Query performance - dynamic indexing?

Question

Hi Experts,
I have a query like this:
WITH /*+materialize*/ A AS (SELECT  A.NAME,
   B.pname,
   B.ptype,
DECODE(C.ttype,'L','Lab','Non-Lab') ttype
,(CASE WHEN(C.ttype = 'L') THEN C.RES_CODE ELSE NULL END) RES_CODE
,(SELECT (first_name|| ' ' ||last_name) FROM resources WHERE name=C.ROLE)ROLE_desc
,D.currency Bill_curr
,D.TOTALCOST Bill_cost
,E.currency Home_curr
,E.TOTALCOST Home_cost
,C.role role_code
,(C.HOURS) AS HRS
,(D.TOTALAMOUNT) 
,(E.TOTALAMOUNT)    Home_amount
FROM    WORKPROG C
,WORKPROG_VAL D
,WORKPROG_VAL E
,PROJECTS A
,periods B
WHERE  C.TNO=D.TNO
AND    C.TNO=E.TNO
AND    D.CURR_TYP='BILL'
AND    E.CURR_TYP='HM'
AND    A.NAME=C.PROJ_CODE
AND    C.PROJ_CODE LIKE 'JA%'
AND C.TDATE BETWEEN b.start_date AND b.end_date AND b.ptype ='SMON'
SELECT
name,b.pname,b.ptype,ttype,res_code,role_desc,role_code,Bill_curr,SUM(Bill_cost),Home_curr,SUM(Home_cost),SUM(hrs),SUM(totalamount),SUM(Home_amount)
FROM A
GROUP BY name,b.pname,b.ptype,ttype,res_code,role_desc,role_code,Bill_curr,Home_currThe problem is - the WITH part takes no time to run, its very fast. But the group by part takes ages. This is because the WITH part returns millions of records.
Is ther eany way that I can dynamically index the result set of WITH query so that the group by runs faster.
Thanks.

Answer 1

Caitanya wrote:
Thanks Someone.
By ages... I mean it kept running for 45 minutes after which I aborted it.
The indexes are being used and there is no 'FULL TABLE ACCESS' in the explain plan. We are using Oracle 9i.It is quite unlikely (although possible) that getting millions of rows is done effectively by using indexes, you should actually look for performing full table scans. Only if the tables contained billions of rows it would be reasonable for the optimizer to choose an index access, but it will take hours to get millions of rows via index lookup and the corresponding single row random access to the tables.
I assume that the statistics are not up to date because a query that returns millions of rows probably won't have a realistic cost of 102. That would mean that Oracle estimates that it takes him the time of 102 single block reads to complete the query - if you take a conservative guess of 10ms per random single block access then this means roughly 1 second.
So please post your explain plan here, and if possible can you get the total number of rows in the tables accessed by the query and the approx. number of rows the query is supposed to return as base information.
Here are some (generic) instructions how to generate a reasonable EXPLAIN PLAN output:
Could you please post an properly formatted explain plan output using DBMS_XPLAN.DISPLAY including the "Predicate Information" section below the plan to provide more details regarding your statement. Please use the \[code\] tag before and \[code\] after or {noformat}{{noformat}code{noformat}}{noformat} before and after to enhance readability of the output provided:
In SQL*Plus:
SET LINESIZE 130
EXPLAIN PLAN FOR <your statement>;
SELECT * FROM TABLE(DBMS_XPLAN.DISPLAY);Note that the package DBMS_XPLAN.DISPLAY is only available from 9i on.
In previous versions you could run the following in SQL*Plus (on the server) instead:
@?/rdbms/admin/utlxplsA different approach in SQL*Plus:
SET AUTOTRACE ON EXPLAIN
<run your statement>;will also show the execution plan.
In order to get a better understanding where your statement spends the time you might want to turn on SQL trace as described here:
[When your query takes too long|http://forums.oracle.com/forums/thread.jspa?threadID=501834]
and post the "tkprof" output here, too.
Regards,
Randolf
Oracle related stuff blog:
http://oracle-randolf.blogspot.com/
SQLTools++ for Oracle (Open source Oracle GUI for Windows):
http://www.sqltools-plusplus.org:7676/
http://sourceforge.net/projects/sqlt-pp/