Specifying Nonstandard Block Sizes for Tablespaces [message #488284] |
Thu, 06 January 2011 03:00 |
dba_7722
Messages: 197 Registered: August 2010 Location: Delhi
|
Senior Member |
|
|
Dear Experts,
We are creating non standard blocking size for 16k and so accordingly have to set the parameter db_16k_cache_size.
Please suggest if have any thumb rule for setting the value for this parameter.
Regards,
Jay vardhan
|
|
|
|
Re: Specifying Nonstandard Block Sizes for Tablespaces [message #488288 is a reply to message #488284] |
Thu, 06 January 2011 03:24 |
John Watson
Messages: 8960 Registered: January 2010 Location: Global Village
|
Senior Member |
|
|
Hi - the rule of thumb is "don't set it".
When this first came out with 9.0 beta, like many DBAs I was going to use it for performance: have my IOTs in a 2k tablespace, with the overflow segment in a 16k tablespace; a table in 4k, with its LOBs in 32k; and so on. But as time went by, the advice from product development was absolutely clear, and when you research this you'll find comments to the effect that "this facility is provided for the sole purpose of transporting tablespaces between databases of different db_block_size. Do not use it for performance tuning".
I'm told that the problem is that the buffer cache block replacement algorithm and latches are not optimized for the non-default buffer pools.
If you must set it, follow the figures in v$buffer_pool_advice.
|
|
|
|
|
|