exists 和 in的性能比较

 测试数据量 TABLE_A 表数据1亿5千万  子查询结果数据量是1百多万
 测试环境oracle11g 单实例环境


结论: 在大数据量的情况下优先考虑exists，而不是in，exists的性能大于in的性能 

 
 1.---in + union --- --两次用时 208.231s    113s
SELECT count(*) FROM TABLE_A   WHERE  (ACNO,DAT,LOGACNO,SENO,PROVICEID) IN
 (
SELECT  trim(ACNO) as ACNO,DAT,LOGACNO,SENO,PROVICEID  FROM TABLE_R  UNION
SELECT  trim(ACNO) as ACNO,DAT,LOGACNO,SENO,PROVICEID FROM TABLE_BAL UNION

SELECT  trim(ORI_ACNO) as ACNO,ORI_DAT,ORI_LOGACNO,ORI_SENO,PROVICEID FROM TABLE_BAL ) ; 

2.---in + union all--- --两次用时 89.496s   109.068s
SELECT count(*) FROM TABLE_A   WHERE  (ACNO,DAT,LOGACNO,SENO,PROVICEID) IN
 (
SELECT  trim(ACNO) as ACNO,DAT,LOGACNO,SENO,PROVICEID  FROM TABLE_R  UNION ALL
SELECT  trim(ACNO) as ACNO,DAT,LOGACNO,SENO,PROVICEID FROM TABLE_BAL UNION ALL
SELECT  trim(ORI_ACNO) as ACNO,ORI_DAT,ORI_LOGACNO,ORI_SENO,PROVICEID FROM TABLE_BAL ) ;  

3.---in + union all + distinct --- --两次用时 85.863s   113.805s
SELECT count(*) FROM TABLE_A   WHERE  (ACNO,DAT,LOGACNO,SENO,PROVICEID) IN
 ( SELECT DISTINCT * FROM (
SELECT  trim(ACNO) as ACNO,DAT,LOGACNO,SENO,PROVICEID  FROM TABLE_R  UNION ALL
SELECT  trim(ACNO) as ACNO,DAT,LOGACNO,SENO,PROVICEID FROM TABLE_BAL UNION ALL 
SELECT  trim(ORI_ACNO) as ACNO,ORI_DAT,ORI_LOGACNO,ORI_SENO,PROVICEID FROM TABLE_BAL )) ; 

4.--- EXISTS + union all + distinct ---  --两次用时8.707s   6.551s
SELECT count(*) FROM TABLE_A A  WHERE EXISTS (
SELECT  *  FROM (
SELECT  trim(ACNO) as ACNO,DAT,LOGACNO,SENO,PROVICEID  FROM TABLE_R  UNION ALL
SELECT  trim(ACNO ) as ACNO,DAT,LOGACNO,SENO,PROVICEID FROM TABLE_BAL  UNION ALL
SELECT  trim(ORI_ACNO) as ACNO,ORI_DAT,ORI_LOGACNO,ORI_SENO,PROVICEID FROM TABLE_BAL  ) B
 WHERE  A.ACNO = B.ACNO AND A.DAT = B.DAT AND A.LOGACNO = B.LOGACNO AND A.SENO=B.SENO AND A.PROVICEID = B.PROVICEID) ;