分页格式1：
SELECT * FROM (SELECT A.*, ROWNUM RN FROM (SELECT * FROM TEST_DISTINCT) A WHERE ROWNUM <= 40) WHERE RN >= 21
该语句最内层SELECT * FROM TEST_DISTINCT表示不翻页遍历整张表，ROWNUM <= 40和RN >= 21控制每页查询的范围。分页的目的就是为了控制查询语句的结果集，上面语句的重点就在ROWNUM <= 40。
格式1展示的是在第二层控制结果集的最大值，在第三次控制结果集的最小值，控制分页的另外一种方式可以在最外层直接控制结果集的范围。
分页格式2：
SELECT * FROM (SELECT A.*, ROWNUM RN FROM (SELECT * FROM TEST_DISTINCT) A) WHERE RN BETWEEN 21 AND 40
对比两个语句的执行计划，发现通常情况下，格式1比格式2的效率更高。
格式1
SQL> SELECT * FROM (SELECT A.*, ROWNUM RN FROM (SELECT * FROM TEST_DISTINCT) A WHERE ROWNUM <= 40) WHERE RN >= 21
2 ;
20 rows selected.
Execution Plan
----------------------------------------------------------
Plan hash value: 3629022454
-------------------------------------------------------------------------------------
| Id | Operation | Name | Rows | Bytes | Cost (%CPU)| Time |
-------------------------------------------------------------------------------------
| 0 | SELECT STATEMENT | | 40 | 1040 | 2 (0)| 00:00:01 |
|* 1 | VIEW | | 40 | 1040 | 2 (0)| 00:00:01 |
|* 2 | COUNT STOPKEY | | | | | |
| 3 | TABLE ACCESS FULL| TEST_DISTINCT | 101K| 1287K| 2 (0)| 00:00:01 |
-------------------------------------------------------------------------------------
Predicate Information (identified by operation id):
---------------------------------------------------
1 - filter("RN">=21)
2 - filter(ROWNUM<=40)
Note
-----
- dynamic sampling used for this statement (level=2)
Statistics
----------------------------------------------------------
7 recursive calls
0 db block gets
144 consistent gets
128 physical reads
0 redo size
1012 bytes sent via SQL*Net to client
534 bytes received via SQL*Net from client
3 SQL*Net roundtrips to/from client
0 sorts (memory)
0 sorts (disk)
20 rows processed
格式2
SQL> SELECT *
2 FROM (SELECT A.*, ROWNUM RN FROM (SELECT * FROM TEST_DISTINCT) A)
3 WHERE RN BETWEEN 21 AND 40;
20 rows selected.
Execution Plan
----------------------------------------------------------
Plan hash value: 2746576408
-------------------------------------------------------------------------------------
| Id | Operation | Name | Rows | Bytes | Cost (%CPU)| Time |
-------------------------------------------------------------------------------------
| 0 | SELECT STATEMENT | | 101K| 2574K| 44 (3)| 00:00:01 |
|* 1 | VIEW | | 101K| 2574K| 44 (3)| 00:00:01 |
| 2 | COUNT | | | | | |
| 3 | TABLE ACCESS FULL| TEST_DISTINCT | 101K| 1287K| 44 (3)| 00:00:01 |
-------------------------------------------------------------------------------------
Predicate Information (identified by operation id):
---------------------------------------------------
1 - filter("RN"<=40 AND "RN">=21)
Note
-----
- dynamic sampling used for this statement (level=2)
Statistics
----------------------------------------------------------
0 recursive calls
0 db block gets
158 consistent gets
153 physical reads
0 redo size
1012 bytes sent via SQL*Net to client
534 bytes received via SQL*Net from client
3 SQL*Net roundtrips to/from client
0 sorts (memory)
0 sorts (disk)
20 rows processed从上面的执行计划比对可以看出，因为格式1的过滤发生在第二层，则内部的结果满足条件后就停止了查询，所以返回的效率相比格式2要高出很多。
另外，简单讨论一下多表等值连接的情况。对于常见的等值表连接查询，CBO一般采用两种相对高效的连接方式NESTED LOOP和HASH JOIN(MERGE JOIN效率比HASH JOIN低，CBO一般不优先考虑)。在这里，由于使用了分页，通过指定的条件限定了结果集的最大值，NESTED LOOP的情况下，当记录数超过条件时可以马上返回结果集，而HASH JOIN必须处理完所有结果集。因此对于大多数情况的分页查询，使用NESTED LOOP作为查询的连接方式效率更高。
强制Oracle使用CBO可以通过Hint，但查看了下这样的结果执行效率反而下降
SELECT /*+ first_rows */ * FROM (SELECT A.*, ROWNUM RN FROM (SELECT * FROM TEST_DISTINCT) A WHERE ROWNUM <= 40) WHERE RN >= 21 Execution Plan ---------------------------------------------------------- Plan hash value: 3629022454 ------------------------------------------------------------------------------------- | Id | Operation | Name | Rows | Bytes | Cost (%CPU)| Time | ------------------------------------------------------------------------------------- | 0 | SELECT STATEMENT | | 40 | 1040 | 44 (3)| 00:00:01 | |* 1 | VIEW | | 40 | 1040 | 44 (3)| 00:00:01 | |* 2 | COUNT STOPKEY | | | | | | | 3 | TABLE ACCESS FULL| TEST_DISTINCT | 101K| 1287K| 44 (3)| 00:00:01 | -------------------------------------------------------------------------------------
创建一个大表用于模拟生产环境测试
SQL> create table t as select * from dba_objects,dba_sequences;
SQL> select count(*) from t;
COUNT(*)
----------
16755730
SQL> SET AUTOT ON
SQL> COL OBJECT_NAME FORMAT A30
SQL> exec dbms_stats.gather_table_stats('SYS','T');对于格式2，可以看到执行计划消耗非常高。
Execution Plan
----------------------------------------------------------
Plan hash value: 1961128543
----------------------------------------------------------------------------
| Id | Operation | Name | Rows | Bytes | Cost (%CPU)| Time |
----------------------------------------------------------------------------
| 0 | SELECT STATEMENT | | 16M| 1470M| 99913 (1)| 00:19:59 |
|* 1 | VIEW | | 16M| 1470M| 99913 (1)| 00:19:59 |
| 2 | COUNT | | | | | |
| 3 | TABLE ACCESS FULL| T | 16M| 479M| 99913 (1)| 00:19:59 |
----------------------------------------------------------------------------
Predicate Information (identified by operation id):
---------------------------------------------------
1 - filter("RN"<=20 AND "RN">=11)
Statistics
----------------------------------------------------------
1 recursive calls
0 db block gets
368199 consistent gets
368191 physical reads
0 redo size
806 bytes sent via SQL*Net to client
523 bytes received via SQL*Net from client
2 SQL*Net roundtrips to/from client
0 sorts (memory)
0 sorts (disk)
10 rows processed而在格式1，效率有明显的提升，一共只发生了96次物理读操作。
select object_id, object_name
from (select rownum rn, object_id, object_name
from (select object_id, object_name from t)
where rownum <= 20)
where rn >= 11;
Execution Plan
----------------------------------------------------------
Plan hash value: 3593519476
----------------------------------------------------------------------------
| Id | Operation | Name | Rows | Bytes | Cost (%CPU)| Time |
----------------------------------------------------------------------------
| 0 | SELECT STATEMENT | | 20 | 1840 | 2 (0)| 00:00:01 |
|* 1 | VIEW | | 20 | 1840 | 2 (0)| 00:00:01 |
|* 2 | COUNT STOPKEY | | | | | |
| 3 | TABLE ACCESS FULL| T | 20 | 600 | 2 (0)| 00:00:01 |
----------------------------------------------------------------------------
Predicate Information (identified by operation id):
---------------------------------------------------
1 - filter("RN">=11)
2 - filter(ROWNUM<=20)
Statistics
----------------------------------------------------------
1 recursive calls
0 db block gets
5 consistent gets
96 physical reads
0 redo size
806 bytes sent via SQL*Net to client
523 bytes received via SQL*Net from client
2 SQL*Net roundtrips to/from client
0 sorts (memory)
0 sorts (disk)
10 rows processed但是格式1的分页如果过于靠后，几乎接近于表的所有行时，其效率与格式2也没有多大的差距了
select object_id, object_name
from (select rownum rn, object_id, object_name
from (select object_id, object_name from t)
where rownum <= 16000011 )
where rn >= 16000000;
Execution Plan
----------------------------------------------------------
Plan hash value: 3593519476
----------------------------------------------------------------------------
| Id | Operation | Name | Rows | Bytes | Cost (%CPU)| Time |
----------------------------------------------------------------------------
| 0 | SELECT STATEMENT | | 16M| 1403M| 95407 (1)| 00:19:05 |
|* 1 | VIEW | | 16M| 1403M| 95407 (1)| 00:19:05 |
|* 2 | COUNT STOPKEY | | | | | |
| 3 | TABLE ACCESS FULL| T | 16M| 457M| 95407 (1)| 00:19:05 |
----------------------------------------------------------------------------
Predicate Information (identified by operation id):
---------------------------------------------------
1 - filter("RN">=16000000)
2 - filter(ROWNUM<=16000011)
Statistics
----------------------------------------------------------
1 recursive calls
0 db block gets
351909 consistent gets
352128 physical reads
0 redo size
953 bytes sent via SQL*Net to client
523 bytes received via SQL*Net from client
2 SQL*Net roundtrips to/from client
0 sorts (memory)
0 sorts (disk)
12 rows processed
