Oracle sql 性能优化调整（34/35/36）

34.       避免在索引列上使用NOT

通常，　我们要避免在索引列上使用NOT, NOT会产生在和在索引列上使用函数相同的

影响. 当ORACLE”遇到”NOT,他就会停止使用索引转而执行全表扫描.

   举例:

 

   低效: (这里,不使用索引)

 

   SELECT …

   FROM DEPT

   WHERE DEPT_CODE NOT = 0;

   

   高效: (这里,使用了索引)

 

  SELECT …

   FROM DEPT

   WHERE DEPT_CODE > 0;

 

   需要注意的是,在某些时候, ORACLE优化器会自动将NOT转化成相对应的关系操作符.

   NOT >  to  <=

   NOT >=  to  <

   NOT <  to  >=

   NOT <=  to  >

 

 

 

     在这个例子中,作者犯了一些错误. 例子中的低效率SQL是不能被执行的.

我做了一些测试:

     

SQL> select * from emp where NOT empno > 1;

no rows selected

Execution Plan

----------------------------------------------------------

   0      SELECT STATEMENT Optimizer=CHOOSE

   1    0   TABLE ACCESS (BY INDEX ROWID) OF 'EMP'

   2    1     INDEX (RANGE SCAN) OF 'EMPNO' (UNIQUE)     

 

SQL> select * from emp where empno <= 1;

no rows selected

Execution Plan

----------------------------------------------------------

   0      SELECT STATEMENT Optimizer=CHOOSE

   1    0   TABLE ACCESS (BY INDEX ROWID) OF 'EMP'

   2    1     INDEX (RANGE SCAN) OF 'EMPNO' (UNIQUE)

 

      两者的效率完全一样，也许这符合作者关于” 在某些时候, ORACLE优化器会自动将NOT转化成相对应的关系操作符” 的观点．

      

35.       用>=替代>

 

如果DEPTNO上有一个索引,

 

高效:

 

   SELECT *

   FROM EMP

   WHERE DEPTNO >=4

   

   低效:

 

   SELECT *

   FROM EMP

   WHERE DEPTNO >3

 

      两者的区别在于, 前者DBMS将直接跳到第一个DEPT等于4的记录而后者将首先定位到DEPTNO=3的记录并且向前扫描到第一个DEPT大于3的记录.

36.       用UNION替换OR (适用于索引列)

通常情况下, 用UNION替换WHERE子句中的OR将会起到较好的效果. 对索引列使用OR将造成全表扫描. 注意, 以上规则只针对多个索引列有效. 如果有column没有被索引, 查询效率可能会因为你没有选择OR而降低.

   在下面的例子中, LOC_ID 和REGION上都建有索引.

高效:

   SELECT LOC_ID , LOC_DESC , REGION

   FROM LOCATION

   WHERE LOC_ID = 10

   UNION

   SELECT LOC_ID , LOC_DESC , REGION

   FROM LOCATION

   WHERE REGION = “MELBOURNE”

 

低效:

   SELECT LOC_ID , LOC_DESC , REGION

   FROM LOCATION

   WHERE LOC_ID = 10 OR REGION = “MELBOURNE”

 

如果你坚持要用OR, 那就需要返回记录最少的索引列写在最前面.

 

注意:

 

WHERE KEY1 = 10   (返回最少记录)

OR KEY2 = 20        (返回最多记录)

 

ORACLE 内部将以上转换为

WHERE KEY1 = 10 AND

((NOT KEY1 = 10) AND KEY2 = 20)        

 

 

 

下面的测试数据仅供参考: (a = 1003 返回一条记录 , b = 1 返回1003条记录)

SQL> select * from unionvsor /*1st test*/

  2   where a = 1003 or b = 1;

1003 rows selected.

Execution Plan

----------------------------------------------------------

   0      SELECT STATEMENT Optimizer=CHOOSE

   1    0   CONCATENATION

   2    1     TABLE ACCESS (BY INDEX ROWID) OF 'UNIONVSOR'

   3    2       INDEX (RANGE SCAN) OF 'UB' (NON-UNIQUE)

   4    1     TABLE ACCESS (BY INDEX ROWID) OF 'UNIONVSOR'

   5    4       INDEX (RANGE SCAN) OF 'UA' (NON-UNIQUE)

Statistics

----------------------------------------------------------

          0  recursive calls

          0  db block gets

        144  consistent gets

          0  physical reads

          0  redo size

      63749  bytes sent via SQL*Net to client

       7751  bytes received via SQL*Net from client

         68  SQL*Net roundtrips to/from client

          0  sorts (memory)

          0  sorts (disk)

       1003  rows processed

SQL> select * from unionvsor /*2nd test*/

  2  where b  = 1 or a = 1003 ;

1003 rows selected.

Execution Plan

----------------------------------------------------------

   0      SELECT STATEMENT Optimizer=CHOOSE

   1    0   CONCATENATION

   2    1     TABLE ACCESS (BY INDEX ROWID) OF 'UNIONVSOR'

   3    2       INDEX (RANGE SCAN) OF 'UA' (NON-UNIQUE)

   4    1     TABLE ACCESS (BY INDEX ROWID) OF 'UNIONVSOR'

   5    4       INDEX (RANGE SCAN) OF 'UB' (NON-UNIQUE)

Statistics

----------------------------------------------------------

          0  recursive calls

          0  db block gets

        143  consistent gets

          0  physical reads

          0  redo size

      63749  bytes sent via SQL*Net to client

       7751  bytes received via SQL*Net from client

         68  SQL*Net roundtrips to/from client

          0  sorts (memory)

          0  sorts (disk)

       1003  rows processed

 

SQL> select * from unionvsor /*3rd test*/

  2  where a = 1003

  3  union

  4   select * from unionvsor

  5   where b = 1;

1003 rows selected.

Execution Plan

----------------------------------------------------------

   0      SELECT STATEMENT Optimizer=CHOOSE

   1    0   SORT (UNIQUE)

   2    1     UNION-ALL

   3    2       TABLE ACCESS (BY INDEX ROWID) OF 'UNIONVSOR'

   4    3         INDEX (RANGE SCAN) OF 'UA' (NON-UNIQUE)

   5    2       TABLE ACCESS (BY INDEX ROWID) OF 'UNIONVSOR'

   6    5         INDEX (RANGE SCAN) OF 'UB' (NON-UNIQUE)

Statistics

----------------------------------------------------------

          0  recursive calls

          0  db block gets

         10  consistent gets   

          0  physical reads

          0  redo size

      63735  bytes sent via SQL*Net to client

       7751  bytes received via SQL*Net from client

         68  SQL*Net roundtrips to/from client

          1  sorts (memory)

          0  sorts (disk)

       1003  rows processed

用UNION的效果可以从consistent gets和 SQL*NET的数据交换量的减少看出