Oralce多个远程库做关联查询的优化 Mr

原数据SQL

SELECT substr(A.NE_NAME, 1, 3) city,
       ROUND(sum(A.COUNTER_1279177418 * 23 + A.COUNTER_1279177419 * 34 +
                 COUNTER_1279177420 * 40 + COUNTER_1279177421 * 54) / 1024 / 1024 / 1024,
             2) +
       ROUND(sum(B.COUNTER_1279178418 * 23 + B.COUNTER_1279178419 * 34 +
                 COUNTER_1279178420 * 40 + COUNTER_1279178421 * 54) / 1024 / 1024 / 1024,
             2) GPRS_RLC_DL,
       ROUND(sum(COUNTER_1279179418 * 22 + COUNTER_1279179419 * 28 +
                 COUNTER_1279179420 * 37 + COUNTER_1279179421 * 44 +
                 COUNTER_1279179422 * 56 + COUNTER_1279179423 * 74 +
                 COUNTER_1279179424 * 56 + COUNTER_1279179425 * 68 +
                 COUNTER_1279179426 * 74) / 1024 / 1024 / 1024,
             YUA 2) +
       ROUND(sum(COUNTER_1279180418 * 22 + COUNTER_1279180419 * 28 +
                 COUNTER_1279180420 * 37 + COUNTER_1279180421 * 44 +
                 COUNTER_1279180422 * 56 + COUNTER_1279180423 * 74 +
                 COUNTER_1279180424 * 56 + COUNTER_1279180425 * 68 +
                 COUNTER_1279180426 * 74) / 1024 / 1024 / 1024,
             2) EDGE_RLC_DL
  FROM (select scan_start_time,
               MO,
               NE_NAME,
               COUNTER_1279177418,
               COUNTER_1279177419,
               COUNTER_1279177420,
               COUNTER_1279177421
          from rcounterdb.OP_HUW_ULT_1275072525_3
         where scan_start_time >=
               TO_DATE(TO_CHAR(SYSDATE - 1, 'YYYY-MM-DD'), 'YYYY-MM-DD')
           and scan_start_time <
               TO_DATE(TO_CHAR(SYSDATE, 'YYYY-MM-DD'), 'YYYY-MM-DD')
           and to_char(scan_start_time, 'hh24') in
               ('10', '08', '09', '18', '19', '20')) A,
       
       (select scan_start_time,
               MO,
               NE_NAME,
               COUNTER_1279178418,
               COUNTER_1279178419,
               COUNTER_1279178420,
               COUNTER_1279178421
          from rcounterdb.OP_HUW_ULT_1275072526_3
         where scan_start_time >=
               TO_DATE(TO_CHAR(SYSDATE - 1, 'YYYY-MM-DD'), 'YYYY-MM-DD')
           and scan_start_time <
               TO_DATE(TO_CHAR(SYSDATE, 'YYYY-MM-DD'), 'YYYY-MM-DD')
           and to_char(scan_start_time, 'hh24') in
               ('10', '08', '09', '18', '19', '20')) B,
       
       (select scan_start_time,
               MO,
               NE_NAME,
               COUNTER_1279179418,
               COUNTER_1279179419,
               COUNTER_1279179420,
               COUNTER_1279179421,
               COUNTER_1279179422,
               COUNTER_1279179423,
               COUNTER_1279179424,
               COUNTER_1279179425,
               COUNTER_1279179426
          from rcounterdb.OP_HUW_ULT_1275072527_3
         where scan_start_time >=
               TO_DATE(TO_CHAR(SYSDATE - 1, 'YYYY-MM-DD'), 'YYYY-MM-DD')
           and scan_start_time <
               TO_DATE(TO_CHAR(SYSDATE, 'YYYY-MM-DD'), 'YYYY-MM-DD')
           and to_char(scan_start_time, 'hh24') in
               ('10', '08', '09', '18', '19', '20')) C,
       
       (select scan_start_time,
               MO,
               NE_NAME,
               COUNTER_1279180418,
               COUNTER_1279180419,
               COUNTER_1279180420,
               COUNTER_1279180421,
               COUNTER_1279180422,
               COUNTER_1279180423,
               COUNTER_1279180424,
               COUNTER_1279180425,
               COUNTER_1279180426
          from rcounterdb.OP_HUW_ULT_1275072528_3
         where scan_start_time >=
               TO_DATE(TO_CHAR(SYSDATE - 1, 'YYYY-MM-DD'), 'YYYY-MM-DD')
           and scan_start_time <
               TO_DATE(TO_CHAR(SYSDATE, 'YYYY-MM-DD'), 'YYYY-MM-DD')
           and to_char(scan_start_time, 'hh24') in
               ('10', '08', '09', '18', '19', '20')) G
 WHERE a.scan_start_time = B.scan_start_time
   AND a.scan_start_time = C.scan_start_time
   AND A.scan_start_time = G.scan_start_time
   AND A.MO = B.MO
   AND a.MO = C.MO
   AND A.MO = G.MO
   AND A.NE_NAME = B.NE_NAME
   AND a.NE_NAME = C.NE_NAME
   AND A.NE_NAME = G.NE_NAME
 group by substr(A.NE_NAME, 1, 3)

分析一下这条SQL的基本操作是将远程数据库的表进行相应的关联查询，分析性能的损失可能是在远程做链接操作所以将远程的表先搬到本地，然后进行相应的关联操作
通过这种方式所进行的优化方案

--A数据信息提取(39s)
create table sniper1 
nologging
parallel (degree 4) 
as 
SELECT A.COUNTER_1279177418,
       A.COUNTER_1279177419,
       a.COUNTER_1279177420,
       a.COUNTER_1279177421,
       a.scan_start_time,
       a.MO,
       a.ne_name
  from rcounterdb.OP_HUW_ULT_1275072525_3 a
 where 1 = 1
   and to_char(a.scan_start_time, 'hh24') in
       ('10', '08', '09', '18', '19', '20')
   AND a.scan_start_time >= TO_DATE('2012-7-1', 'YYYY-MM-DD')
   and a.scan_start_time < TO_DATE('2012-8-1', 'YYYY-MM-DD')
--B信息数据提取(4min)
create table sniper1 
create table sniper2 
nologging
parallel (degree 4) 
as 
select B.COUNTER_1279178418,
       B.COUNTER_1279178419,
       b.COUNTER_1279178420,
       b.COUNTER_1279178421,
       b.scan_start_time,
       b.mo,
       b.ne_name
  from rcounterdb.OP_HUW_ULT_1275072526_3 B
--C信息数据提取(5min)
 create table sniper3 
nologging
parallel (degree 4) 
 as 
 select
 c.COUNTER_1279179418 ,COUNTER_1279179419 ,
          c.COUNTER_1279179420 ,c.COUNTER_1279179421, c.COUNTER_1279179422 ,
         c.COUNTER_1279179423 ,COUNTER_1279179424,COUNTER_1279179425,
          c.COUNTER_1279179426,c.scan_start_time,c.mo,c.ne_name
from
 rcounterdb.OP_HUW_ULT_1275072527_3 C
--信息数据提取(5min)
create table sniper4
nologging
parallel (degree 4) 
as 
select g.COUNTER_1279180418,
       g.COUNTER_1279180419,
       g.COUNTER_1279180420,
       g.COUNTER_1279180421,
       g.COUNTER_1279180422,
       g.COUNTER_1279180423,
       g.COUNTER_1279180424,
       g.COUNTER_1279180425,
       g.COUNTER_1279180426,
       g.scan_start_time,
       g.mo,
       g.ne_name
  from rcounterdb.OP_HUW_ULT_1275072528_3 G
--合并数据量大的中间部分
SELECT substr(A.NE_NAME, 1, 3) city,
       ROUND(sum(A.COUNTER_1279177418 * 23 + A.COUNTER_1279177419 * 34 +
                 COUNTER_1279177420 * 40 + COUNTER_1279177421 * 54) / 1024 / 1024 / 1024,
             2) +
       ROUND(sum(B.COUNTER_1279178418 * 23 + B.COUNTER_1279178419 * 34 +
                 COUNTER_1279178420 * 40 + COUNTER_1279178421 * 54) / 1024 / 1024 / 1024,
             2) GPRS_RLC_DL,
       ROUND(sum(COUNTER_1279179418 * 22 + COUNTER_1279179419 * 28 +
                 COUNTER_1279179420 * 37 + COUNTER_1279179421 * 44 +
                 COUNTER_1279179422 * 56 + COUNTER_1279179423 * 74 +
                 COUNTER_1279179424 * 56 + COUNTER_1279179425 * 68 +
                 COUNTER_1279179426 * 74) / 1024 / 1024 / 1024,
             2) +
       ROUND(sum(COUNTER_1279180418 * 22 + COUNTER_1279180419 * 28 +
                 COUNTER_1279180420 * 37 + COUNTER_1279180421 * 44 +
                 COUNTER_1279180422 * 56 + COUNTER_1279180423 * 74 +
                 COUNTER_1279180424 * 56 + COUNTER_1279180425 * 68 +
                 COUNTER_1279180426 * 74) / 1024 / 1024 / 1024,
             2) EDGE_RLC_DL
  FROM sniper1 A, sniper2 B, sniper3 C, sniper4 G
 WHERE a.scan_start_time = B.scan_start_time
   AND a.scan_start_time = C.scan_start_time
   AND A.scan_start_time = G.scan_start_time
   AND A.MO = B.MO
   AND a.MO = C.MO
   AND A.MO = G.MO
   AND A.NE_NAME = B.NE_NAME
   AND a.NE_NAME = C.NE_NAME
   AND A.NE_NAME = G.NE_NAME
 group by substr(A.NE_NAME, 1, 3)

调整后的SQL变为

select X.CITY_NAME,
       X.CITY_ID,
       X.city_code,
       nvl(GSM网数据总流量GB, 0) + nvl(EDGE_RLC_DL + gprs_RLC_DL, 0) all_gprs,
       nvl(GSM网EDGE数据总流量GB, 0) + nvl(EDGE_RLC_DL, 0) EDGE,
       ROUND((nvl(GSM网EDGE数据总流量GB, 0) + nvl(EDGE_RLC_DL, 0)) /
             (nvl(GSM网数据总流量GB, 0) + nvl(EDGE_RLC_DL + gprs_RLC_DL, 0)) * 100,
             2) EDGE流量占比
  FROM (select c.city_name 城市,
               C.CITY_ID,
               c.city_desc city_code,
               ROUND((SUM(nvl(gprs_rlc_up, 0)) + SUM(nvl(gprs_rlc_dl, 0)) +
                     SUM(nvl(edge_rlc_up, 0)) + SUM(nvl(edge_rlc_dl, 0))) / 1024 / 1024,
                     2) GSM网数据总流量GB,
               ROUND((SUM(nvl(edge_rlc_up, 0)) + SUM(nvl(edge_rlc_dl, 0))) / 1024 / 1024,
                     2) GSM网EDGE数据总流量GB,
               ROUND((SUM(nvl(edge_rlc_up, 0)) + SUM(nvl(edge_rlc_dl, 0))) /
                     (SUM(nvl(gprs_rlc_up, 0)) + SUM(nvl(gprs_rlc_dl, 0)) +
                     SUM(nvl(edge_rlc_up, 0)) + SUM(nvl(edge_rlc_dl, 0))) * 100,
                     2) EDGE数量流量占比
          from unite.upr_c2_ne_vendor T, config.uc_citys c
         where t.city_code = c.city_code
           and to_char(t.GENTIME, 'hh24') in
               ('10', '08', '09', '18', '19', '20')
           and t.gentime >= TO_DATE('2012-7-1', 'YYYY-MM-DD')
           and t.gentime < TO_DATE('2012-8-1', 'YYYY-MM-DD')
           and t.vendor_name != 'HUAWEI'
         GROUP BY c.city_name, C.CITY_ID, c.city_desc) m,
       (SELECT substr(A.NE_NAME, 1, 3) city,
               ROUND(sum(A.COUNTER_1279177418 * 23 +
                         A.COUNTER_1279177419 * 34 + COUNTER_1279177420 * 40 +
                         COUNTER_1279177421 * 54) / 1024 / 1024 / 1024,
                     2) +
               ROUND(sum(B.COUNTER_1279178418 * 23 +
                         B.COUNTER_1279178419 * 34 + COUNTER_1279178420 * 40 +
                         COUNTER_1279178421 * 54) / 1024 / 1024 / 1024,
                     2) GPRS_RLC_DL,
               ROUND(sum(COUNTER_1279179418 * 22 + COUNTER_1279179419 * 28 +
                         COUNTER_1279179420 * 37 + COUNTER_1279179421 * 44 +
                         COUNTER_1279179422 * 56 + COUNTER_1279179423 * 74 +
                         COUNTER_1279179424 * 56 + COUNTER_1279179425 * 68 +
                         COUNTER_1279179426 * 74) / 1024 / 1024 / 1024,
                     2) +
               ROUND(sum(COUNTER_1279180418 * 22 + COUNTER_1279180419 * 28 +
                         COUNTER_1279180420 * 37 + COUNTER_1279180421 * 44 +
                         COUNTER_1279180422 * 56 + COUNTER_1279180423 * 74 +
                         COUNTER_1279180424 * 56 + COUNTER_1279180425 * 68 +
                         COUNTER_1279180426 * 74) / 1024 / 1024 / 1024,
                     2) EDGE_RLC_DL
          FROM sniper1 A, sniper2 B, sniper3 C, sniper4 G
         WHERE a.scan_start_time = B.scan_start_time
           AND a.scan_start_time = C.scan_start_time
           AND A.scan_start_time = G.scan_start_time
           AND A.MO = B.MO
           AND a.MO = C.MO
           AND A.MO = G.MO
           AND A.NE_NAME = B.NE_NAME
           AND a.NE_NAME = C.NE_NAME
           AND A.NE_NAME = G.NE_NAME
         group by substr(A.NE_NAME, 1, 3)) b,
       CONFIG.UC_CITYS X
 WHERE X.CITY_DESC = B.city(+)
   and x.city_id != 0
   AND X.CITY_ID = M.CITY_ID(+)
 order by x.city_id

调整后SQL的执行时间减少了一半左右。