Multiple dbwr processes

Question

Hi,
just looking for peoples thoughts on how many dbwr processes a system should have.
We have a database instance doing OLTP 24 * 7, with a single dbwr but linux ASYNC I/O enabled.
The server has 8 logical cpus and the thought is that there should be a dbwr process for each cpu.
My understanding however was that with ASYNC I/O enabled at OS level that the database required just a single DBWR process.
regards
Alan

Answer 1

Pavan Kumar wrote:
Hi,
I read this with respect to ORACLE 8i Verison and check with respect 9i.
Will the result of cpu_count parameter be different from the fixed table you mentioned?
Are you sure on this.. ??
Refer : http://www.pvmehta.com/new/db_writer_process%20or%20dbwr_io_slaves.htm (History ;-))
Let me know ur views..waiting for your comments.
- Pavan Kumar NPavan,
AFAIK, the db_writer_processes will be equal to the number o LRU Latches (db_block_lru_latches), since each latch set will be handled by one db_writer process. But From 9i version the "db_block_lru_latches" is hidden and the db_writer_processes must be less than the number of CPUs .Either you read and understood it wrong or I am missing something? Where it is mentioned that LRU latches are going to control the DBWR process?
Here is the link to 8i,
http://download-west.oracle.com/docs/cd/A87860_01/doc/server.817/a76961/ch131.htm#50255
DB_BLOCK_LRU_LATCHES specifies the maximum number of LRU latch sets. The buffers of a buffer pool are equally divided among the working LRU latch sets of the buffer pool so that each buffer is protected by one LRU latch. Normally, the more latches you specify, the less contention exists for those latches. However, too many latches may result in small LRU lists, potentially reducing the cache life of a database block. The maximum of (CPU_COUNT x 2 x 3) ensures that the number of latches does not exceed twice the product of the number of CPUs and the number of buffer pools.LRU latch control the buffers movement in the LRU list within one workarea. Workareas are 2*cpu_count( I need to check that whether the same is true for 10g or not) so this means that there will be 1:1 relationship between LRU latch and one workarea. Now , here there "may be" a possibility that one DBWR may not be sufficient for the entire cache so there may be a need for multiples of them.But still, that doesn't signfy anywhere that the LRU latches and DBWRs are supposed to be in 1:1 ratio or the former control or determines the other.Itmay be a need that one has to go for mor e number of DBWRs. But that still doesn't mean that there must to be a DBWR for one workarea set in correlation with LRU latch.
I couldn't find any given reference in 9i or beyond as the parameter was deprecated from 9i. Can you point me to the 9i docs where it is mentioned?
I didin't understand the link that youhave given.Its seems to be never updated beyond oracle 8 and a lot has changed since.I don't see that its of much use.
Lastly, I asked about the X$ table that you mentioned for the number of CPUs,
Oracle Database 10g Enterprise Edition Release 10.2.0.1.0 - Production
With the Partitioning, OLAP and Data Mining options
SQL> show parameter count
NAME                                 TYPE        VALUE
active_instance_count                integer
cpu_count                            integer     2
db_file_multiblock_read_count        integer     16
plsql_native_library_subdir_count    integer     0
SQL> select KVIIVAL from X$KVII where KVIITAG='ksbcpu'
  2  /
no rows selected
SQL>  select KVIIVAL from X$KVII where KVIITAG='KSBCPU';
no rows selected
SQL>I have got no result from the query of yours while simply the parameter is giving number of cpus.
Cheers
Aman....