分区表、分区索引3(海量数据测试)

建立分区表

create table F_IMEI_CELL_DY_20120508BAK
(
  IMEI_KEY         VARCHAR2(32),
  DAY_KEY          NUMBER(22),
  TACTYPE_KEY      NUMBER(22),
  CELL_KEY         NUMBER(22),
  USER_ID          VARCHAR2(32),
  CELL_ENAME       VARCHAR2(32),
  SCAN_START_TIME  DATE,
  SCAN_STOP_TIME   DATE,
  MNS_TYPE         NUMBER(22),
  GRID_KEY         NUMBER(22),
  CITY_KEY         NUMBER(22),
  GSM_ID           VARCHAR2(32),
  CHRG             NUMBER(22,4) default 0,
  COMM             NUMBER(22,4) default 0,
  NP_MOBILE_SYS    VARCHAR2(32),
  IMEICELL_NUM     NUMBER(22),
  LOCATION_CELLNUM NUMBER(22),
  BTS_KEY          NUMBER(22),
  BTS_ENAME        VARCHAR2(64),
  LOCATION_KEY     NUMBER(22),
  LOCATION_NAME    VARCHAR2(128),
  IF_TD_COVER      NUMBER(22),
  BELONG_CITY_KEY  NUMBER(22),
  PHONESYS_KEY     NUMBER(22)
)
partition by range (SCAN_START_TIME)
(
  partition P_1D_20120401 values less than (TO_DATE(' 2012-04-02 00:00:00', 'SYYYY-MM-DD HH24:MI:SS', 'NLS_CALENDAR=GREGORIAN'))
    tablespace F_TEWMTYPE_CITY_DY
    pctfree 10
    initrans 1
    maxtrans 255
    storage
    (
      initial 64K
      minextents 1
      maxextents unlimited
    ),
  partition P_1D_20120402 values less than (TO_DATE(' 2012-04-03 00:00:00', 'SYYYY-MM-DD HH24:MI:SS', 'NLS_CALENDAR=GREGORIAN'))
    tablespace F_TEWMTYPE_CITY_DY
    pctfree 10
    initrans 1
    maxtrans 255
    storage
    (
      initial 64K
      minextents 1
      maxextents unlimited
    )
  ................
);

汇总测试数据

INSERT INTO casdw.F_IMEI_CELL_DY_20120508bak (IMEI_KEY,USER_ID,CELL_KEY,scan_start_time,scan_stop_time,DAY_KEY,TACTYPE_KEY,
                                                                                                CELL_ENAME,BTS_KEY,BTS_ENAME,MNS_TYPE,GRID_KEY,GSM_ID,CITY_KEY,COMM,IMEICELL_NUM,
                                                                                                IF_TD_COVER,belong_city_key,phonesys_key    )
SELECT V_IMEI_CELL_GPRS_CS_TEST.IMEI,
             V_IMEI_CELL_GPRS_CS_TEST.USR_ID,
             V_IMEI_CELL_GPRS_CS_TEST.CELL_KEY,
             V_IMEI_CELL_GPRS_CS_TEST.scan_start_time,
             max(substr(V_IMEI_CELL_GPRS_CS_TEST.scan_start_time + interval '1' day,1,10)),
          max(V_IMEI_CELL_GPRS_CS_TEST.DAY_KEY),
          max(V_IMEI_CELL_GPRS_CS_TEST.TACTYPE_KEY),
          max(V_IMEI_CELL_GPRS_CS_TEST.CELL_ENAME),
          max(V_IMEI_CELL_GPRS_CS_TEST.BTS_KEY),
          max(V_IMEI_CELL_GPRS_CS_TEST.BTS_ENAME),
          max(V_IMEI_CELL_GPRS_CS_TEST.MNS_TYPE),
          max(V_IMEI_CELL_GPRS_CS_TEST.GRID_KEY),
          max(V_IMEI_CELL_GPRS_CS_TEST.GSM_ID),
             max(V_IMEI_CELL_GPRS_CS_TEST.CITY_KEY),
             sum(NVL(V_IMEI_CELL_GPRS_CS_TEST.COMM,0))/1024/1024,1,
             max(V_IMEI_CELL_GPRS_CS_TEST.IF_TD_COVER),
             max(V_IMEI_CELL_GPRS_CS_TEST.belong_city_key),
             max(V_IMEI_CELL_GPRS_CS_TEST.phonesys_key)
FROM casdw.V_IMEI_CELL_GPRS_CS_TEST
WHERE  scan_start_time = to_date('&1','yyyy-mm-dd')
group by V_IMEI_CELL_GPRS_CS_TEST.scan_start_time, V_IMEI_CELL_GPRS_CS_TEST.CELL_KEY, V_IMEI_CELL_GPRS_CS_TEST.USR_ID, V_IMEI_CELL_GPRS_CS_TEST.IMEI;

汇总语句中select的执行计划

未建索引时汇总入库耗时：48分14秒

[tmn@zhfx1 zhaoxj]$ tail -f 20120410.log
(12-05-09 09:14:00) Start 

old  23: WHERE  scan_start_time = to_date('&1','yyyy-mm-dd')
new  23: WHERE  scan_start_time = to_date('2012-04-10','yyyy-mm-dd')

12177142 rows created.

Commit complete.
(12-05-09 10:02:14) Over 

清空partition (P_1D_20120410)数据

alter table F_IMEI_CELL_DY_20120508bak truncate partition P_1D_20120410

创建local索引

CREATE UNIQUE INDEX F_IMEI_CELL_DY_0508BAK_IDX 
ON F_IMEI_CELL_DY_20120508BAK (SCAN_START_TIME,DAY_KEY, IMEI_KEY, TACTYPE_KEY, CELL_KEY, USER_ID)
TABLESPACE F_TEWMTYPE_CITY_DY_IDX  LOCAL

创建索引后汇总入库耗时:109分26秒

(12-05-09 10:49:03) Start synchro

old  23: WHERE  scan_start_time = to_date('&1','yyyy-mm-dd')
new  23: WHERE  scan_start_time = to_date('2012-04-10','yyyy-mm-dd')

12177142 rows created.

Commit complete.
(12-05-09 12:38:29) Over synchro

INSERT 12177142 条数据时有索引比无索引多用接近一倍时间。注：服务器性能对测试结果有一定影响

数据数量：长沙分区总数(12177142);全表分区总数(25844518)

有索引时查询结果为一条数据时的速度：0.453secends

select cell_ename,grid_key from F_IMEI_CELL_DY_20120508bak partition (P_1D_20120410)
where scan_start_time = to_date('2012-04-10','yyyy-mm-dd') 
      and DAY_key='20120410'
      and IMEI_KEY='5305215221089010'
      and TACTYPE_KEY='27237'
      and CELL_KEY='329312'
      and USER_ID='3110071623961137'

执行计划

有索引时查询结果为一条数据时的速度：328.172secends

select cell_ename,grid_key from F_IMEI_CELL_DY_20120508bak partition (P_1D_20120410)
where scan_start_time = to_date('2012-04-10','yyyy-mm-dd') 
      and DAY_key='20120410'
      and CELL_KEY='329312'
      and USER_ID='3110071623961137'

执行计划

　　表的组合索引顺序为：SCAN_START_TIME,DAY_KEY, IMEI_KEY, TACTYPE_KEY, CELL_KEY, USER_ID。

时间为什么相差这么大？

　　ORACLE采用自下而上的顺序解析WHERE子句,根据这个原理,表之间的连接必须写在其他WHERE条件之前, 那些可以过滤掉最大数量记录的条件必须写在WHERE子句的末尾. 即:最大地减少数据范围的条件写在后面,oracle对sql的解析是从后向前的