Oracle Tuning Tools

Performance tools can examine execution strategies and can be used for SQL tuning, the tools can give you a good estimate of resource usage used by the queries.

The common tuning tools are

explain plan - lets you see the execution plan used by the query
autotrace - automatically produces an execution plan when you execute any SQL.
SQL trace - traces the execution of SQL statements
TKPROF - formats the trace files into a readable form
dbms_monitor - setup up end-to-end tracing
Statspack - performance diagnosis tool
AWR - automatic load repository

EXPLAIN Plan

The explain plan lets you see the execution plan used by the optimizer when it executed your query, it is idea when you are using hints as you can see if the hint is being used or not. The output of the explain plan goes into a table called plan_table, the explain plan out will tell you

The tables used in the query and the order in which they are accessed
The operations performed on the output of each step of the plan.
The specific access and join methods used for each table mentioned
The cost of each operation.

To create your own explain plan you must execute a oracle script which will create the plan table where the output of the explain plan is stored.

Setting up explain plan

@$ORACLE_HOME/rdbms/admin/utlxplan.sql

Creating the EXPLAIN plan

explain plan for
select * from employees where lname = 'valle';

Display explain plan

select * from plan_table (dbms_xplan.display);

Note: i have used the dbms_xplan package in the above statement

When reading the plan there are some principle to consider

Each step in the plan returns output in the form of a set of rows to the parent step
Read the plan outward starting from the line that is indented the most
If two operations are at the same level in terms of indentation, read the top one first
The numbering of the steps in the plan is misleading, start reading the plan output from inside out, read the most indented operations first

example

select statement
  hash join
    nested loops
      table access full department
      index unique scan employee_pk
    index fast full scan dept_id_pk

Oracle does a full table scan of the department table
Oracle performs an index unique scan of the employees table using its primary key index
Oracle performs a nested loop operation to join the rows from step 1 and 2
Oracle performs a fast full index scan of the department_id using its primary key
The final step oracle performs a hash join of the set from step3 and the rows resulting from step 4

Autotrace

The autotrace facility enables you to produce explain plans automatically when you execute a SQL statement. make sure that the plan table has been created (see above regarding running script utlxplan.sql).

Set privilege	grant plustrace to vallep; Note: you can also grant to public
Turn off	set autotrace off
Turn on	set autotrace on Note: this turns on explain and statistics
Turn on explain	set autotrace {on\|off\|trace[only]} explain
Turn on statistics	set autotrace {on\|off\|trace[only]} statistics
Traceonly	set autotrace traceonly

SQL Trace and TKPROF

SQL trace helps you trace the execution of SQL statements, TKPROF formats the output from the trace into a readable format. SQL trace enables you to track the following variables in a SQL statement

CPU and elapsed times
Parsed and execution counts for each SQL statement
Number of physical and logical reads
Execution plan for all the SQL statements
Library cache hit ratios

The explain plan gives you important information regarding the access path that the optimizer used, the SQL trace gives you the breakdown of the resources used CPU and I/O.

Collecting trace statistics imposes a performance penalty, you can control the collection of statistics by two parameters

Turn on statistics collection

alter system set statistics_level = typical;
alter system set statistics_level = all;

Turn off statistics collection

alter system set statistics_level = basic;

Turn on timed statistics

alter system set time_statistics = true;

Note: default is set to false, even if the statistics_level is set to basic (off) time statistics will be collected

You can turn on tracing for both the session or instance level, remember that turning on for the instance will use a lot of disk space and system resources.

Instance

alter system set sql_trace = true;
alter system set sql_trace = false;

Session

alter session set sql_trace = true;
alter session set sql_trace = false;

dbms_system.set_sql_trace_in_session(<sid><serial#>,true);
dbms_system.set_sql_trace_in_session(<sid><serial#>,false);

The trace will create a file located in the user_dump_dest and will format the file as db_name_ora_nnnnn.trc, the file size will generally be much larger in size that other files in this area.

TKPROF uses the trace file along with the following parameters

FILENAME	input trace file
EXPLAIN	the explain plan for the SQL statement
RECORD	creates a SQL script with all the nonrecursive SQL statements
WAITS	Records a summary of wait events
SORT	Presents sort data based on one or more items
TABLE	The name of the table TKPROF temporarily puts the executions plans
SYS	Enables and disables listing SQL statements issued by SYS
PRINT	Lists only a specified number of SQL statements instead of all statements
INSERT	Creates scripts that stores the trace information in the database.

TKPROF examples

tkprof finance_ora_16340.trc test.txt sys=no explain=y

Note: the output will be dumped into the test.txt file

End-to-End Tracing

By using a new attribute client_identifier you can trace a users session through multiple database sessions. you use the package dbms_monitor or OEM to setup the tracing. You require three attributes to trace the session

Client Identifier
Service Name
Combination of service name, module name and action name.

Below is an example on how to use the end-to-end tracing

setup the service name, module name, action name and the client id	dbms_monitor.serv_mod_act_trace_enable ( service_name => 'myservice', module_name => 'batch_job', action_name => 'batch_insert' ); dbms_monitor.client_id_trace_enable ( client_id => 'vallep' );
set the UID using a trigger	create or replace trigger logon_trigger after logon on database declare user_id varchar(64); begin select ora_login_user \|\| ':' \|\| sys_context('userenv','os_user') into user_id from dual; dbms_session.set_identifier(uid); end;
Obtain the sid and serial#	dbms_monitor.session_trace_enable ( session_id => 111, serial_num => 23, waits => true, binds => false );
Combine the multiple trace files in one file	c:\> trcsess output="vallep.trc" service="myservice" module="batch_job" action="batch_insert"
run TKPROF against the consolidated file	c:\> tkprof vallep.trc output=vallep.rpt sort=(EXELA, PRSELA, FCHELA) Note: there are many options to the sort parameter please see the Oracle documentation for more information. EXELA - elapsed time executing PRSELA - elapsed time parsing FCHELA - elapsed time fetching

Inefficient SQL

You can use the view v$sql to find inefficient SQL code, the view gathers important information regarding the disk reads and memory reads for a SQL statement. This view holds information on statements since startup, it also ages out older statements. The view will give you information on the following

rows_processed - total number of rows processed by the statement
sql_text - the SQL text of the statement (first 1,000 characters)
buffer_gets - total number of logical reads (high CPU use)
disk_reads - total number of disk reads (high I/O use)
sorts - number of sort for the statement (high sort ratios)
cpu_time - total parse and execution time
elapsed_time - elapsed parse and execution time
parse_calls - combined soft and hard parse calls
executions - number of times the statement was executed
loads - number of times the statement was flushed out of the shared pool then reloaded
sharable_memory - total memory used by the shared cursor
persistant_memory - total persistent memory used by the cursor
runtime_memory - total runtime memory used by the cursor

High disk reads	select sql_text, executions, buffer_gets, disk_reads from v$sql where buffer_gets > 100000 or disk_reads > 100000 order by buffer_gets + 100 * disk_reads desc;
High disk reads and parsed calls and row processed	select sql_text, rows_processed, buffer_gets, disk_reads, parsed_calls from v$sql where buffer_gets > 100000 or disk_reads > 100000 order by buffer_gets + 100 * disk_reads desc;
TOP 5 CU time and elapsed time	select sql_text, executions, round(elapsed_time/1000000, 2) elapsed_seconds, round(cpu_time/1000000, 2) cpu_secs from (select * from v£sql order by elapsed_time desc) where rownum < 6;

SQL Tuning Advisor

When you have identified bad SQL, you can use the SQL tuning advisor to perform an in depth analysis to come up with a better execution plan.

See the Advisors for more detailed information.

Statspack

Statspack is a diagnostic tool that captures and stores the V$ table information and allows you generate reports at a later date, although it has been replaced with AWR many dba's still use this tool, I will only give a brief overview of this tool as you will more than likely start using the AWR tool as its now Oracle preferred method of collecting statistics.

To install statspack you simply run the following as sys with sysdba privilege "$oracle_home\rdbms\admin\spcreate.sql", you can use "spdrop.sql" to remove it. The install script will ask you three pieces of information

The PERFSTAT user password
The default tablespace that you will use for the PERFSTAT schema
The default temporary tablespace that you will use for the PERFSTAT schema

Once the installation has finished check the "spcpkg.lis" file for any errors, the below commands can be run to obtain snapshots and generate reports.

Create snapshot

exec statspack.snap

Run report

@$oracle_home\rdbms\admin\spreport.sql

Note: when you run the report it will ask for two snapshot points to compare.

AWR

See AWR for more information on how to setup and how to run reports.