Ora-4031 after reducing the max sga

Question

Oracle 10g R2 and Baan 5c on AIX 5.3 L (p550, 3792M , 4 lcpu)
After I reduced the max_sga_size (so to avoid the paging), now I see more ora-4031 (out of shared memory loading library cache object) warnings in the udump trc file, not in alert file. In addition, they appeared in the log.ora.sql of Baan as an error message, so the query requests were hung or stopped.
Reading from other posts and metalink note: 146599.1, it indicated that the system has extensive fragmentation problem. (REQUEST_FAILURES is > 0 and LAST_FAILURE_SIZE is > SHARED_POOL_RESERVED_MIN_ALLOC. )
So I used the tips form AskTom to find the query scripts that may have bind variable problem and I did found some, most of them in 2~4 multiples. However, not all of them gave ora-4031 warning/error, actually only very few.
So what should I do for the best practice of the performance? Correct the problems with the bind variable of all queries or
increase the max SGA again.
All the queries were produced by the Baan´s processes or modules.

Answer 1

I got hit this thread while searching some other subjects.
(This is kind of old... but answered for a long time)
I think this is somewhat interesting problem.
Here is one query showing in the query of mulitple copies in system by using asktom technique. The system return 3 copies.
SELECT /*+ FIRST_ROWS INDEX(A TTSSOC@$IDX@) */ A.T$CLST FROM >BAANDB.TTSSOC@ A WHERE (A.T$CLST = :@ OR A.T$CLST = :@) AND A.T$ORNO = :@I think there is a chance that you're hitting bind mismatch problem.
Is [email protected]$CLST(What an annoying names~) column is VARCHAR2 type? If yes, following thread will be helpful.
Re: Is there any way to avoid hard parsing caused by "BIND_MISMATCH" in 10g