一.前言：
一开始分区表和位图索引怎么会挂钩呢？可能现实就是这么的不期而遇；比如说一张表的字段是年月日—‘yyyy-mm-dd’，重复率高吧，适合建位图索引吧，而且这张表数据量也不小，也适合转换成分区表吧！下面我来比较一下分区表和分区字段位图索引的性能！
二.实验
生产上的表结构以及索引：
create table LOT_WIN_RESULT_DETAIL
(
id INTEGER not null,
rpt_date DATE,
sys_game_level_code_id INTEGER,
game_desc VARCHAR2(50),
periods_no VARCHAR2(50),
start_tt VARCHAR2(10),
end_tt VARCHAR2(10),
ok_dtt VARCHAR2(10),
ok_ind CHAR(1),
open_code VARCHAR2(500),
remark VARCHAR2(50),
json_result VARCHAR2(4000),
getdata_dtt TIMESTAMP(6),
last_modfiy_user_id INTEGER,
last_modfiy_user_name VARCHAR2(50),
last_modfiy_dtt TIMESTAMP(6),
platform_ident VARCHAR2(45)
)
tablespace NB_TBS_YOBET
pctfree 10
initrans 1
maxtrans 255
storage
(
initial 24M
next 1M
minextents 1
maxextents unlimited
);
-- Create/Recreate indexes
create bitmap index IX1_LOT_WIN_DTL on LOT_WIN_RESULT_DETAIL (PLATFORM_IDENT) tablespace NB_INX_TBS_YOBET
;
create index IX2_LOT_WIN_DTL on LOT_WIN_RESULT_DETAIL (SYS_GAME_LEVEL_CODE_ID, PERIODS_NO) tablespace NB_INX_TBS_YOBET
;
create bitmap index IX4_LOT_WIN_DTL on LOT_WIN_RESULT_DETAIL (RPT_DATE) tablespace NB_INX_TBS_YOBET
;
create index LX_LOT_WIN_RESULT_DETAIL_TIME on LOT_WIN_RESULT_DETAIL (GETDATA_DTT) tablespace NB_INX_TBS_YOBET
;
-- Create/Recreate primary, unique and foreign key constraints
alter table LOT_WIN_RESULT_DETAIL add constraint PK_LOT_WIN_RESULT_DETAIL primary key (ID) using index tablespace NB_TBS_YOBET
;
模拟建一张一模一样的分区表（ LOT_WIN_RESULT_DETAIL_part），以rpt_date时间字段分区，且把数据量插入过来：
create table LOT_WIN_RESULT_DETAIL_part
(
id INTEGER not null,
rpt_date DATE,
sys_game_level_code_id INTEGER,
game_desc VARCHAR2(50),
periods_no VARCHAR2(50),
start_tt VARCHAR2(10),
end_tt VARCHAR2(10),
ok_dtt VARCHAR2(10),
ok_ind CHAR(1),
open_code VARCHAR2(500),
remark VARCHAR2(50),
json_result VARCHAR2(4000),
getdata_dtt TIMESTAMP(6),
last_modfiy_user_id INTEGER,
last_modfiy_user_name VARCHAR2(50),
last_modfiy_dtt TIMESTAMP(6),
platform_ident VARCHAR2(45)
)PARTITION BY RANGE(RPT_DATE) INTERVAL (NUMTODSINTERVAL(1, 'DAY'))
(
PARTITION RPT_DATE_20191218 VALUES LESS THAN(TO_DATE('2019-12-19', 'YYYY-MM-DD'))
) tablespace NB_TBS_YOBET ;
create bitmap index IX1 on LOT_WIN_RESULT_DETAIL_part (PLATFORM_IDENT)
local tablespace NB_INX_TBS_YOBET
;
create index IX2 on LOT_WIN_RESULT_DETAIL_part (SYS_GAME_LEVEL_CODE_ID, PERIODS_NO)
local tablespace NB_INX_TBS_YOBET
;
create index LX3 on LOT_WIN_RESULT_DETAIL_part (GETDATA_DTT) lcoal tablespace NB_INX_TBS_YOBET
;
-- Create/Recreate primary, unique and foreign key constraints
alter table LOT_WIN_RESULT_DETAIL_part
add constraint PK_LOT_WIN_RESULT_DETAIL_p primary key (ID)
using index
tablespace NB_TBS_YOBET
;
insert into LOT_WIN_RESULT_DETAIL_part select * from LOT_WIN_RESULT_DETAIL;
数据库都是89万。
BEGIN
DBMS_STATS.GATHER_TABLE_STATS( OWNNAME => 'RACTTFC',
TABNAME => 'LOT_WIN_RESULT_DETAIL_part',
CASCADE => TRUE);
END;
/
BEGIN
DBMS_STATS.GATHER_TABLE_STATS( OWNNAME => 'RACTTFC',
TABNAME => 'LOT_WIN_RESULT_DETAIL_part',
CASCADE => TRUE);
END;
/
2.1 时间分区字段范围查询
语句如下：
select rpt_date,
periods_no,
open_code,
json_result,
remark,
ok_dtt as start_tt,
GAME_DESC,
rownums
from (select rpt_date,
periods_no,
a.open_code,
a.json_result,
a.remark,
a.ok_dtt,
a.GAME_DESC,
ROW_NUMBER() OVER(ORDER BY to_number(periods_no) desc) as rownums
from lot_win_result_detail a
where rpt_date >=
to_date('2019-12-23 00:00:00', 'yyyy-mm-dd HH24:mi:ss')
and rpt_date <=
to_date('2019-12-24 23:59:59', 'yyyy-mm-dd HH24:mi:ss')
and sys_game_level_code_id = 5827
and ok_ind >= '1'
and PLATFORM_IDENT = 'af') a
where rownums <= 10
order by to_number(periods_no) desc;
执行计划如下：
采用了bitmap 索引范围扫描，CPU成本3652；
select rpt_date,
periods_no,
open_code,
json_result,
remark,
ok_dtt as start_tt,
GAME_DESC,
rownums
from (select rpt_date,
periods_no,
a.open_code,
a.json_result,
a.remark,
a.ok_dtt,
a.GAME_DESC,
ROW_NUMBER() OVER(ORDER BY to_number(periods_no) desc) as rownums
from lot_win_result_detail_part a
where rpt_date >=
to_date('2019-12-23 00:00:00', 'yyyy-mm-dd HH24:mi:ss')
and rpt_date <=
to_date('2019-12-24 23:59:59', 'yyyy-mm-dd HH24:mi:ss')
and sys_game_level_code_id = 5827
and ok_ind >= '1'
and PLATFORM_IDENT = 'af') a
where rownums <= 10
order by to_number(periods_no) desc；
基于分区过略，且基数2881比2671大，最终消耗的成本1442比3652低的多，且最开始进行的分区表索引范围扫描成本105比非分区表的539 小的多。
2.2时间分区字段等值
select rpt_date,
periods_no,
open_code,
json_result,
remark,
ok_dtt as start_tt,
GAME_DESC,
rownums
from (select rpt_date,
periods_no,
a.open_code,
a.json_result,
a.remark,
a.ok_dtt,
a.GAME_DESC,
ROW_NUMBER() OVER(ORDER BY to_number(periods_no) desc) as rownums
from lot_win_result_detail a
where rpt_date =
to_date('2019-12-23', 'yyyy-mm-dd')
and sys_game_level_code_id = 5827
and ok_ind >= '1'
and PLATFORM_IDENT = 'af') a
where rownums <= 10
order by to_number(periods_no) desc;
单个值的查询的时候，bitmap没有进行范围扫描，进行了单个等值查询，通过索引范围扫描，然后再通过bimap索引转换成ROWID,最后通过又通过bitmap回表，CPU 耗费成本2515；
同理，换成分区表：
select rpt_date,
periods_no,
open_code,
json_result,
remark,
ok_dtt as start_tt,
GAME_DESC,
rownums
from (select rpt_date,
periods_no,
a.open_code,
a.json_result,
a.remark,
a.ok_dtt,
a.GAME_DESC,
ROW_NUMBER() OVER(ORDER BY to_number(periods_no) desc) as rownums
from lot_win_result_detail_part a
where rpt_date =
to_date('2019-12-23', 'yyyy-mm-dd')
and sys_game_level_code_id = 5827
and ok_ind >= '1'
and PLATFORM_IDENT = 'af') a
where rownums <= 10
order by to_number(periods_no) desc；
整个执行计划是走的单个分区，通过索引范围扫描，然后再通过bimap索引转换成ROWID,最后通过又通过bitmap回表，CPU 耗费成本763，远远小于2515，且单个的索引扫描51 远远小于上述的551索引范围扫描（这个是分区表利用本地索引的优势）；
整个两次对比结果bitmap无论是在范围查询还是单个的等值查询都是完败。
2.3 分区表分区索引变成全局索引
drop index IX222;
create index IX222 on LOT_WIN_RESULT_DETAIL_part (SYS_GAME_LEVEL_CODE_ID, PERIODS_NO) tablespace NB_INX_TBS_YOBET;
BEGIN
DBMS_STATS.GATHER_TABLE_STATS( OWNNAME => 'RACTTFC',
TABNAME => 'LOT_WIN_RESULT_DETAIL_part',
CASCADE => TRUE);
END;
/
select rpt_date,
periods_no,
open_code,
json_result,
remark,
ok_dtt as start_tt,
GAME_DESC,
rownums
from (select rpt_date,
periods_no,
a.open_code,
a.json_result,
a.remark,
a.ok_dtt,
a.GAME_DESC,
ROW_NUMBER() OVER(ORDER BY to_number(periods_no) desc) as rownums
from lot_win_result_detail a
where rpt_date >=
to_date('2019-12-23 00:00:00', 'yyyy-mm-dd HH24:mi:ss')
and rpt_date <=
to_date('2019-12-24 23:59:59', 'yyyy-mm-dd HH24:mi:ss')
and sys_game_level_code_id = 5827
and ok_ind >= '1'
and PLATFORM_IDENT = 'af') a
where rownums <= 10
order by to_number(periods_no) desc;
分区范围内的全表扫描，并没有用到新建的哪个索引，CPU成本4587，远高于bitmap的3652；
单个值的查询：
select rpt_date,
periods_no,
open_code,
json_result,
remark,
ok_dtt as start_tt,
GAME_DESC,
rownums
from (select rpt_date,
periods_no,
a.open_code,
a.json_result,
a.remark,
a.ok_dtt,
a.GAME_DESC,
ROW_NUMBER() OVER(ORDER BY to_number(periods_no) desc) as rownums
from lot_win_result_detail_part a
where rpt_date =date'2019-12-23'
and sys_game_level_code_id = 5827
and ok_ind >= '1'
and PLATFORM_IDENT = 'af') a
where rownums <= 10
order by to_number(periods_no) desc;
同理一样的全局索引没有使用，2193和2515是一个数量级的，差距不是很明显。
三.结论
综上所述，bitmap 在进行等值与以及范围查询的时候，整个执行过程大致一样，但是，主要是分区表能够在分区范围内利用本地索引进行扫描（全局索引几乎是几个量级），非分区表没有这种优势；
索引在rang范围查询字段方面。建议使用分区加local 索引，会造成BITMAP INDEX RANGE SCAN并没有PARTITION RANGE ITERATOR 加local 索引高效。