Pga_aggregate_target and hash_join cost (anomalies)

Question

Hi,
I've got some performance issues with queries using bind variables (BVP + histograms = evil).
I did some test cases in test environment where the only difference with PROD was pga_aggregate_target .
And was able to make plan changes w/o changing query .
Whole test was:
alter session set workarea_size_policy=manual;
alter session set hash_area_size=11131000;
and with that settings optimizer choose:
TABLE ACCESS FULL with HJ
over
nested loop with index Range Scan.
My question is can query plan depend on current hash_area_size value derivered from PAT parameter (is that dynamic during session life) ?
I can imagine that session can have different hash_area_size in different point in time when workarea_size_policy=auto,
because the amount of allocated memory depends on whole system load.
But this leads to anonalies in plans is that possible ?
During my test, query was reparsed because of hash_area_size change, nothing more.
DB 9.2.0.8 EE
Regards.
GG
Edited by: user10388717 on May 29, 2009 6:25 AM

Answer 1

But this leads to anonalies in plans is that possible ?Yes, it may affect plans in runtime. There are different causes to not shared cursors due to some changes besides query text change. It may be checked through [V$SQL_SHARED_CURSOR|http://download.oracle.com/docs/cd/B10501_01/server.920/a96536/ch3199.htm#1125735] view.