一. 官網的說明
http://download.oracle.com/docs/cd/E11882_01/server.112/e17110/initparams042.htm#REFRN10025
1.1 CURSOR_SHARING
| Property |
Description |
| Parameter type |
String |
| Syntax |
CURSOR_SHARING = { SIMILAR | EXACT | FORCE } |
| Default value |
EXACT |
| Modifiable |
ALTER SESSION, ALTER SYSTEM |
| Basic |
No |
CURSOR_SHARING determines what kind of SQL statements can share the same cursors.
Values:
（1）FORCE
Allows the creation of a new cursor if sharing an existing cursor, or if the cursor plan is not optimal.
（2）SIMILAR
Causes statements that may differ in some literals, but are otherwise identical, to share a cursor, unless the literals affect either the meaning of the statement or the degree to which the plan is optimized.
（3）EXACT
Only allows statements with identical text to share the same cursor.
--只有SQL 語句完全相同的情況下，才會使用相同的cursor，即執行計划。
Notes:
（1）If you set CURSOR_SHARING, then Oracle recommends the FORCE setting unless you are in a DSS environment. FORCE limits the growth of child cursors that can occur when the setting is SIMILAR.
（2）The value of the CURSOR_SHARING parameter has performance implications. Refer to Oracle Database Performance Tuning Guide before setting this parameter.
1.2 When to Set CURSOR_SHARING to a Nondefault Value
The best practice is to write sharable SQL and use the default of EXACT for CURSOR_SHARING. However, for applications with many similar statements, setting CURSOR_SHARING can significantly improve cursor sharing, resulting in reduced memory usage, faster parses, and reduced latch contention. Consider this approach when statements in the shared pool differ only in the values of literals, and when response time is poor because of a very high number of library cache misses.
Setting CURSOR_SHARING to FORCE or SIMILAR has the following drawbacks:
（1）The database must perform extra work during the soft parse to find a similar statement in the shared pool.
（2）There is an increase in the maximum lengths (as returned by DESCRIBE) of any selected expressions that contain literals in a SELECT statement. However, the actual length of the data returned does not change.
（3）Star transformation is not supported.
（4）If stored outlines were generated with CURSOR_SHARING set to EXACT, then the database does not use stored outlines generated with literals. To avoid this problem, generate outlines with CURSOR_SHARING set to FORCE or SIMILAR and use the CREATE_STORED_OUTLINES parameter.
When deciding whether to set CURSOR_SHARING to FORCE or SIMILAR, consider the performance implications of each setting.
When CURSOR_SHARING is set to FORCE, the database uses one parent cursor and one child cursor for each distinct SQL statement. The database uses the same plan for each execution of the same statement.
When set to SIMILAR, database behavior depends on the presence of histograms:
（1）Histogram absent for column with system-generated bind value
Only one parent cursor and one child cursor exists for each distinct SQL statement. In this case, all executions of a SQL statement use the same plan.
（2）Histogram present for column with system-generated bind value
If the same SQL statement is executed multiple times, each execution has its own child cursor. In this case, the database peeks at bind variable values and create a new child cursor for each distinct value. Thus, each statement execution uses a plan based on the specific literals in the statement.
For example, consider the following statement:
SELECT * FROM hr.employees WHERE employee_id = 101
If FORCE is used, or if SIMILAR is used when no histogram exists, then the database optimizes this statement as if it contained a bind variable and uses bind peeking to estimate cardinality. Statements that differ only in the bind variable share the same execution plan.
If SIMILAR is used, and if a histogram does exist, then the database does not treat the statement as if a bind variable were used. The same query for a different employee may not use the same plan.
If you set CURSOR_SHARING, then Oracle recommends the FORCE setting unless you are in a DSS environment. FORCE limits the growth of child cursors that can occur when the setting is SIMILAR.
Also, function-based indexes may not work when using SIMILAR because the database converts index parameters to bind variables.
For example, if the index is SUBSTR(id,1,3), then the database converts it to SUBSTR("ID",:SYS_B_0,:SYS_B_1)=:id, rendering the index invalid.
Oracle 綁定變量 詳解
http://blog.csdn.net/tianlesoftware/archive/2010/09/01/5856430.aspx
二. 測試
2.1 cursor_sharing=exact，這是cursor_sharing的默認值
2.1.1 查看cursor_sharing 值
SYS@anqing2(rac2)> show parameter cursor_sharing
NAME TYPE VALUE
------------------------- -------------------- ---------------------
cursor_sharing string EXACT
2.1.2 查看當前硬解析值
SYS@anqing2(rac2)> select name,value from v$sysstat where name like '%parse%';
NAME VALUE
------------------------------ ----------
parse time cpu 1882056
parse time elapsed 2648194
parse count (total) 12780229
parse count (hard) 9890010（硬解析次數）
parse count (failures) 71
2.1.3 執行一條select 語句，然后查看硬解析次數
SYS@anqing2(rac2)> select * from ta where id=168;
ID NAME
---------- ------------------------------
168 dave
SYS@anqing2(rac2)> select name,value from v$sysstat where name like '%parse%';
NAME VALUE
------------------------------ ----------
parse time cpu 1882061
parse time elapsed 2648196
parse count (total) 12780360
parse count (hard) 9890021
parse count (failures) 71
-- 這里硬解析的次數加一，因為之前SQL 沒有解析過，所以需要進行硬解析之后才能執行。
2.1.4 執行與之前類似的SQL，謂詞值不一樣
SYS@anqing2(rac2)> select * from ta where id=198;
ID NAME
---------- ------------------------------
198 dave
SYS@anqing2(rac2)> select name,value from v$sysstat where name like '%parse%';
NAME VALUE
------------------------------ ----------
parse time cpu 1882061
parse time elapsed 2648196
parse count (total) 12780482
parse count (hard) 9890022
parse count (failures) 71
-- 硬解析次數又加1了，沒有重用之前的執行計划
2.1.5 執行相同的SQL 語句
SYS@anqing2(rac2)> select * from ta where id=198;
ID NAME
---------- ------------------------------
198 dave
SYS@anqing2(rac2)> select name,value from v$sysstat where name like '%parse%';
NAME VALUE
------------------------------ ----------
parse time cpu 1882061
parse time elapsed 2648196
parse count (total) 12780543
parse count (hard) 9890022
parse count (failures) 71
-- 測試硬解析沒有變化。 重用之前的cursor。
總結：
在這種模式下,只有SQL 語句完全相同的情況下，才會使用相同的cursor，即執行計划。
這種模式下，表有統計信息和沒有統計信息的執行計划是有出入的。 所以該模式下的表，需要定期的去收集統計信息。
2.2 cursor_sharing=force
--修改cursor_sharing 值
SYS@anqing2(rac2)> alter session set cursor_sharing='force';
Session altered.
SYS@anqing2(rac2)> show parameter cursor_sharing
NAME TYPE VALUE
------------------------- --------------------- ---------------
cursor_sharing string force
--查看硬解析次數
SYS@anqing2(rac2)> select name,value from v$sysstat where name like '%parse%';
NAME VALUE
------------------------------ ----------
parse time cpu 1882075
parse time elapsed 2648219
parse count (total) 12782090
parse count (hard) 9890067 （硬解析次數）
parse count (failures) 71
-- select 查詢
SYS@anqing2(rac2)> select * from ta where id=88;
ID NAME
---------- ------------------------------
88 dave
SYS@anqing2(rac2)> select name,value from v$sysstat where name like '%parse%';
NAME VALUE
------------------------------ ----------
parse time cpu 1882075
parse time elapsed 2648219
parse count (total) 12782215
parse count (hard) 9890068 -- 硬解析次數加一
parse count (failures) 71
-- 執行相同的select，但謂詞值不一樣
SYS@anqing2(rac2)> select * from ta where id=99;
ID NAME
---------- ------------------------------
99 dave
SYS@anqing2(rac2)> select name,value from v$sysstat where name like '%parse%';
NAME VALUE
------------------------------ ----------
parse time cpu 1882075
parse time elapsed 2648219
parse count (total) 12782285
parse count (hard) 9890068
parse count (failures) 71
--注意，這里的硬解析次數沒有變化，這個就是force 的作用。只要sql語句相同，不管謂詞值是否相同，都會當成相同的sql，重用之前的cursor，不會進行硬解析。
-- 查看child cursor 信息
SYS@anqing2(rac2)> select sql_text,child_number from v$sql where sql_text like 'select * from ta where%';
SQL_TEXT CHILD_NUMBER
---------------------------------------- ------------
select * from ta where id=:"SYS_B_0" 0
select * from ta where id=:"SYS_B_0" 1
select * from ta where id=:"SYS_B_0" 2
注意：
對於相同的SQL，oracle 在這里將不同的謂詞值改成了變量，這樣SQL_TEXT 就相同，正常情況下，應該使用同一個cursor，即執行計划，但是在我上面的查詢中，Oracle 並沒有重用，而是重新生成了一個child_cursor. 這就說明Oracle 認為這個cursor 並不是最優的，所有重新生成了一個。
可以通過如下SQL 查看為什么沒有重用之前的cursor：
SQL>select * from v$sql_shared_cursor where sql_id='c9swtz4spq3xz';
如果這里有Y，就是導致不能重用的原因。
Oracle 性能相關的幾個 視圖 和 參數
http://blog.csdn.net/tianlesoftware/archive/2010/09/06/5867276.aspx
總結：
Allows the creation of a new cursor if sharing an existing cursor, or if the cursor plan is not optimal.
When CURSOR_SHARING is set to FORCE, the database uses one parent cursor and one child cursor for each distinct SQL statement. The database uses the same plan for each execution of the same statement.
FORCE limits the growth of child cursors that can occur when the setting is SIMILAR.
當cursor_sharing 設置為force時， Oracle 會把相同SQL的不同謂詞值轉換成變量，這樣SQL_TEXT就看上去一樣。 Oracle 就會使用一個相同的cursor。 這樣他們的執行計划也是一樣的。
當Oracle 認為存在的cursor 不是最優的時候，就會重新創建一個child cursor，而不重用之前的已經存在cursor。 可以通過v$sql_shared_cursor 查看為什么沒有重用。
這樣就會和我們上面查詢的一樣，會有多個child cursor，但是他們的parent cursor是一樣的。這個child cursor 不是無限增常的，force 和similar 都會限制child cursor 的增長。
2.3 cursor_sharing=similar
在這種模式下，對表做統計和沒做統計分析是不一樣的。有關統計信息，參考我的Blog：
Oracle Statistic 統計信息 小結
http://blog.csdn.net/tianlesoftware/archive/2009/10/14/4668723.aspx
2.3.1 沒有統計信息，沒有直方圖的情況
-- 查詢ta的統計信息
SYS@anqing2(rac2)> exec dbms_stats.delete_table_stats('SYS','TA');
PL/SQL procedure successfully completed.
-- 修改cursor_sharing 模式
SYS@anqing2(rac2)> alter session set cursor_sharing='similar';
Session altered.
SYS@anqing2(rac2)> show parameter cursor_sharing
NAME TYPE VALUE
---------------------- ------------------ ----------------
cursor_sharing string similar
在similer模式下，且沒有直方圖，等於force 模式。 看測試：
-- 查看硬解析次數
SYS@anqing2(rac2)> select name,value from v$sysstat where name like '%parse%';
NAME VALUE
------------------------------ ----------
parse time cpu 1882222
parse time elapsed 2648406
parse count (total) 12793472
parse count (hard) 9890538
parse count (failures) 74
--做一次select 查詢
SYS@anqing2(rac2)> select * from ta where id=238;
ID NAME
---------- ------------------------------
238 dave
SYS@anqing2(rac2)> select name,value from v$sysstat where name like '%parse%';
NAME VALUE
------------------------------ ----------
parse time cpu 1882222
parse time elapsed 2648406
parse count (total) 12793476
parse count (hard) 9890539 硬解析加一
parse count (failures) 74
-- 在次select
SYS@anqing2(rac2)> select * from ta where id=2397;
ID NAME
---------- ----------
2397 dave
SYS@anqing2(rac2)> select name,value from v$sysstat where name like '%parse%';
NAME VALUE
------------------------------ ----------
parse time cpu 1882337
parse time elapsed 2648558
parse count (total) 12823605
parse count (hard) 9890751
parse count (failures) 75
-- 這里還是有增加硬解析，有點奇怪
-- 查看對應的SQL
SYS@anqing2(rac2)> select sql_text,child_number from v$sql where sql_text like 'select * from ta where%';
SQL_TEXT CHILD_NUMBER
---------------------------------------- ------------
select * from ta where id=238 0
select * from ta where id=2396 0
select * from ta where id=2391 0
select * from ta where id=2397 0
這里並沒有將謂詞值轉換成變量，也就是說similar 沒有生效，沒有達到force的效果。
在如下Blog 上有對Force 沒有生效的一個討論：
http://forums.oracle.com/forums/message.jspa?messageID=3569923#3569923
摘取一段如下：
Interesting. You may need to flush the shared pool "harder"... Try your testcase again, but issue the ALTER SYSTEM FLUSH SHARED_POOL twice in a row before changing the cursor_sharing setting. It looks like that the statement doesn't show up in V$SQLAREA any longer but still seems not to be flushed entirely and I can find some left-overs of it in the underlying X$KGL library cache fixed tables with the same hash value. After the second "flush shared pool" these are gone and then the cursor sharing works as expected.
解決方法就是在指定cursor_sharing 之前，執行2次flush shared_pool. 測試一下看看。
SYS@anqing2(rac2)> alter system flush shared_pool;
System altered.
SYS@anqing2(rac2)> alter system flush shared_pool;
System altered.
SYS@anqing2(rac2)> alter session set cursor_sharing=similar;
Session altered.
SYS@anqing2(rac2)> select name,value from v$sysstat where name like '%parse%';
NAME VALUE
------------------------------ ----------
parse count (hard) 9890936
SYS@anqing2(rac2)> select * from ta where id=232;
ID NAME
---------- ----------
232 dave
SYS@anqing2(rac2)> select name,value from v$sysstat where name like '%parse%';
NAME VALUE
------------------------------ ----------
parse count (hard) 9890946
--從硬解析來看，數量有增加
-- 在查詢一次
SYS@anqing2(rac2)> select * from ta where id=23218;
ID NAME
---------- ----------
23218 dave
SYS@anqing2(rac2)> select name,value from v$sysstat where name like '%parse%';
NAME VALUE
------------------------------ ----------
parse count (hard) 9890954
--硬解析又增加了，如果說只有這一條SQL執行，不會增加硬解析。 這個是其他的SQL 帶來的硬解析，我們可以查看v$sql驗證：
SYS@anqing2(rac2)> select sql_text,child_number,executions from v$sql where sql_text like 'select * from ta where%';
SQL_TEXT CHILD_NUMBER EXECUTIONS
---------------------------------------- ------------ ----------
select * from ta where id=:"SYS_B_0" 0 2
-- 在這里，已經使轉成成了變量，並且執行了2次。 這說明，我們的force是生效的。
2.3.2 有統計信息和直方圖
當cursor_sharing 設置為similar，並且表上存在histograms時， 這時候的similar 等於 默認值 exact. 下面我們驗證一下。
-- 分析表
SYS@anqing2(rac2)> exec dbms_stats.gather_table_stats('SYS','TA', METHOD_OPT =>'FOR COLUMNS SIZE 10 ID');
PL/SQL procedure successfully completed.
--這里我手工的指定了收集的類型。
-- 查看直方圖: histograms
這部分內容，具體可以參考我的blog： Oracle Statistic 統計信息 小結
http://blog.csdn.net/tianlesoftware/archive/2009/10/14/4668723.aspx
SYS@anqing2(rac2)> SELECT column_name, num_distinct, num_buckets, histogram
2 FROM DBA_TAB_COL_STATISTICS WHERE table_name = 'TA' AND column_name = 'ID';
COLUMN_NAME NUM_DISTINCT NUM_BUCKETS HISTOGRAM
------------------------------ ------------ ----------- ---------------
ID 1005510 10 HEIGHT BALANCED
-- 先flush shared_pool 兩次
SYS@anqing2(rac2)> alter system flush shared_pool;
System altered.
SYS@anqing2(rac2)> alter system flush shared_pool;
System altered.
SYS@anqing2(rac2)> show parameter cursor_sharing
NAME TYPE VALUE
------------------------ -------------------------------- ----------
cursor_sharing string SIMILAR
-- select 查詢和對應的解析
SYS@anqing2(rac2)> select sql_text,child_number,executions from v$sql where sql_text like 'select count(*) from ta %';
SQL_TEXT CHILD_NUMBER EXECUTIONS
--------------------------------------------- ------------ ----------
select count(*) from ta where id>:"SYS_B_0" 0 3
SYS@anqing2(rac2)> select count(*) from ta where id>10000;
COUNT(*)
----------
990000
SYS@anqing2(rac2)> select sql_text,child_number,executions from v$sql where sql_text like 'select count(*) from ta %';
SQL_TEXT CHILD_NUMBER EXECUTIONS
--------------------------------------------- ------------ ----------
select count(*) from ta where id>:"SYS_B_0" 0 3
select count(*) from ta where id>:"SYS_B_0" 1 1
--因為ID值和之前的不一樣，所以沒有重用之前的cursor，而是重新創建了一個
SYS@anqing2(rac2)> select count(*) from ta where id>100;
COUNT(*)
----------
999900
SYS@anqing2(rac2)> select sql_text,child_number,executions from v$sql where sql_text like 'select count(*) from ta %';
SQL_TEXT CHILD_NUMBER EXECUTIONS
--------------------------------------------- ------------ ----------
select count(*) from ta where id>:"SYS_B_0" 0 3
select count(*) from ta where id>:"SYS_B_0" 1 1
select count(*) from ta where id>:"SYS_B_0" 2 1
-- 這次ID 值又不一樣，又重新生成了一個cursor
..... 中間又生成了N 次...
SYS@anqing2(rac2)> select count(*) from ta where id>90808;
COUNT(*)
----------
909192
SYS@anqing2(rac2)> select sql_text,child_number,executions from v$sql where sql_text like 'select count(*) from ta %';
SQL_TEXT CHILD_NUMBER EXECUTIONS
--------------------------------------------- ------------ ----------
select count(*) from ta where id>:"SYS_B_0" 0 3
select count(*) from ta where id>:"SYS_B_0" 1 1
select count(*) from ta where id>:"SYS_B_0" 2 1
select count(*) from ta where id>:"SYS_B_0" 3 1
select count(*) from ta where id>:"SYS_B_0" 4 1
select count(*) from ta where id>:"SYS_B_0" 5 1
select count(*) from ta where id>:"SYS_B_0" 6 1
select count(*) from ta where id>:"SYS_B_0" 7 1
-- 這個就像我們開始說的，雖然會對變量進行轉換，但不會重用，只有謂詞值一樣的時候才重用之前的cursor。
SYS@anqing2(rac2)> select count(*) from ta where id>90808;
COUNT(*)
----------
909192
SYS@anqing2(rac2)> select sql_text,child_number,executions from v$sql where sql_text like 'select count(*) from ta %';
SQL_TEXT CHILD_NUMBER EXECUTIONS
--------------------------------------------- ------------ ----------
select count(*) from ta where id>:"SYS_B_0" 0 3
select count(*) from ta where id>:"SYS_B_0" 1 1
select count(*) from ta where id>:"SYS_B_0" 2 1
select count(*) from ta where id>:"SYS_B_0" 3 1
select count(*) from ta where id>:"SYS_B_0" 4 1
select count(*) from ta where id>:"SYS_B_0" 5 1
select count(*) from ta where id>:"SYS_B_0" 6 1
select count(*) from ta where id>:"SYS_B_0" 7 2
-- 注意這里，因為我們執行了一個一樣的SQL,包括謂詞，這里的重用了之前的cursor。
總結：
當cursor_sharing 設置為similar時，Oracle 會對SQL 的位置值轉換成常量，這個和force 一樣。 但是不同的是：
（1）當字段上有histograms時， 這時等於exact，雖然SQL text 一樣，但謂詞值不同就不會重用cursor。 只有SQL 語句完全一樣，包括謂詞值，才會重用cursor。
（2）當沒有statstics時，即沒有histograms，這時等於force，只要SQL 語句一樣，就會重用cursor。
三. similar 模式對expdp 帶來的影響
在eygle 的blog上看到similar 對增加expdp備份的時間。
原理鏈接如下：
http://www.eygle.com/archives/2011/01/cursor_sharing_expdp.html
相關鏈接：
Oracle 10g Data Pump Expdp/Impdp 詳解
http://blog.csdn.net/tianlesoftware/archive/2009/10/15/4674224.aspx
Oracle expdp/impdp 使用示例
http://blog.csdn.net/tianlesoftware/archive/2011/03/18/6260138.aspx
3.1按照默認的expdp
SYS@anqing2(rac2)> show parameter cursor_sharing
NAME TYPE VALUE
------------------------------------ ----------- ------------------------------
cursor_sharing string EXACT
[oracle@rac2 backup]$ expdp /'/ as sysdba/' directory=backup full=y dumpfile=fullexp.dmp logfile=fullexp.log
Export: Release 10.2.0.4.0 - Production on Friday, 17 June, 2011 13:54:40
...............
Job "SYS"."SYS_EXPORT_FULL_01" successfully completed at 14:05:49
測試庫，全備花了11分鍾。
3.2 用similar 模式 expdp 備份
cursor_sharing是一個動態參數，可以在執行expdp之前進行修改，然后執行導出, 備份完在改回來。
SYS@anqing2(rac2)> alter system set cursor_sharing=similar scope=memory;
System altered.
SYS@anqing2(rac2)> show parameter cursor_sharing
NAME TYPE VALUE
------------------------------------ ----------- ------------------------------
cursor_sharing string SIMILAR
[oracle@rac2 backup]$ expdp /'/ as sysdba/' directory=backup full=y dumpfile=fullexp.dmp logfile=fullexp.log
Export: Release 10.2.0.4.0 - Production on Friday, 17 June,2011 13:40:37
....
Job "SYS"."SYS_EXPORT_FULL_01" successfully completed at 13:52:42
備份了12分鍾.
因為庫不大，所以差距不明顯. 不過如果庫大的話，一般也不會用expdp來備份。所以這里作為一個知識點，了解一下。
zhuan :http://www.cnblogs.com/zlja/archive/2011/06/17/2449633.html