SQL Optimization - Exit on First Match

Question

Hi,
I have a requirement where a query, sometimes, takes more than 25 seconds. This should come out in less than 1 second.
About the Query :
SELECT 1 FROM DUAL
WHERE     exists
(SELECT TM.AD
FROM     TM,
     GM
WHERE     TM.AD = GM.AD
AND     TM.LOA = :b1
and     GM.soid='Y');
The way this query has been written, it fetches only 1 row. The plan of this query is (not from production but from my test instance as I could reproduce this but the number of rows differ) :
Rows Row Source Operation
1 FILTER (cr=26 pr=0 pw=0 time=433 us)
1 FAST DUAL (cr=0 pr=0 pw=0 time=12 us)
1 NESTED LOOPS (cr=26 pr=0 pw=0 time=398 us)
9 TABLE ACCESS BY INDEX ROWID TM (cr=6 pr=0 pw=0 time=150 us)
9 INDEX RANGE SCAN TM_LOA (cr=2 pr=0 pw=0 time=21 us)(object id 56302)
1 TABLE ACCESS BY INDEX ROWID GM (cr=20 pr=0 pw=0 time=258 us)
9 INDEX UNIQUE SCAN PK_GM (cr=11 pr=0 pw=0 time=123 us)(object id 56304)
The plan of Production is exactly the same. The issue here is :
1. LOA has an Index and for certain values of LOA, the number of records are around 1000. Issue is normally reported when the number of rows fetched are more than 800.
2. The clustering factor of LOA index is not good and from the plan, it is observed that for every row fetched from an Index, approx equal number of blocks are read from table.
3. AD column of GM is a Primary Key
Also, the problem is visible, when the disk reads of this query is very high i.e. if the CR is 800, PR is 700. For any subsequent executions, it gets the results in less than a second.
In my view, it is the table access of TM that is causing an increase in response time and therefore, if I can eliminate these (unwanted) table access. One way is reorganizing the table to improve the CF, but it can have a negative impact. Therefore, optimizing the query seems to be a better option. Based on the Query Plan, I assume, the optimizer gets 1000 rows from an Index and Table TM, then joins to GM. Fetching these 1000 rows seems to be an issue. The query can be optimized, if the search from TM exits immediately a matching row is found in GM table. Therefore, instead of fetching 1000 rows, it matches each and every row and exits immediately when the first match is found. AD in TM is not Unique, therefore, for each AD from TM, it checks for the existence in GM. So, in case there are 10 matching AD from TM and GM, the search should complete immediately on the first matching AD.
Would appreciate help on this.
Regards

Answer 1

Hi,
Will check for the performance with FIRST_ROWS and arrays, but, feel that these will not yield any benefit as 1) The code is directly run on the server, and 2) It is doing a proper index scan, but the code needs a modification to exit immediately as the first match is found.
A pl/sql representation of this code is pasted below :
create or replace function check_exists(la in varchar2)
return number
as
cursor tm_csr is
select ad from tm
where LOA = la;
l_number number:=0;
l_ad tm.ad%type;
begin
open tm_csr;
loop
fetch tm_csr into l_ad;
begin
select 1 into l_number from gm
where gm.ad = l_ad
AND     GM.soid='Y';
exception when others then
l_number:=0;
end;
exit when tm_csr%notfound or l_number=1;
end loop;
close tm_csr;
return l_number;
end;
The code, while not a feasible solution but is just a representation of the requirement, fetches AD from TM. Then it checks for the existence in GM and if a matching row is found, exits from the LOOP.
Edited by: Vivek Sharma on Jul 1, 2009 12:20 PM