db_file_multiblock_read_count

一、参数说明：

初始化参数db_file_multiblock_read_count是用来约束Oracle进行多数据块读取时的行为，所谓多数据块读取，就是Oracle在一次I/O时，可以读取多个数据块，从而用最小的I/O完成数据的读取。

db_file_multiblock_read_count的设置要受OS最大IO能力影响，也就是说，如果你系统的硬件IO能力有限，即使设置再大的db_file_multiblock_read_count也是没用的。

理论上，最大db_file_multiblock_read_count和系统IO能力应该有如下关系：

Max(db_file_multiblock_read_count) = MaxOsIOsize/db_block_size

当然这个Max(db_file_multiblock_read_count)还要受Oracle的限制。

在这里引用一下Oracle 11g r2的Reference中关于db_file_multiblock_read_count的介绍：



二、实验说明：

在这里引用一下谭大师的《让Oracle跑得更快2》中的一个例子：

----创建一张表jack并插入一下数据----

1 SQL> create table jack(x int,y int);

Table created.

SQL> insert into jack values(1,1);

1 row created.

SQL> insert into jack values(2,1);

1 row created.

----将表jack中的每个数据块存放在记录数收缩到最小，以便于记录分布在尽可能多的数据块上----

13 SQL> alter table jack minimize records_per_block;

Table altered.

----再次插入一下数据，并做一下数据分析----

17 SQL> insert into jack select rownum+2,1 from all_objects where rownum<=254;

254 rows created.

SQL> create index jack_ind on jack(x);

Index created.

SQL> exec dbms_stats.gather_table_stats(user,'jack');

PL/SQL procedure successfully completed.

----下面的查询说明jack表占用的数据块数为128个。

29 SQL> select count(distinct dbms_rowid.rowid_block_number(rowid)) from jack;

COUNT(DISTINCTDBMS_ROWID.ROWID_BLOCK_NUMBER(ROWID))
---------------------------------------------------
128

----清楚缓存，并将数据块间隔着读入内存中----

35 SQL> alter system flush buffer_cache;

System altered.

SQL> alter session set db_file_multiblock_read_count = 64;

Session altered.

SQL> declare
2      l_y number;
3  begin
4       for i in 1..64 loop
5          select y into l_y from jack where x = i*4;
6       end loop;
7  end;
8  /

PL/SQL procedure successfully completed.

----查看一下trace文件的路径----

54 SQL> @showtrace

trace_file_name
--------------------------------------------------------------------------------
/u01/app/oracle/diag/rdbms/yft/yft/trace/yft_ora_4556.trc

[oracle@yft ~]$ cat showtrace.sql
SELECT  d.VALUE
|| '/'
|| LOWER (RTRIM(i.INSTANCE,CHR(0)))
|| '_ora_'
|| p.spid
|| '.trc' as "trace_file_name"
FROM (SELECT p.spid
FROM v$mystat m,v$session s,v$process p
WHERE m.statistic# = 1 AND s.SID = m.SID AND p.addr = s.paddr) p,
(SELECT t.INSTANCE
FROM v$thread t,v$parameter v
WHERE v.NAME = 'thread'
AND (v.VALUE = 0 OR t.thread# = TO_NUMBER (v.VALUE))) i,
(SELECT VALUE
FROM v$parameter
WHERE NAME = 'user_dump_dest') d;

----打开10046事件，并进行一次全表扫描----

78 SQL> alter session set events '10046 trace name context forever,level 12';

Session altered.

SQL> set autotrace traceonly statistics;
SQL> select * from jack;

256 rows selected.

Statistics
----------------------------------------------------------
1  recursive calls
0  db block gets
199  consistent gets
123  physical reads
0  redo size
4829  bytes sent via SQL*Net to client
606  bytes received via SQL*Net from client
19  SQL*Net roundtrips to/from client
0  sorts (memory)
0  sorts (disk)
256  rows processed

SQL> alter session set events '10046 trace name context off';

Session altered.
---查看10046事件中的信息----

105 [oracle@yft ~]$ cat /u01/app/oracle/diag/rdbms/yft/yft/trace/yft_ora_4556.trc
WAIT #14: nam='db file sequential read' ela= 41 file#=6 block#=157 blocks=1 obj#=75062 tim=1358893053963775
WAIT #14: nam='db file sequential read' ela= 83 file#=6 block#=159 blocks=1 obj#=75062 tim=1358893053963907
WAIT #14: nam='db file sequential read' ela= 104 file#=6 block#=161 blocks=1 obj#=75062 tim=1358893053964108
WAIT #14: nam='db file scattered read' ela= 124 file#=6 block#=163 blocks=2 obj#=75062 tim=1358893053964318
FETCH #14:c=1000,e=698,p=5,cr=8,cu=0,mis=0,r=15,dep=0,og=1,plh=949574992,tim=1358893053964360
WAIT #14: nam='SQL*Net message from client' ela= 129 driver id=1650815232 #bytes=1 p3=0 obj#=75062 tim=1358893053964536
WAIT #14: nam='SQL*Net message to client' ela= 3 driver id=1650815232 #bytes=1 p3=0 obj#=75062 tim=1358893053964625
WAIT #14: nam='db file sequential read' ela= 92 file#=6 block#=166 blocks=1 obj#=75062 tim=1358893053964774
WAIT #14: nam='db file sequential read' ela= 91 file#=6 block#=168 blocks=1 obj#=75062 tim=1358893053964930
FETCH #14:c=0,e=393,p=2,cr=8,cu=0,mis=0,r=15,dep=0,og=1,plh=949574992,tim=1358893053964999
WAIT #14: nam='SQL*Net message from client' ela= 129 driver id=1650815232 #bytes=1 p3=0 obj#=75062 tim=1358893053965174
WAIT #14: nam='db file scattered read' ela= 221 file#=6 block#=171 blocks=2 obj#=75062 tim=1358893053965485
WAIT #14: nam='SQL*Net message to client' ela= 3 driver id=1650815232 #bytes=1 p3=0 obj#=75062 tim=1358893053965536
WAIT #14: nam='db file sequential read' ela= 64 file#=6 block#=174 blocks=1 obj#=75062 tim=1358893053965695
WAIT #14: nam='db file sequential read' ela= 85 file#=6 block#=178 blocks=1 obj#=75062 tim=1358893053965849

这个例子展示了Oracle多个数据块读取的工作机制，当内存中已经有了某个数据块时，Oracle将不再从磁盘中读取它。这里使用一个循环来通过索引块访问的方式(每次读取一个数据块)，
将间隔的数据块读入到内存中，这样即使db_file_multiblock_read_count设置为64时，执行jack表的全表扫描时，由于已经没有连续的数据块可供读取了，所以Oracle每次也只能将一个数据块读取到内存。
在等待时间中每一个WAIT#中blocks=1说明每一次I/O读取的数据块都为1，而且数据块的序号正好间隔为1，说明它们之间的那个数据块已经读取到内存中了。

在这里要说明一下SQL> alter table jack minimize records_per_block;该语句的作用：

SQL> create table echo (x int,y int);

Table created.

SQL> insert into echo values(1,1);

1 row created.

SQL> insert into echo values(2,1);

1 row created.

SQL> select count(distinct dbms_rowid.rowid_block_number(rowid)) from echo;

COUNT(DISTINCTDBMS_ROWID.ROWID_BLOCK_NUMBER(ROWID))
---------------------------------------------------
1

SQL> insert into echo select rownum+2,1 from all_objects where rownum<=254;

254 rows created.

SQL> select count(distinct dbms_rowid.rowid_block_number(rowid)) from echo;

COUNT(DISTINCTDBMS_ROWID.ROWID_BLOCK_NUMBER(ROWID))
---------------------------------------------------

在这里很明显可以看到没有使用alter table jack minimize records_per_block语句时，echo表占用的数据块数为1.                         1

三、设置db_file_multiblock_read_count为不同的值，观察SQL的性能变化，写出几种类型的SQL会从这个参数中受益

----创建一张echo表----
1 SQL> create table echo as select * from dba_objects;

Table created.

SQL> set autotrace trace exp;
----将参数设置成16，此时没有索引，只能进行全表扫描----
6 SQL> alter session set db_file_multiblock_read_count=16;

Session altered.

SQL> select * from echo;

Execution Plan
----------------------------------------------------------
Plan hash value: 642657756

--------------------------------------------------------------------------
| Id  | Operation      | Name | Rows  | Bytes | Cost (%CPU)| Time     |
--------------------------------------------------------------------------
|   0 | SELECT STATEMENT  |     | 63977 |    12M|   234   (1)| 00:00:03 |
|   1 |  TABLE ACCESS FULL| ECHO | 63977 |    12M|   234   (1)| 00:00:03 |
--------------------------------------------------------------------------

Note
-----
- dynamic sampling used for this statement (level=2)

SQL> alter session set db_file_multiblock_read_count=60;

Session altered.

SQL> select * from echo;

Execution Plan
----------------------------------------------------------
Plan hash value: 642657756

--------------------------------------------------------------------------
| Id  | Operation      | Name | Rows  | Bytes | Cost (%CPU)| Time     |
--------------------------------------------------------------------------
|   0 | SELECT STATEMENT  |     | 63977 |    12M|   194   (1)| 00:00:03 |
|   1 |  TABLE ACCESS FULL| ECHO | 63977 |    12M|   194   (1)| 00:00:03 |
--------------------------------------------------------------------------

Note
-----
- dynamic sampling used for this statement (level=2)

SQL> alter session set db_file_multiblock_read_count=128;

Session altered.

SQL> select * from echo;

Execution Plan
----------------------------------------------------------
Plan hash value: 642657756

--------------------------------------------------------------------------
| Id  | Operation      | Name | Rows  | Bytes | Cost (%CPU)| Time     |
--------------------------------------------------------------------------
|   0 | SELECT STATEMENT  |     | 63977 |    12M|   186   (1)| 00:00:03 |
|   1 |  TABLE ACCESS FULL| ECHO | 63977 |    12M|   186   (1)| 00:00:03 |
--------------------------------------------------------------------------

Note
-----
- dynamic sampling used for this statement (level=2)
-----当参数设定为16,60,128时，CBO计算出的FTS成本分别是234,194,186，很显然当参数的值越高CBO更倾向于全表扫描。
68
----给表创建一个主键----
69 SQL> alter table echo add constraint pk_echo primary key (object_id);

Table altered.

SQL> alter session set db_file_multiblock_read_count=16;

Session altered.

SQL> select count(*) from echo;

Execution Plan
----------------------------------------------------------
Plan hash value: 1123611804

-------------------------------------------------------------------------
| Id  | Operation          | Name    | Rows    | Cost (%CPU)| Time    |
-------------------------------------------------------------------------
|   0 | SELECT STATEMENT      |     |     1 |    38   (0)| 00:00:01 |
|   1 |  SORT AGGREGATE       |     |     1 |         |        |
|   2 |   INDEX FAST FULL SCAN| PK_ECHO | 63977 |    38   (0)| 00:00:01 |
-------------------------------------------------------------------------

Note
-----
- dynamic sampling used for this statement (level=2)

SQL> alter session set db_file_multiblock_read_count=60;

Session altered.

SQL> select count(*) from echo;

Execution Plan
----------------------------------------------------------
Plan hash value: 1123611804

-------------------------------------------------------------------------
| Id  | Operation          | Name    | Rows    | Cost (%CPU)| Time    |
-------------------------------------------------------------------------
|   0 | SELECT STATEMENT      |     |     1 |    31   (0)| 00:00:01 |
|   1 |  SORT AGGREGATE       |     |     1 |         |        |
|   2 |   INDEX FAST FULL SCAN| PK_ECHO | 63977 |    31   (0)| 00:00:01 |
-------------------------------------------------------------------------

Note
-----
- dynamic sampling used for this statement (level=2)

SQL> alter session set db_file_multiblock_read_count=128;

Session altered.

SQL> select count(*) from echo;

Execution Plan
----------------------------------------------------------
Plan hash value: 1123611804

-------------------------------------------------------------------------
| Id  | Operation          | Name    | Rows    | Cost (%CPU)| Time    |
-------------------------------------------------------------------------
|   0 | SELECT STATEMENT      |     |     1 |    30   (0)| 00:00:01 |
|   1 |  SORT AGGREGATE       |     |     1 |         |        |
|   2 |   INDEX FAST FULL SCAN| PK_ECHO | 63977 |    30   (0)| 00:00:01 |
-------------------------------------------------------------------------

Note
-----
- dynamic sampling used for this statement (level=2)
----进行INDEX_FFS查询时，参数的值越大CBO计算的成本越低
138
----执行INDEX RANGE SCAN查询----
139 SQL> alter session set db_file_multiblock_read_count=16;

Session altered.

SQL> select * from echo where object_id<1000;

Execution Plan
----------------------------------------------------------
Plan hash value: 3487819792

---------------------------------------------------------------------------------------
| Id  | Operation            | Name    | Rows  | Bytes | Cost (%CPU)| Time     |
---------------------------------------------------------------------------------------
|   0 | SELECT STATEMENT        |          |   942 |   190K|    29    (0)| 00:00:01 |
|   1 |  TABLE ACCESS BY INDEX ROWID| ECHO    |   942 |   190K|    29    (0)| 00:00:01 |
|*  2 |   INDEX RANGE SCAN        | PK_ECHO |   942 |       |     4    (0)| 00:00:01 |
---------------------------------------------------------------------------------------

Predicate Information (identified by operation id):
---------------------------------------------------

2 - access("OBJECT_ID"<1000)

Note
-----
- dynamic sampling used for this statement (level=2)

SQL> alter session set db_file_multiblock_read_count=60;

Session altered.

SQL> select * from echo where object_id<1000;

Execution Plan
----------------------------------------------------------
Plan hash value: 3487819792

---------------------------------------------------------------------------------------
| Id  | Operation            | Name    | Rows  | Bytes | Cost (%CPU)| Time     |
---------------------------------------------------------------------------------------
|   0 | SELECT STATEMENT        |          |   942 |   190K|    29    (0)| 00:00:01 |
|   1 |  TABLE ACCESS BY INDEX ROWID| ECHO    |   942 |   190K|    29    (0)| 00:00:01 |
|*  2 |   INDEX RANGE SCAN        | PK_ECHO |   942 |       |     4    (0)| 00:00:01 |
---------------------------------------------------------------------------------------

Predicate Information (identified by operation id):
---------------------------------------------------

2 - access("OBJECT_ID"<1000)

Note
-----
- dynamic sampling used for this statement (level=2)

SQL> alter session set db_file_multiblock_read_count=128;

Session altered.

SQL> select * from echo where object_id<1000;

Execution Plan
----------------------------------------------------------
Plan hash value: 3487819792

---------------------------------------------------------------------------------------
| Id  | Operation            | Name    | Rows  | Bytes | Cost (%CPU)| Time     |
---------------------------------------------------------------------------------------
|   0 | SELECT STATEMENT        |          |   942 |   190K|    29    (0)| 00:00:01 |
|   1 |  TABLE ACCESS BY INDEX ROWID| ECHO    |   942 |   190K|    29    (0)| 00:00:01 |
|*  2 |   INDEX RANGE SCAN        | PK_ECHO |   942 |       |     4    (0)| 00:00:01 |
---------------------------------------------------------------------------------------

Predicate Information (identified by operation id):
---------------------------------------------------

2 - access("OBJECT_ID"<1000)

Note
-----
- dynamic sampling used for this statement (level=2)
----当参数的值改变时，CBO计算的成本没有发生变化
219
----执行INDEX FULL SCAN查询----
220 SQL> alter session set db_file_multiblock_read_count=16;

Session altered.

SQL> select object_id from echo order by object_id;

Execution Plan
----------------------------------------------------------
Plan hash value: 1544245908

----------------------------------------------------------------------------
| Id  | Operation     | Name    | Rows  | Bytes | Cost (%CPU)| Time       |
----------------------------------------------------------------------------
|   0 | SELECT STATEMENT |       | 63977 |   812K|   165   (0)| 00:00:02 |
|   1 |  INDEX FULL SCAN | PK_ECHO | 63977 |   812K|   165   (0)| 00:00:02 |
----------------------------------------------------------------------------

Note
-----
- dynamic sampling used for this statement (level=2)

SQL> alter session set db_file_multiblock_read_count=60;

Session altered.

SQL> select object_id from echo order by object_id;

Execution Plan
----------------------------------------------------------
Plan hash value: 1544245908

----------------------------------------------------------------------------
| Id  | Operation     | Name    | Rows  | Bytes | Cost (%CPU)| Time       |
----------------------------------------------------------------------------
|   0 | SELECT STATEMENT |       | 63977 |   812K|   165   (0)| 00:00:02 |
|   1 |  INDEX FULL SCAN | PK_ECHO | 63977 |   812K|   165   (0)| 00:00:02 |
----------------------------------------------------------------------------

Note
-----
- dynamic sampling used for this statement (level=2)

SQL> alter session set db_file_multiblock_read_count=128;

Session altered.

SQL> select object_id from echo order by object_id;

Execution Plan
----------------------------------------------------------
Plan hash value: 1544245908

----------------------------------------------------------------------------
| Id  | Operation     | Name    | Rows  | Bytes | Cost (%CPU)| Time       |
----------------------------------------------------------------------------
|   0 | SELECT STATEMENT |       | 63977 |   812K|   165   (0)| 00:00:02 |
|   1 |  INDEX FULL SCAN | PK_ECHO | 63977 |   812K|   165   (0)| 00:00:02 |
----------------------------------------------------------------------------

Note
-----
- dynamic sampling used for this statement (level=2)
----当参数的值改变时，CBO计算的成本没有发生变化

总结：在这里只列举是4种查询方式，可以看出当参数的值越大时，FTS和INDEX_FFS的成本就会越低，执行计划就越向这边倾斜。

四、总结

对于OLTP数据库，每次用户读取的记录数非常少，这个值可以考虑设置小一点；而对于OLAP数据库，因为查询的量非常大，索引可以考虑设置大一些，但是需要注意多数据块读取只发生在以下两种情况下：

FTS(FULL TABLE SCAN)

INDEX_FFS(INDEX FAST FULL SCAN)

关于这个参数，在Oracle 10G r2及以后的版本里，Oracle不建议修改它的默认值，当设置这个参数为默认值时，Oracle会通过收集SQL的I/O情况，来动态设置这个参数的值；如果手工修改了它的默认值，Oracle将使用这个新的值。