DB_WRITER_PROCESSES or DBWR_IO_SLAVES?

Hi folks,

We have a very write intensive application (a lot of insert operations) and the most popular wait event we encountered is "free buffer wait".
The parameter db_writer_processes was initially set to 4 (against 4 CPUs until yesterday).
Actually,after some observations we made,we didn't see a difference between multiple DBWRs and one DBWR regarding the wait event "free buffer wait" and we still get a huge amount of
this wait.
Now the parameter db_writer_process = 1.

What do you recommend from your experience : multiple DBWRs or working with io slaves (dbwr_io_slaves)?

Following some details:
Server - Linux 2.6.9-42.ELsmp i386 athlon
CPUs - 8 (was increased yesterday from 4)
Physical Memory (RAM) - 32GB
Database version - 10.2.0.3.0

Following the init*.ora parameters:

Code:

---

bilminer.__db_cache_size=956301312
bilminer.__java_pool_size=16777216
bilminer.__large_pool_size=16777216
bilminer.__shared_pool_size=251658240
bilminer.__streams_pool_size=0
*.audit_file_dest='/b4bdb/app01/oracle/admin/bilminer/adump'
*.background_dump_dest='/b4bdb/app01/oracle/admin/bilminer/bdump'
*.compatible='10.2.0.3.0'
*.control_files='/b4bdb/or01/bilminer/control01.ctl','/b4bdb/or02/bilminer/control02.ctl'
*.core_dump_dest='/b4bdb/app01/oracle/admin/bilminer/cdump'
*.db_block_checksum='FALSE'
*.db_block_size=16384
*.db_domain=''
*.db_file_multiblock_read_count=64
*.db_name='bilminer'
*.db_writer_processes=1 # changed by roee 26/5/2007 was 4
*.disk_asynch_io=FALSE
*.dispatchers='(PROTOCOL=TCP) (SERVICE=bilminerXDB)'
*.fast_start_parallel_rollback='FALSE'
*.job_queue_processes=10
*.log_archive_dest='/b4bdb/oa01/bilminer'
*.log_archive_format='%t_%r_%s.arc'
*.log_archive_max_processes=1
*.open_cursors=500
*.optimizer_index_caching=90
*.optimizer_index_cost_adj=10
*.pga_aggregate_target=827326464
*.processes=2000
*.remote_login_passwordfile='EXCLUSIVE'
*.remote_os_authent=true
*.sga_max_size=1258291200
*.sga_target=1258291200
*.undo_management='AUTO'
*.undo_retention=21600
*.undo_tablespace='UNDOTBS1'
*.user_dump_dest='/b4bdb/app01/oracle/admin/bilminer/udump'
*._optim_peek_user_binds=FALSE
*.log_checkpoint_interval=999999999
*.log_checkpoint_timeout=0

---

Thanks in advance,
Nir

Hi

If you are using Linux you can enable Asynch IO or Direct IO to achieve better write performance

Also, isnt your db_file_multiblock_read_count too high?

Hi pando,

Thanks a lot for your advice!

Regarding to DB_FILE_MULTIBLOCK_READ_COUNT - You are right.
I'll check it tomorrow. This is an OLTP environment and not a DSS environment.
Thanks for this remark also.

Best regards,
Nir