Query Performance Please Help

Question

Hi can any body tell me how do I improve the performance of this query.This query takes forever to execute.
PLEASE HELP
select substr(d.name,1,14) "dist",
sum(r.room_net_sq_foot) "nsf",
sum(r.student_station_count) "sta",
sum(distinct(r.cofte)) "fte"
from b_fish_report r,
g_efis_organization d
where substr(r.organization_code,-2,2) = substr(d.code,-2,2) and
d.organization_type = 'CNTY' and
r.room_satisfactory_flag = 'Y' and
substr(d.code,-2,2) between '01' and '72'
-- rownum < 50
group by d.name, r.organization_code
order by d.name
It has nonunique Indexes on Organization code
Thanks
Asma.

Answer 1

Asma,
I tried your SQL on my tables T1 and T2. Indexes are on C1,C2,C3 and N1,N2,N3. The data in T1 and T2 are shown below with the explain plan (also called EP) listed. You really need to do an explain plan (free TOAD is easiest to do this in) and respond showing your EP results.
By simply changing the optimizer mode to RULE I was able to get it to use indexes on both T1 and T2.
T1 data
C1     C2     C3     N1     N2
001     Y     AAA     1     11
002     Y     BBB     2     22
003     Y     CCC     3     33
111     N     DDD     4     44
222     N     EEE     5     55
333     Y     FFF     6     66
070     Y     GGG     7     77
071     N     HHH     8     88
072     Y     III     9     99
TEST     TEST     TEST     10     100
T2 data
C1     C2     C3     N1     N2
001     CNTY     AAA     1     11
002     CNTY     BBB     2     22
003     CNTY     CCC     3     33
111     XXX     DDD     4     44
222     XXX     EEE     5     55
333     CNTY     FFF     6     66
070     CNTY     GGG     7     77
071     XXX     HHH     8     88
072     CNTY     III     9     99
TEST     TEST     TEST     10     100
This is the results when I run the SQL based on this data ...
dist     nsf     sta     fte
AAA     1     11     10
BBB     2     22     20
CCC     3     33     30
FFF     6     66     60
GGG     7     77     70
III     9     99     90
--[SQL 1] : with CHOOSE as the optimizer mode, which is normally the DEFAULT if no hint is specified
select /*+ CHOOSE */
substr(d.c3,1,14) "dist",
sum(r.n1) "nsf",
sum(r.n2) "sta",
sum(distinct(r.n3)) "fte"
from t1 r, t2 d
where substr(r.c1,-2,2) = substr(d.c1,-2,2) and
d.c2 = 'CNTY' and
r.c2 = 'Y' and
substr(d.c1,-2,2) between '01' and '72'
group by d.c3, r.c1
order by d.c3
This is what the EP shows for your SQL (which will probably be the same for you once you do an EP on your actuall sql) ...
SELECT STATEMENT Optimizer=CHOOSE (Cost=4 Card=1 Bytes=37)
SORT (GROUP BY) (Cost=4 Card=1 Bytes=37)
NESTED LOOPS (Cost=2 Card=1 Bytes=37)
TABLE ACCESS (FULL) OF T1 (Cost=1 Card=1 Bytes=12)
TABLE ACCESS (BY INDEX ROWID) OF T2 (Cost=1 Card=1 Bytes=25)
INDEX (RANGE SCAN) OF I_NU_T2_C2 (NON-UNIQUE)
Notice the FULL table scan of T1 which you don't want, and neither C1 index is getting used (I've explained why below).
--[SQL 2] : only changed the hint to RULE ...
select /*+ RULE */
substr(d.c3,1,14) "dist",
sum(r.n1) "nsf",
sum(r.n2) "sta",
sum(distinct(r.n3)) "fte"
from t1 r, t2 d
where substr(r.c1,-2,2) = substr(d.c1,-2,2) and
d.c2 = 'CNTY' and
r.c2 = 'Y' and
substr(d.c1,-2,2) between '01' and '72'
group by d.c3, r.c1
order by d.c3
SELECT STATEMENT Optimizer=HINT: RULE
SORT (GROUP BY)
NESTED LOOPS
TABLE ACCESS (BY INDEX ROWID) OF T2
INDEX (RANGE SCAN) OF I_NU_T2_C2 (NON-UNIQUE)
TABLE ACCESS (BY INDEX ROWID) OF T1
INDEX (RANGE SCAN) OF I_NU_T1_C2 (NON-UNIQUE)
Though the C2 index is getting used (your r.c2 = 'Y' part in the where clause) the main problem your having here is the JOIN column (C1 in both tables) is not getting used. So the join you have ...
where substr(r.c1,-2,2) = substr(d.c1,-2,2)
isn't using an index and you want it too. There are 2 solutions to correct this..
Solution #1
The first is to make a function-based index for data. Since your doing SUBSTR on C1 that C1 index does not contain that partial information so it will not use it. Below is the syntax to make a function based index for this partial data ...
CREATE INDEX I_NU_T1_C1_SUBSTR ON T1 (SUBSTR(C1,-2,2));
CREATE INDEX I_NU_T2_C1_SUBSTR ON T2 (SUBSTR(C1,-2,2));
or also this way if it's still not using the above indexes ...
CREATE INDEX I_NU_T1_C1_SUBSTR ON T1 (SUBSTR(C1,-2,2),C1);
CREATE INDEX I_NU_T2_C1_SUBSTR ON T2 (SUBSTR(C1,-2,2),C1);
Solution #2
The second solution is to make another column in both table and place this 2 digit information in it, and then index this new column. That way the join will look like ...
where r.c_new_column = d.c_new_column
and
r.c_new_column between '01' and '72'
also with this new column the BETWEEN clause at the end you will not need the substring as well. Also remember BETWEEN on character values is different than numbers.
Final Notes
I just tried creating the functional index and I can't get it to be used it for some reason (I might not have the right amount of data), but I really think that is your best option here. As long as it uses the functional index you won't have to change your code. You might want to try using INDEX() in the hint to get it to be used, but hopefully it will use it right away. Try all 4 types of optimizer modes (CHOOSE, RULE, ALL_ROWS, FIRST_ROWS) in your primary hints to see if it will use the new function-based index.
You really do need to get explain plan going. Even if you make these functional indexes you won't know if its going to be using them until you look at the EP results. You can do EP manually (the SQL of how to produce the results is in OTN, though I find free TOAD is by far the easiest) and you will still need to have run the utlxplan.sql script. Oracle I do think has some GUI tools, maybe in OEM, that have explain plan built in as well.
I hope this helps ya,
Tyler D.