Query running slow after upgrading

Question

Hi,
We have one query that is running very slow in the 10.2.0.4 database after upgrading from 9.2.0.8 HP-UX B11.31. In 9.2.0.8 database it used to take 5 mins. But in the new 10.2.0.4 database it takes around 36 mins.The query is given below.
SELECT a.transaction_date,
                a.ORD_PROD_BE_ID,
                a.SLS_POSTD_DT_BE_ID,
                 a.net_trd_amt,
                a.trd_actl_un_qty,
                t.FISC_MO_CD, 
                t.FY_CD
FROM fact_net_trd_sls a,
            dim_tm_mv t,
            dim_prod b,
            (SELECT DISTINCT kit_prod_cd FROM kal_kit_bom) c
WHERE b.be_id = a.ORD_PROD_BE_ID
AND b.end_date >SYSDATE
AND t.be_id = a.SLS_POSTD_DT_BE_ID
AND t.end_date > SYSDATE
AND c.kit_prod_cd (+)= b.base_prod_cd
AND t.FY_CD IN (SELECT DISTINCT FY_CD FROM DIM_tm_MV T WHERe T.DAY_STRT_PRD_OF_TM=TRUNC(SYSDATE))
and nvl2(c.kit_prod_cd,'K','FG') = 'FG'
ORDER BY  a.ORD_PROD_BE_ID,a.TRANSACTION_DATE;The plan in 9i is below,
PLAN_TABLE_OUTPUT
| Id  | Operation                           |  Name                  | Rows  | Bytes |TempSpc| Cost  |
|   0 | SELECT STATEMENT                    |                        |  2861K|  1585M|       |   170K|
|   1 |  VIEW                               |                        |  2861K|  1585M|       |   170K|
|   2 |   SORT UNIQUE                       |                        |  2861K|   654M|  1397M|   170K|
|*  3 |    FILTER                           |                        |       |       |       |       |
|*  4 |     HASH JOIN OUTER                 |                        |       |       |       |       |
|*  5 |      HASH JOIN                      |                        |  1998K|   445M|  4056K| 63419 |
|*  6 |       TABLE ACCESS BY INDEX ROWID   | DIM_PROD               | 53189 |  3428K|       |   283 |
|   7 |        BITMAP CONVERSION TO ROWIDS  |                        |       |       |       |       |
|   8 |         BITMAP INDEX FULL SCAN      | XN3_DIM_PROD           |       |       |       |       |
|   9 |       TABLE ACCESS BY INDEX ROWID   | FACT_NET_TRD_SLS       |  4081 |   203K|       |   118 |
|  10 |        NESTED LOOPS                 |                        |  1998K|   320M|       | 57607 |
|* 11 |         HASH JOIN SEMI              |                        |   490 | 57330 |       |   208 |
|* 12 |          TABLE ACCESS FULL          | DIM_TM_MV              | 15180 |  1260K|       |   195 |
|  13 |          TABLE ACCESS BY INDEX ROWID| DIM_TM_MV              |     1 |    32 |       |     1 |
|* 14 |           INDEX RANGE SCAN          | XN4_DIM_TM_MV          |     1 |       |       |     1 |
|* 15 |         INDEX RANGE SCAN            | FACT_NET_TRD_SLS_IDX3  |  4081 |       |       |    54 |
|  16 |      TABLE ACCESS FULL              | KAL_KIT_BOM            | 14175 | 85050 |       |    24 |
Predicate Information (identified by operation id):
   3 - filter(NVL2("KAL_KIT_BOM"."KIT_PROD_CD",'K','FG')='FG')
   4 - access("B"."BASE_PROD_CD"=("KAL_KIT_BOM"."KIT_PROD_CD"(+)))
   5 - access("B"."BE_ID"="A"."ORD_PROD_BE_ID")
   6 - filter("B"."END_DATE">SYSDATE@!)
  11 - access("SYS_ALIAS_0000"."FY_CD"="T"."FY_CD")
  12 - filter("SYS_ALIAS_0000"."END_DATE">SYSDATE@!)
  14 - access("T"."DAY_STRT_PRD_OF_TM"=TRUNC(SYSDATE@!))
  15 - access("SYS_ALIAS_0000"."BE_ID"="A"."SLS_POSTD_DT_BE_ID")
Note: cpu costing is off
36 rows selected.The plan in 10g is below.
Operation           Object Name     Rows     Bytes     Cost       Object Node      In/Out  PStart    PStop
SELECT STATEMENT Optimizer Mode=ALL_ROWS                             34                           1382                                                       
  VIEW                   34           19 K        1382                                                       
    SORT UNIQUE                               34           7 K          1382                                                       
      FILTER                                                                                                                             
        HASH JOIN OUTER                   34           7 K          1375                                                       
          MAT_VIEW ACCESS BY INDEX ROWID           WHSUSR.DIM_TM_MV 1              81           1                                                              
            NESTED LOOPS                      24           5 K          1350                                                       
              NESTED LOOPS                    740         105 K     1336                                                       
                MERGE JOIN CARTESIAN                              1              95           1332                                                       
                  TABLE ACCESS BY INDEX ROWID              WHSUSR.DIM_PROD      52 K        3 M        279                                                         
                    BITMAP CONVERSION TO ROWIDS                                                                                                                      
                      BITMAP INDEX FULL SCAN     WHSUSR.XN3_DIM_PROD                                                                                                           
                  BUFFER SORT                   1              30           9223372036 G                                                   
                    SORT UNIQUE                               1              30           1                                                              
                      MAT_VIEW ACCESS BY INDEX ROWID               WHSUSR.DIM_TM_MV 1              30           1                                                                           
                        INDEX RANGE SCAN               WHSUSR.XN4_DIM_TM_MV      1                              1                                                              
                TABLE ACCESS BY INDEX ROWID                TRANSDATA.FACT_NET_TRD_SLS             740         36 K        4                                                                    
                  INDEX RANGE SCAN     TRANSDATA.FACT_NET_TRD_SLS_IDX2 2 K                         1                                                              
              INDEX RANGE SCAN         WHSUSR.XN1_DIM_TM_MV      1                              1                                                              
          TABLE ACCESS FULL               EDW_DBA.KAL_KIT_BOM            14 K        83 K        24                                                            We have checked that the statistics are uptodate. Please help with suggestions.
Thanks in advane.

Answer 1

Most likely, the index stats are not up-to-date, please double check.
h5. In 9i, FACT_NET_TRD_SLS_IDX3 is used to scan the fact.
Time dimension is used to filter out most records in fact table first, then join to Product dimension.
h5. In 10g, FACT_NET_TRD_SLS_IDX2 is used to scan the fact.
Time dimension "MERGE JOIN CARTESIAN" to Product dimension first to form a data set, and then use it to filter the fact table.
If the stats for FACT_NET_TRD_SLS_IDX2 let ORACLE believe that much less records will remain/survive after the join, 10g will use it.
It seems to me that cartesian join Time x Product is going to generate a much bigger data set than estimated, then the nested loop after that will take a long time to iterate through.
Why don't you partition your fact table BY RANGE(SLS_POSTD_DT_BE_ID) to make it simpler?