Oracle 12C In-Memory特性研究

一、Oracle In-Memory

行格式与列格式:

Oracle 数据库传统上以行格式存储数据。在行格式数据库中,数据库中存储的每个新事务或新记录都表示为表中的一个新行,而在查询数据时是利用传统BUFFER CACHE。

列格式数据库将表以单独的列结构存储到内存中。列格式适用于报表类,分析,选择少量列但是查询要访问大部分的数据的场景。
在这里插入图片描述

1.1 In-Memory 开启方法

启用IMO非常简单,12.1.0.2及之后版本下,设置INMEMORY_SIZE 为非0值便可启用IM column store特性。

INMEMORY_SIZE 是个实例级参数,默认为0,设置一个非0值时,最小值为100M。

通常情况下,sys用户下的对象及SYSTEM、SYSAUX表空间上的对象无法使用IMO特性,但通过设置“_enable_imc_sys”隐含参数也可以使用

开启DB In-Memory过程如下:
1、修改INMEMORY_SIZE参数:
SQL> ALTER SYSTEM SET INMEMORY_SIZE=1G SCOPE=SPFILE;

2、检查sga参数的设置,确保在设置完inmemroy_size参数之后数据库实例还可以正常启动。如果数据库使用了ASMM,则需要检查sga_target参数。如果使用了AMM,则需要检查MEMORY_TARGET参数,同时也需要检查SGA_MAX_TARGET(或MEMORY_MAX_TARGET)。

备注:从 12.2 开始,可以动态增加 In-Memory 区域的大小,为此,只需 通过 ALTER SYSTEM 命令增加 INMEMORY_SIZE 参数值即可

3、重启数据库实例

4、查看IM特性是否开启

SQL> SHOW PARAMETER inmemory;
NAME                                 TYPE        VALUE
------------------------------------ ----------- ------------------------------
inmemory_adg_enabled                 boolean     TRUE
inmemory_clause_default              string
inmemory_expressions_usage           string      ENABLE
inmemory_force                       string      DEFAULT
inmemory_max_populate_servers        integer     2
inmemory_query                       string      ENABLE
inmemory_size                        big integer 1G
inmemory_trickle_repopulate_servers_ integer     1
percent
inmemory_virtual_columns             string      MANUAL
optimizer_inmemory_aware             boolean     TRUE

1.2 开启与关闭IM column store

1、TABLE 级启用:

可以通过如下初始建表或后续修改表 inmemory 属性的方式进行启用:
create table test (id number) inmemory;
alter table test inmemory;

2、COLUMN 级启用:

仅启用表中某列前,该表必须先设置为 inmemory 模式:
alter table imo_t1 inmemory (id) no inmemory (name,type); alter table imo_t2 inmemory (name) no inmemory (id,type);

SELECT table_name, segment_column_id seg_col_id, column_name, inmemory_compression FROM v$im_column_level WHERE owner = 'IMOTEST' and table_name in ('IMO_T1','IMO_T2') ORDER BY 1,3;

                
TABLE_NAME SEG_COL_ID COLUMN_NAME   INMEMORY_COMPRESSION    
---------- ---------- --------------- -----------------------   
IMO_T1           1          ID          DEFAULT 
IMO_T1           2          NAME        NO INMEMORY 
IMO_T1           3          TYPE        NO INMEMORY 
IMO_T2           1          ID          NO INMEMORY 
IMO_T2           2          NAME        DEFAULT 
IMO_T2           3          TYPE        NO INMEMORY

3、表空间级启用:

可以通过如下初始创建表空间或后续修改表空间 inmemory 属性的方式进行启用,在属性为 inmemory 的表空间中创建的对象自动加载 inmemory 属性,除非显示设置对象为 no inmemory:
create tablespace imotest datafile '/u01/app/oracle/oradata/orcl/imotest01.dbf' size 100M default inmemory;alter tablespace imotest default inmemory;
select tablespace_name, def_inmemory from dba_tablespaces where tablespace in ('IMOTEST');

TABLESPACE_NAME                 DEF_INMEMORY
------------------------------ ---------------
IMOTEST                         ENABLED

4、如果需要知道具体哪些列开启了IM column store则需要到V$IM_COLUMN_LEVEL中进行查看

SQL> SELECT TABLE_NAME, COLUMN_NAME, INMEMORY_COMPRESSION FROM V$IM_COLUMN_LEVEL WHERE TABLE_NAME = 'TBNAME';

5、查询有关IM列表属性

SQL> SELECT OWNER, SEGMENT_NAME,bytes,INMEMORY_SIZE,POPULATE_STATUS, BYTES_NOT_POPULATED FROM V$IM_SEGMENTS;

OWNER      SEGMENT_NA      BYTES INMEMORY_SIZE POPULATE_STAT BYTES_NOT_POPULATED
---------- ---------- ---------- ------------- ------------- -------------------
CS         MEM         391880704      52428800 COMPLETED                       0
CS         TEST         99426304      15138816 COMPLETED                       0
CS         DUP           5095424       3407872 COMPLETED                       0

6、关闭inmemory

alter table test no inmemory;

1.3 inmemory优先级调整

启用了 IMO 的对象,会按照一定的优先级进入 SGA 中配置好的 IN-MEMORY 区域,同时,在 IN-MEMORY 区域用满后,依次置换出优先级较低的对象。下表为关于 IMO 对象优先级说明:

优先级描述

PRIORITY NONE	缺省级别;执行 SQL 引起对象扫描后,触发进入 IN-MEMORY
PRIORITY CRITICAL	最高优先级;数据库启动后立即进入 IN-MEMORY
PRIORITY HIGH	在具有 CRITICAL 优先级的对象之后进入 IN-MEMORY
PRIORITY MEDIUM	在具有 CRITICAL、HIGH 优先级的对象之后进入 IN-MEMORY
PRIORITY LOW	在具有 CRITICAL、HIGH、MEDIUM 优先级的对象之后进入 IN-MEMORY

修改示例:

SQL> alter table test inmemory priority high;

SQL> col owner format a30

SQL> col segment_name format a30

SQL> SELECT INMEMORY,INMEMORY_PRIORITY,INMEMORY_COMPRESSION,INMEMORY_DISTRIBUTE, INMEMORY_DUPLICATE
FROM USER_TABLES WHERE TABLE_NAME = 'TEST';
INMEMORY INMEMORY INMEMORY_COMPRESS INMEMORY_DISTRI INMEMORY_DUPL
-------- -------- ----------------- --------------- -------------
ENABLED  HIGH     FOR QUERY LOW     AUTO            NO DUPLICATE

1.4 加载对象到IM

1、通过全表扫描objects加载数据到IM

SQL> SELECT /*+ FULL (s) */ COUNT(*) FROM table s;
查询V$IM_SEGMENTS中的数据
SQL> col owner for a10

SQL> col SEGMENT_NAME for a10

SQL> SELECT OWNER, SEGMENT_NAME, POPULATE_STATUS, BYTES_NOT_POPULATED FROM V$IM_SEGMENTS;
 
OWNER      SEGMENT_NA POPULATE_ BYTES_NOT_POPULATED

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

CS    TEST      COMPLETED                   0

2、DBMS_INMEMORY来加载数据

DBMS_INMEMORY包提供了两个PROCEDURE来手工加载数据到IM中:

POPULATE:强制加载给定的表

REPOPULATE:强制重新加载给定的表,在该表已经至少加载过一次之后才可使用。

手工执行DBMS_INMEMORY.POPULATE procedure来加载TEST表到IM中
SQL> EXEC DBMS_INMEMORY.POPULATE('CS','TEST');

通过DBMS_INMEMORY.REPOPULATE prcedure及FORCE=>TRUE选项来加载TEST表到IM中。FORCE=>TRUE选项强制进行一次完整重新载入,类似于full refresh

SQL> EXEC DBMS_INMEMORY.REPOPULATE('CS','TEST', FORCE=>TRUE);

3、设置PRIORITY clause自动加载数据
将测试表TEST的PRIORITY级别改为HIGH。

SQL> ALTER TABLE test INMEMORY PRIORITY HIGH|MEDIUM|LOW;

二、In-Memory测试

2.1 全字段查询

通过CTAS方式创建表TEST,普通行式扫描与开启inmemory特性分别测试,比对执行计划。

create table test as select * from dba_objects;

no inmemory查询测试

set timing on
set autotrace traceonly
set pagesize 200 linesize 200
col TABLE_NAME for a20
col INMEMORY_PRIORITY for a20
col INMEMORY_DISTRIBUTE for a20
col INMEMORY_COMPRESSION for a20

全字段查询,执行计划

SQL> select count(*) from test;
COUNT(*)
----------
     74329
Elapsed: 00:00:00.01

Execution Plan
----------------------------------------------------------
Plan hash value: 1950795681

-------------------------------------------------------------------
| Id  | Operation          | Name | Rows  | Cost (%CPU)| Time     |
-------------------------------------------------------------------
|   0 | SELECT STATEMENT   |      |     1 |   403   (1)| 00:00:01 |
|   1 |  SORT AGGREGATE    |      |     1 |            |          |
|   2 |   TABLE ACCESS FULL| TEST | 74329 |   403   (1)| 00:00:01 |
-------------------------------------------------------------------


Statistics
----------------------------------------------------------
          0  recursive calls
          5  db block gets
       1460  consistent gets
          0  physical reads
          0  redo size
        544  bytes sent via SQL*Net to client
        608  bytes received via SQL*Net from client
          2  SQL*Net roundtrips to/from client
          0  sorts (memory)
          0  sorts (disk)
          1  rows processed

开启in memory查询测试

SQL> alter table test inmemory;
SQL> select TABLE_NAME,INMEMORY_PRIORITY,INMEMORY_DISTRIBUTE,INMEMORY_COMPRESSION from user_tables;

TABLE_NAME           INMEMORY_PRIORITY    INMEMORY_DISTRIBUTE  INMEMORY_COMPRESSION
-------------------- -------------------- -------------------- --------------------
TEST                 NONE                 AUTO                 FOR QUERY LOW

已开启inmemory
SQL> select pool,ALLOC_BYTES/1024/1024,USED_BYTES/1024/1024,POPULATE_STATUS,con_id 
from V$INMEMORY_AREA;

POOL          ALLOC_BYTES/1024/1024 USED_BYTES/1024/1024 POPULATE_STATUS      CON_ID
------------  ------------          --------------------- ---------- ------------------
1MB POOL          815                    0                DONE                      3
64KB POOL         192                    0                DONE                      3
--因为只是把该表设置了INMEMORY,但是未查询过,所以查询V$INMEMORY_AREA中未使用相关内存--

全字段查询,执行计划
第一次执行:

SQL> select count(*) from test;

  COUNT(*)
----------
     74329

Elapsed: 00:00:00.05

Execution Plan
----------------------------------------------------------
Plan hash value: 1950795681

----------------------------------------------------------------------------
| Id  | Operation                   | Name | Rows  | Cost (%CPU)| Time     |
----------------------------------------------------------------------------
|   0 | SELECT STATEMENT            |      |     1 |   403   (1)| 00:00:01 |
|   1 |  SORT AGGREGATE             |      |     1 |            |          |
|   2 |   TABLE ACCESS INMEMORY FULL| TEST | 74329 |   403   (1)| 00:00:01 |
----------------------------------------------------------------------------


Statistics
----------------------------------------------------------
          5  recursive calls
          5  db block gets
       1467  consistent gets
          0  physical reads
          0  redo size
        544  bytes sent via SQL*Net to client
        608  bytes received via SQL*Net from client
          2  SQL*Net roundtrips to/from client
          2  sorts (memory)
          0  sorts (disk)
          1  rows processed

第二次执行:

Elapsed: 00:00:00.03

Execution Plan
----------------------------------------------------------
Plan hash value: 1950795681

----------------------------------------------------------------------------
| Id  | Operation                   | Name | Rows  | Cost (%CPU)| Time     |
----------------------------------------------------------------------------
|   0 | SELECT STATEMENT            |      |     1 |    16   (7)| 00:00:01 |
|   1 |  SORT AGGREGATE             |      |     1 |            |          |
|   2 |   TABLE ACCESS INMEMORY FULL| TEST | 74329 |    16   (7)| 00:00:01 |
----------------------------------------------------------------------------


Statistics
----------------------------------------------------------
          0  recursive calls
          3  db block gets
          9  consistent gets
          0  physical reads
          0  redo size
        544  bytes sent via SQL*Net to client
        608  bytes received via SQL*Net from client
          2  SQL*Net roundtrips to/from client
          0  sorts (memory)
          0  sorts (disk)
          1  rows processed

查询inmemory内存使用:

SQL> select pool,ALLOC_BYTES/1024/1024,USED_BYTES/1024/1024,POPULATE_STATUS,con_id 
from V$INMEMORY_AREA;

POOL                       ALLOC_BYTES/1024/1024 USED_BYTES/1024/1024 POPULATE_STATUS                CON_ID
-------------------------- --------------------- -------------------- -------------------------- ----------
1MB POOL                                     815                    4 DONE                                3
64KB POOL                                    192                  .25 DONE                                3
--再次查看,已经使用了分配的In-Memory中内存

性能比对:

sql planno-inmemoryinmemory
consistent gets14609
physical reads00
Cost40316

结果:
开启inmemory之后性能提升162倍

2.2 索引字段比较测试

在已开启IMO特性条件下,通过给表test增加列索引,比较IM与索引执行计划
1、在object_name上创建索引

SQL> create index idx_test_OBname on test(OBJECT_NAME);

2、查看表TEST是否加载到IM中

col owner for a20
col SEGMENT_NAME for a20
col POPULATE_STATUS for a20
col BYTES_NOT_POPULATED for 99999

SELECT OWNER, SEGMENT_NAME, POPULATE_STATUS, BYTES_NOT_POPULATED FROM V$IM_SEGMENTS;
OWNER                SEGMENT_NAME         POPULATE_STATUS      BYTES_NOT_POPULATED
-------------------- -------------------- -------------------- -------------------
CS                   TEST                 COMPLETED                              0
表已被加入到IM中

3、使用索引简单列查询

SQL> set autotrace traceonly
SQL> select count(*) from test where object_name='TEST';


Execution Plan
----------------------------------------------------------
Plan hash value: 3197243274

-------------------------------------------------------------------------------------
| Id  | Operation         | Name            | Rows  | Bytes | Cost (%CPU)| Time     |
-------------------------------------------------------------------------------------
|   0 | SELECT STATEMENT  |                 |     1 |    35 |     3   (0)| 00:00:01 |
|   1 |  SORT AGGREGATE   |                 |     1 |    35 |            |          |
|*  2 |   INDEX RANGE SCAN| IDX_TEST_OBNAME |    10 |   350 |     3   (0)| 00:00:01 |
-------------------------------------------------------------------------------------

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

   2 - access("OBJECT_NAME"='TEST')


Statistics
----------------------------------------------------------
          2  recursive calls
          0  db block gets
          6  consistent gets
          2  physical reads
          0  redo size
        542  bytes sent via SQL*Net to client
        607  bytes received via SQL*Net from client
          2  SQL*Net roundtrips to/from client
          0  sorts (memory)
          0  sorts (disk)
          1  rows processed
          
          逻辑读:6
          物理:2

4、强制全表扫描查询

SQL> select /*+full(s)*/count(*) from test s where object_name='test';

Execution Plan
----------------------------------------------------------
Plan hash value: 1950795681

------------------------------------------------------------------------------------
| Id  | Operation                   | Name | Rows  | Bytes | Cost (%CPU)| Time     |
------------------------------------------------------------------------------------
|   0 | SELECT STATEMENT            |      |     1 |    35 |   124   (1)| 00:00:01 |
|   1 |  SORT AGGREGATE             |      |     1 |    35 |            |          |
|*  2 |   TABLE ACCESS INMEMORY FULL| TEST |     9 |   315 |   124   (1)| 00:00:01 |
------------------------------------------------------------------------------------

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

   2 - inmemory("OBJECT_NAME"='test')
       filter("OBJECT_NAME"='test')


Statistics
----------------------------------------------------------
          1  recursive calls
          0  db block gets
         10  consistent gets
          0  physical reads
          0  redo size
        541  bytes sent via SQL*Net to client
        607  bytes received via SQL*Net from client
          2  SQL*Net roundtrips to/from client
          0  sorts (memory)
          0  sorts (disk)
          1  rows processed
          
逻辑读:10
物理读:0

总结:在数据离散度较高,且通过索引条件过滤的扫描场景中,IM特性对性能并没有提升,传统的索引+行式存储的执行计划已经足够,在默认情况下还是会根据查询索引返回rowid的方式查找数据。

2.3 批量update测试

1、创建测试表disk、emp,分别为im表以及rows表,总数相同

SQL> create table disk as select * from dba_objects;
SQL> insert into disk select * from disk;

74360 rows created.

SQL> /

148720 rows created.

SQL> /

297440 rows created.

SQL> /

594880 rows created.

SQL> /

1189760 rows created.

SQL> commit;

Commit complete.

SQL> select count(*) from disk;

  COUNT(*)
----------
   2379520
   
  SQL> create table mem as select * from disk inmemory; 

更新磁盘表
update disk set owner=‘tom’;
更新内存表
update mem set owner=‘tom’;

普通表:
时间:14.49
逻辑读:3202294+894128
物理读:5602
cost:403

IM表:
时间:18.94
逻辑读:3231179+254
物理读:92476
cost:483

总结:IM特性在update性能有所下降

2.4 大批量insert测试

从磁盘插到磁盘
insert into disk select * from cs;
从磁盘插到内存
insert into mem select * from cs;

普通表:
时间:0.13
逻辑读:9332+2903
物理读:2895
cost:404

IM表:
时间:0.17
逻辑读:3231179+254
物理读:92476
cost:483

2.5 大量delete测试

delete from disk where rownum <10000;
delete from mem where rownum <10000;
普通表:
时间:0.08
逻辑读:11470+202
物理读:253
cost:12986

IM表:
时间:0.07
逻辑读:11472+9
物理读:0
cost:493

2.6 压缩方式测试

压缩方式描述
NO MEMCOMPRESSIMO 中存储无压缩
MEMCOMPRESS FOR DML最小化压缩,优化 DML 操作
MEMCOMPRESS FOR QUERY LOW缺省方式:查询性能最优、空间压缩效果好于DML方式
MEMCOMPRESS FOR QUERY HIGH查询性能次优(excellent)、空间压缩效果好于 QUERY LOW
MEMCOMPRESS FOR CAPACITY LOW查询性能良好(good)、空间压缩效果好于 QUERY HIGH
MEMCOMPRESS FOR CAPACITY HIGH缺省设置、空间压缩效果最优

查看压缩比

col owner for a5
col SEGMENT_NAME for a10
col POPULATE_STATUS for a10
col INMEMORY_COMPRESSION for a10
SQL> SELECT V.OWNER, V.SEGMENT_NAME,V.BYTES/1024/1024 ORIG_SIZE_MB,V.INMEMORY_SIZE/1024/1024 IN_MEM_SIZE_MB,BYTES_NOT_POPULATED,POPULATE_STATUS status,INMEMORY_COMPRESSION,V.BYTES/V.INMEMORY_SIZE COMP_RATIO FROM V$IM_SEGMENTS V WHERE SEGMENT_NAME = 'TEST';

    OWNER SEGMENT_NAME         ORIG_SIZE_MB IN_MEM_SIZE_MB BYTES_NOT_POPULATED STATUS               INMEMORY_COMPRESS COMP_RATIO
----- -------------------- ------------ -------------- ------------------- -------------------- ----------------- ----------
CS    TEST                   1466.78125        110.625                   0 COMPLETED            FOR QUERY LOW     13.2590395

表TEST在磁盘上占用1466MB,采用默认压缩方式,内存中占用110.625MB,压缩比为13.26:1

压缩比测试:
表TEST大小1466.78125MB,数据行9514112

1、NO MEMCOMPRESS:

alter table test inmemory MEMCOMPRESS NO MEMCOMPRESS;
逻辑读:14
物理读:0
COST: 126
时间: 0.01
压缩比:1.19

2、MEMCOMPRESS FOR DML:

alter table test inmemory MEMCOMPRESS FOR DML;
逻辑读:14 
物理读:0
COST: 126
时间:0.01
压缩比:1.22

3、MEMCOMPRESS FOR QUERY LOW:

alter table test inmemory MEMCOMPRESS FOR QUERY LOW;
逻辑读:14
物理读:0
COST:125
时间:0.01
压缩比:7.20

4、MEMCOMPRESS FOR QUERY HIGH:

alter table test inmemory MEMCOMPRESS FOR QUERY high;
逻辑读:14
物理读:0
COST: 125
时间:0.01
压缩比:9.67

5、MEMCOMPRESS FOR CAPACITY LOW:

alter table test inmemory MEMCOMPRESS FOR CAPACITY LOW;
逻辑读:14
物理读:0
COST: 125
时间:0.01
压缩比:12.91

6、MEMCOMPRESS FOR CAPACITY HIGH:

alter table test inmemory MEMCOMPRESS FOR CAPACITY HIGH;
逻辑读:14
物理读:0
COST: 25
时间:0.01
压缩比:19.14

测试结果汇总:

压缩方式压缩比
NO MEMCOMPRESS1.19
MEMCOMPRESS FOR DML1.22
MEMCOMPRESS FOR QUERY LOW7.20
MEMCOMPRESS FOR QUERY HIGH9.67
MEMCOMPRESS FOR CAPACITY LOW12.91
MEMCOMPRESS FOR CAPACITY HIGH19.14

总结:压缩方式不同,表加载到IM中的时间也会不一样。压缩比越大加载到内存中的时间越长。而对于select查询消耗时间不影响。

2.7 查询大量列性能测试

三、评估对象在IMO大小

DBMS_COMPRESSION.GET_COMPRESSION_RATIO
在对一张表使用COMPRESSION clause进行IM压缩级别设置之前,我们可以通过Oracle的COMPRESSION ADVISOR对表放入到IM中的大小进行提前计算。

SET SERVEROUTPUT ON

DECLARE

l_blkcnt_cmp PLS_INTEGER;

l_blkcnt_uncmp PLS_INTEGER;

l_row_cmp PLS_INTEGER;

l_row_uncmp PLS_INTEGER;

l_cmp_ratio PLS_INTEGER;

l_comptype_str VARCHAR2(100);

BEGIN

dbms_compression.get_compression_ratio (

-- Input parameters

scratchtbsname => 'CS',

ownname => 'CS',

objname => 'TEST',

subobjname => NULL,

comptype => dbms_compression.comp_inmemory_QUERY_LOW,

-- Output parameter

blkcnt_cmp => l_blkcnt_cmp,

blkcnt_uncmp => l_blkcnt_uncmp,

row_cmp => l_row_cmp,

row_uncmp => l_row_uncmp,

cmp_ratio => l_cmp_ratio,

comptype_str => l_comptype_str,

subset_numrows => dbms_compression.comp_ratio_allrows);

dbms_output.put_line('Comp. ratio (QUERY LOW):'||l_cmp_ratio);

END;

/

Comp. ratio (QUERY LOW):7
估计结果压缩比7,实际压缩比为7.20

#五、RAC环境测试
rac环境独有参数DUPLICATE clause、DISTRIBUTE clause:

DUPLICATE clause:
此参数为EXADATA一体机专用,在RAC环境中,每个节点拥有自己的IM Area。一个objects根据DUPLICATE clause的设置将一样的数据加载到多个IM Area中。

默认是NO DUPLICATE设置,表示在数据库的IM中对一个objects在所有节点中合起来只保存一份。举例说明,比如三节点的RAC中,对于分区表SALES来讲可能2012年份的数据在1节点,2013年份的数据在2节点,2014年份的数据在3节点,每个分区只保存在一个节点上。

为了提升可用性,也可以设置为DUPLICAET ALL,在每个节点上都保存一份。举例说明,还是刚才那个SALES表的请款下,1,2,3三个节点各保存一份完整sales表数据到各自的IM中。在任意一个节点上都可以获取查询需要的数据。

在设置为DUPLICATE ALL的情况下

DISTRIBUTE clause:

如果一个objects因为太大无法被加载到一个IM Area中,还可以通过DISTRIBUTE clause的设置将它分成几个数据片分别加载到不同的节点中。

默认情况下DISTRIBUTE clause的默认值为AUTO-DISTRIBUTE,这时候是否将objects分布式分布在不同的节点上由Oracle内部算法决定。这个参数对于单实例没有影响,在RAC环境中,默认存在IM中的表会分布在各个节点之中。

RAC环境并不像单实例一样只需修改表的IM属性即可启用,如果要使用IMO,必须在系统层面修改“并行度策略”为自动。下面对参数“parallel_degree_policy”,分别测试2个场景

1、先查看要测试的表TEST是否已加载到IM STORE

col SEGMENT_NAME for a5
col POPULATE_STATUS for a10
col INMEMORY_DISTRIBUTE for a10
col INMEMORY_DUPLICATE for a20

SQL> select INST_ID,SEGMENT_NAME,INMEMORY_SIZE/1024/1024,BYTES/1024/1024,BYTES_NOT_POPULATED/1024/1024,INMEMORY_DISTRIBUTE,INMEMORY_DUPLICATE,POPULATE_STATUS from gv$im_segments;

   INST_ID SEGME INMEMORY_SIZE/1024/1024 BYTES/1024/1024 BYTES_NOT_POPULATED/1024/1024 INMEMORY_D INMEMORY_DUPLICATE   POPULATE_S
---------- ----- ----------------------- --------------- ----------------------------- ---------- -------------------- ----------
         2 TEST                    9.125             192                     123.53125 AUTO       NO DUPLICATE         COMPLETED
         1 TEST                  16.1875             192                    66.9140625 AUTO       NO DUPLICATE         COMPLETED

1、开启AUTO DOP

alter system set parallel_degree_policy=AUTO sid='*';
alter system set parallel_force_local=false sid='*';
alter system flush buffer_cache;
alter system flush shared_pool;

2、查看执行计划

SQL> select /*+ parallel */count(*) from test;

  COUNT(*)
----------
   1455168


Execution Plan
----------------------------------------------------------
Plan hash value: 2661943167

-----------------------------------------------------------------------------------------------------------------
| Id  | Operation                       | Name     | Rows  | Cost (%CPU)| Time     |    TQ  |IN-OUT| PQ Distrib |
-----------------------------------------------------------------------------------------------------------------
|   0 | SELECT STATEMENT                |          |     1 |     9   (0)| 00:00:01 |        |      |            |
|   1 |  SORT AGGREGATE                 |          |     1 |            |          |        |      |            |
|   2 |   PX COORDINATOR                |          |       |            |          |        |      |            |
|   3 |    PX SEND QC (RANDOM)          | :TQ10000 |     1 |            |          |  Q1,00 | P->S | QC (RAND)  |
|   4 |     SORT AGGREGATE              |          |     1 |            |          |  Q1,00 | PCWP |            |
|   5 |      PX BLOCK ITERATOR          |          | 90948 |     9   (0)| 00:00:01 |  Q1,00 | PCWC |            |
|   6 |       TABLE ACCESS INMEMORY FULL| TEST     | 90948 |     9   (0)| 00:00:01 |  Q1,00 | PCWP |            |
-----------------------------------------------------------------------------------------------------------------

Note
-----
   - automatic DOP: Computed Degree of Parallelism is 2
   - parallel scans affinitized for inmemory


Statistics
----------------------------------------------------------
        232  recursive calls
          4  db block gets
        163  consistent gets
         15  physical reads
          0  redo size
        545  bytes sent via SQL*Net to client
        551  bytes received via SQL*Net from client
          2  SQL*Net roundtrips to/from client
         16  sorts (memory)
          0  sorts (disk)
          1  rows processed   
逻辑读:163
物理读:15

3、不使用并行查询查询

SQL> select count(*) from test;

  COUNT(*)
----------
   1455168


Execution Plan
----------------------------------------------------------
Plan hash value: 1950795681

----------------------------------------------------------------------------
| Id  | Operation                   | Name | Rows  | Cost (%CPU)| Time     |
----------------------------------------------------------------------------
|   0 | SELECT STATEMENT            |      |     1 |    16   (0)| 00:00:01 |
|   1 |  SORT AGGREGATE             |      |     1 |            |          |
|   2 |   TABLE ACCESS INMEMORY FULL| TEST | 90948 |    16   (0)| 00:00:01 |
----------------------------------------------------------------------------

Note
-----
   - automatic DOP: Computed Degree of Parallelism is 1 because of no expensive parallel operation


Statistics
----------------------------------------------------------
          3  recursive calls
          0  db block gets
       9081  consistent gets
       9077  physical reads
          0  redo size
        545  bytes sent via SQL*Net to client
        551  bytes received via SQL*Net from client
          2  SQL*Net roundtrips to/from client
          0  sorts (memory)
          0  sorts (disk)
          1  rows processed
逻辑读:9081
物理读:9077

4、关闭AUTO DOP

alter system set parallel_degree_policy=MANUAL sid='*';
alter system set parallel_force_local=false sid='*';
alter system flush buffer_cache;
alter system flush shared_pool;

5、查看执行计划

SQL> select /*+ parallel */count(*) from test;

  COUNT(*)
----------
   1455168


Execution Plan
----------------------------------------------------------
Plan hash value: 2661943167

-----------------------------------------------------------------------------------------------------------------
| Id  | Operation                       | Name     | Rows  | Cost (%CPU)| Time     |    TQ  |IN-OUT| PQ Distrib |
-----------------------------------------------------------------------------------------------------------------
|   0 | SELECT STATEMENT                |          |     1 |     9   (0)| 00:00:01 |        |      |            |
|   1 |  SORT AGGREGATE                 |          |     1 |            |          |        |      |            |
|   2 |   PX COORDINATOR                |          |       |            |          |        |      |            |
|   3 |    PX SEND QC (RANDOM)          | :TQ10000 |     1 |            |          |  Q1,00 | P->S | QC (RAND)  |
|   4 |     SORT AGGREGATE              |          |     1 |            |          |  Q1,00 | PCWP |            |
|   5 |      PX BLOCK ITERATOR          |          | 90948 |     9   (0)| 00:00:01 |  Q1,00 | PCWC |            |
|   6 |       TABLE ACCESS INMEMORY FULL| TEST     | 90948 |     9   (0)| 00:00:01 |  Q1,00 | PCWP |            |
-----------------------------------------------------------------------------------------------------------------

Note
-----
   - automatic DOP: Computed Degree of Parallelism is 2
   - parallel scans affinitized for inmemory


Statistics
----------------------------------------------------------
        209  recursive calls
          4  db block gets
      17909  consistent gets
      17766  physical reads
          0  redo size
        545  bytes sent via SQL*Net to client
        551  bytes received via SQL*Net from client
          2  SQL*Net roundtrips to/from client
         16  sorts (memory)
          0  sorts (disk)
          1  rows processed
    逻辑读:17909
    物理读:17766

总结:
1、调整系统参数(parallel_degree_policy)之后,rac环境下的IM列查询大幅降低了逻辑读与物理读。

2、在多实例的并发查询中实例之间传输的并不是IMCU,而是每个节点都会对本节点的数据运行相同的sql语句,之后把自己的结果集发送给发起sql语句的实例,组成最终的结果返回给用户。

parallel_degree_policyAUTOMANUAL
逻辑读16317909
物理读1517766

备注:
在没有显式使用并行sql时,rac环境im全表扫描并没有使用并行。oracle的优化器会通过一系列的计算比较cost。

SQL> select count(*) from test;
执行计划Note
-----
automatic DOP: 
Computed Degree of Parallelism is 1 because of no expensive parallel operation

In some cases, even with Auto DOP set correctly, the optimizer may calculate the cost of a serial access to be less than the cost of a parallel access. This has been identified as bug 18960760, and will usually only happen when very smalltables in the IM column store are involved in the query.

来自Oracle Blog
https://blogs.oracle.com/in-memory/oracle-database-in-memory-on-rac-part-i

四、参考文献

1、Oracle Database In-Memory on RAC - Part I
https://blogs.oracle.com/in-memory/oracle-database-in-memory-on-rac-part-i

2、Oracle 12c DB In-Memory入门实验手册
https://blog.csdn.net/badly9/article/details/49777993

3、Oracle12c IMO 测试
https://www.jianshu.com/p/966ee0182e1c

4、rac并行查询
http://blog.sina.com.cn/s/blog_74a7d3390102wegl.html

5、Oracle In-Memory白皮书
http://www.oracle.com/technetwork/cn/database/in-memory/overview/twp-oracle-database-in-memory-2245633-zhs.pdf

评论 2
添加红包

请填写红包祝福语或标题

红包个数最小为10个

红包金额最低5元

当前余额3.43前往充值 >
需支付:10.00
成就一亿技术人!
领取后你会自动成为博主和红包主的粉丝 规则
hope_wisdom
发出的红包
实付
使用余额支付
点击重新获取
扫码支付
钱包余额 0

抵扣说明:

1.余额是钱包充值的虚拟货币,按照1:1的比例进行支付金额的抵扣。
2.余额无法直接购买下载,可以购买VIP、付费专栏及课程。

余额充值