Running 9.2.0.2 on AIX 4.3.3. We recently experienced a situation where
a datawarehouse database crashed due to a bug dealing with DBWR_IO_SLAVES > 0.
Oracle's recommended fix was to set the DBWR_IO_SLAVES to 0, which I did.
Now, my understanding is that if AIO is enabled on the box, the rule of thumb
appears to be to use DB_WRITER_PROCESSES and set DBWR_IO_SLAVES = 0.
Now, the Sr DBA here is screaming about the performance since I made the change,
in particular, he says he's seeing high redo latch contention and redo log
buffer waits which he says validates his contention that we need to get back to
using DBWR_IO_SLAVES.
I admit I'm confused. I don't clearly see the relationship that he complaining
about.
My idea of 'monitoring' was to assess free buffer waits, and to monitor the batch
schedules to see if any degradation in timing had occurred, and since the change,
neither has been an issue.
Am I missing something? What stats should I be assessing to contrast the
use of DB_WRITER_PROCESSES vs DBWR_IO_SLAVES? And yes, I've googled and
metalinked and and most information is mainly pertinent to 7.3.4 and 8i and of
a more general nature.
And yes, DISK_ASCYCH_IO is set to TRUE.
Thanks.
--------------------------------------------
Jeffery D Thomas
DBA
Thomson Information Services
Thomson, Inc.
Email: [EMAIL PROTECTED]
Indy DBA Master Documentation available at:
http://gkmqp.tce.com/tis_dba
--------------------------------------------
