Join SQL Statement

Question

Hello all,
Good day.
I've one question on how to create the optimal index to support the below JOIN statement.
SELECT *      
FROM
     "TABLE1" T_1 INNER JOIN "TABLE2" T_2
     ON T_1.FIELD_JOIN1 = T_2. FIELD_JOIN1 AND
     ON T_1.FIELD_JOIN2 = T_2. FIELD_JOIN2
WHERE
     T_1.FIELD_WHERE3 and T_1. FIELD_WHERE4 and T_2. FIELD_WHERE5 and T_2. FIELD_WHERE6
I noticed that in many cases, there will be 2 indexes are created on each table: one index to support the fields in JOIN Condition and the other to support the WHERE Clause.
My initial thought is to combine the fields in the WHERE clause and JOIN Condition into one secondary index in order to make the index a 'covering index'.
My proposed indexes:
Index on TABLE1: MANDT, FIELD_WHERE3, FIELD_WHERE4, FIELD_JOIN1, FIELD_JOIN2
Index on TABLE2: MANDT, FIELD_WHERE5, FIELD_WHERE6, FIELD_JOIN1, FIELD_JOIN2
Let say all the fields in the JOIN Condition and WHERE Clause are very selective, is this recommended if we
1) Create an index which contains the fields for both JOIN condition and WHERE clause?
2) and the fields of the index will follow the above sequence, meaning the fields in WHERE clause are located at the beginning of the index?
Thanks,
KP

Answer 1

Hi Volker, Hi Harald,
Thanks for your reply.
I wrote one SQL statement and tested it in ST05 and the execution plan is as follow.
- how many rows in MSEG: 36,000,000 
- how many rows on WHERE MSEG: 4,000  << Smaller
- how many rows in MKPF: 5,700,000 
- how many rows on WHERE MKPF: 3,200,000
--> No of entries after joining the above tables plus the 'where' selection: 62 records
SELECT
FROM
  "MKPF" T_00
      INNER JOIN "MSEG" T_01
            ON T_01 . "MANDT" = '100' AND
                   T_00 . "MANDT" = T_01 . "MANDT" AND
                   T_00 . "MBLNR" = T_01 . "MBLNR" AND
                   T_00 . "MJAHR" = T_01 . "MJAHR"                           
WHERE
      T_00 . "MANDT" = '100' AND
      T_01 . "MATNR" = '000000111216070006' AND
      T_01 . "WERKS" = '1201' AND 
      T_00 . "BUDAT" >= '20090401'
From the execution plan above, the Outer table is MSEG and Nested Loop Join method is being used. I think correct access plan is used as only 62 records returned.
SELECT STATEMENT ( Estimated Costs = 19 , Estimated #Rows = 182 )
5 NESTED LOOPS
          ( Estim. Costs = 19 , Estim. #Rows = 182 )
            Estim. CPU-Costs = 1,646,549 Estim. IO-Costs = 19
2 TABLE ACCESS BY INDEX ROWID MSEG
              ( Estim. Costs = 18 , Estim. #Rows = 363 )
              Estim. CPU-Costs = 412,866 Estim. IO-Costs = 18
1 INDEX RANGE SCAN MSEG~M
( Estim. Costs = 1 , Estim. #Rows = 363 )
Search Columns: 3
Estim. CPU-Costs = 23,106 Estim. IO-Costs = 1
Access Predicates
4 TABLE ACCESS BY INDEX ROWID MKPF
Estim. CPU-Costs = 3,399 Estim. IO-Costs = 0
Filter Predicates
3 INDEX RANGE Scan MKPF~3
I think there is a room to further improve the above statement by
1) Append the fields in the JOIN condition to the index MSEG~M, so the field sequence of this index would be
     MANDT, MATNR , WERKS, MBLNR, MJAHR 
2) Append the fields (BUDAT) in the WHERE condition to the index MSEG~M, so the field sequence of this index would be
     MANDT, BUDAT, MBLNR, MJAHR      (The distinct number of field BUDAT is 840)
     --> I doubt that whether to put BUDAT at the beginning of the index because operator GE ">=" is used with BUDAT. Do you have any comment on this?
3) Let's assume BUDAT is used with Equal operator "=", will this field sequence look fine?
    MANDT, BUDAT, MBLNR, MJAHR
Your comments will be much appreciated. Thank you.
Best regards,
KP