SQL-Tuning Tips

Question

Hi there,
i need some general (i can't supply you with an execution plan) advice in tuning this statement running in 9i CBO mode:
SELECT
      TAB_C1.CLIENT_NR, TAB_C1.DD_NR, last_day(trunc(TAB_C1.BELEGDAT)),
      0, TAB_C2.VALUE_DIM ,
      nvl(SUM(DECODE(TAB_A.FOLGBU || TAB_A.ZUABNEUT,'0A',0,'1A',0,'0Z',1,'1Z',0) * TAB_C2.VALUE),0) ,
      nvl(SUM(DECODE(TAB_A.FOLGBU || TAB_A.ZUABNEUT,'0A',1,'1A',0,'0Z',0,'1Z',0) * TAB_C2.VALUE),0) ,
      nvl(SUM(DECODE(TAB_A.FOLGBU || TAB_A.ZUABNEUT,'0A',0,'1A',0,'0Z',0,'1Z',1) * TAB_C2.VALUE),0) ,
      nvl(SUM(DECODE(TAB_A.FOLGBU || TAB_A.ZUABNEUT,'0A',0,'1A',1,'0Z',0,'1Z',0) * TAB_C2.VALUE),0) ,
      0 ,max(TAB_C1.ROWSEQ * 1000) + TAB_C2.BLENDTYPE * 100 + TAB_C2.VALUE_DIM
  FROM TAB_C1,
       TAB_A,
       TAB_C2,
       TAB_C3,
       TAB_B
WHERE TAB_C1.ACC_TYPE = 'N'
       AND TAB_C1.CANCEL = 0
       AND TAB_A.CLIENT_NR = TAB_C1.CLIENT_NR
       AND TAB_A.ACC_MODE = TAB_C1.ACC_MODE
       AND TAB_C3.CLIENT_NR = TAB_C1.CLIENT_NR
       AND TAB_C3.ACC_TYPE = TAB_C1.ACC_TYPE
       AND TAB_C3.ACC_NR = TAB_C1.ACC_NR
       AND TAB_C3.POS = TAB_C1.POS
       AND TAB_C3.ACC_DAT = TAB_C1.ACC_DAT
       AND TAB_C2.CLIENT_NR = TAB_C1.CLIENT_NR
       AND TAB_C2.ACC_TYPE = TAB_C1.ACC_TYPE
       AND TAB_C2.ACC_NR = TAB_C1.ACC_NR
       AND TAB_C2.POS = TAB_C1.POS
       AND TAB_C2.ACC_DAT = TAB_C1.ACC_DAT
       AND TAB_B.CLIENT_NR = TAB_C2.CLIENT_NR
       AND TAB_B.AMOUNT_KEY = TAB_C2.VALUE_DIM
       AND TAB_B.AMOUNT_TYPE = 0
GROUP BY TAB_C1.CLIENT_NR,
          releasetype,
          releaseid,
          releasegrp,
          releaseseq,
          last_day(trunc(TAB_C1.BELEGDAT)),
          blendtype,
          VALUE_DIM;
Table           Rows     PK
TAB_A          300     CLIENT_NR, ACC_MODE
TAB_B          600     CLIENT_NR, AMOUNT_KEY
TAB_C1          200.000     CLIENT_NR, ACC_TYPE, ACC_NR, POS, ACC_DAT
TAB_C2          1 Mio.     CLIENT_NR, ACC_TYPE, ACC_NR, POS, ACC_DAT, COL_BP, VALUE_DIM, COL_BT
TAB_C3          350.000     CLIENT_NR, ACC_TYPE, ACC_NR, POS, ACC_DAT, COL_RI, COL_RT, COL_RG, COL_RS, COL_RLIn my opinion i should try to avoid the function calls (nvl(sum...), but i dont't have an idea how to realize this the best way. Maybe you can give me some advice in solve this or point me out some other options to improve performance. The view is used to fill a table.
Thank you in advance
Kind regards
Matthias

Answer 1

I would be really surprised if changing anything in the select list would improve performance much. The Oracle built-in functions are for the most part about as fast as selecting the column without using a function.
I would concentrate my efforts on the from and where parts, which is the part that causes all of the I/O, which is generally the most expensive and time consuming part of the query. Without looking too closely at your query, and without knowing anything about your data, I notice that tab_b and tab_c3 are not used in your select list, and only tab_b has a selective predicate against it. Are you sure you need these two tables?
Can you add more selective predicates against some or all of the tables without changing the results of the query? For example, does the tab_b.amount_type = 0 imply anything about perhaps a date range or account range in the other tables.
You say that
the restrictions:
TAB_C1.CANCEL = 0
TAB_B.AMOUNT_TYPE = 0
have very low selectivity so that indexing one of these columns seems to be not very helpfulbut you also have a predicate on TAB_C1.ACC_TYPE = 'N' AND TAB_C1.CANCEL = 0. Is that combination more selective?
The list goes on, but as others have said, without more information we are only guessing.
John