processed by the
TKPROF utility like any other SQL Trace file.By default statistics are gathered at the session level. The
DBMS_MONITOR package allows this to be altered to follow the client identifier, service or combinations of the service, module and action.BEGIN
-- Enable/Disable Client Identifier Statistics.
DBMS_MONITOR.client_id_stat_enable (client_id => 'my_id');
DBMS_MONITOR.client_id_stat_disable (client_id => 'my_id');
-- Enable/Disable Service, Module and Action Statistics (various overloads).
DBMS_MONITOR.serv_mod_act_stat_enable (
service_name => 'my_service');
DBMS_MONITOR.serv_mod_act_stat_enable (
service_name => 'my_service',
module_name => 'my_module');
DBMS_MONITOR.serv_mod_act_stat_enable (
service_name => 'my_service',
module_name => 'my_module',
action_name => 'INSERT');
DBMS_MONITOR.serv_mod_act_stat_disable (
service_name => 'my_service',
module_name => 'my_module',
action_name => 'INSERT');
END;
/The gathered statistics can be displayed using the following views.
DBA_ENABLED_AGGREGATIONS- Accumulated global statistics.V$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
The
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);
