Re: Significance of specific database parameter difference

Well, I mostly agree with you, especially with optimizer_index_cost_adj. The value derived from the default system statistics is around 10, meaning that db file single block read is about 10% of the cost of the db file scattered read (multi-block read). However, Ion the systems with a lot of memory, I still like to set optimizer_index_caching to 80, meaning that 80% of the index blocks are cached, because that usually reflects the situation.

The problem with optimizer_index_cost_adj is the same as with the rule based optimizer: it sometimes selects the wrong index. That parameter does a very good approximation of RBO and makes CBO emulate the RBO very, very well. However, with the data models not observing the 3rd normal form and being generally crappy, RBO can be a shot in a foot. Or even both feet.  With the reporting and DW databases, I find OPTIMIZER_DYNAMIC_SAMPLING=11 being very helpful. No need to mess with the optimizer parameters.

As the idiot who wrote a white paper 20 years ago to make people aware of the optimizer_index_caching and optimizer_index_cost_adj parameters introduced initially in Oracle8 and documented in Oracle8i, I always wince when I see people crippling themselves with those parameters today.  Please, for heaven's sake, stop!  Using an instance-level parameter to issue a broad directive to every SQL statement about a condition which cannot be true in all cases is inappropriate micromanagement.  Let the CBO do what it is designed to do, which is to use specific data conditions portrayed by system-, table-, index-, and column-statistics to make decisions on a case-by-case basis for each individual SQL statement.

Also, big kudos on getting rid of the "underscore" parameters!  Using undocumented "underscore" parameters not justified by directives from Oracle Support cases or documented configuration requirements from a vendor (i.e. SAP for R3, Oracle for EBS, etc) is a stunningly bad idea.  Undocumented parameters must be justified, else removed, especially when upgrading.

Everything else looks sensible to me.

On 4/20/2021 4:05 AM, Pap wrote:

Hello All, we are moving from HP to Exadata - X8. Along with this we are also moving from current Oracle version 11.2.0.4 to 19C. And below are a few of the differences in parameters we see between old and new databases. New database(which is not yet live) holds mostly default setup only. But the existing database has below non default setup(Around ~12 parameters), And I don't know if those non default setups in the old/existing database is really helping us anyway. So wanted to understand if it's safe to go by all the defaults(as it's mentioned below) for the new database? Or any of these non default parameters from existing databases is really critical to be double checked/verified thoroughly, so that we may need to move those as it is to the new database?

Existing environment (11.2, HP):-

_buffered_publisher_flow_control_threshold = 500000

_capture_publisher_flow_control_threshold =500000

_enable_query_rewrite_on_remote_objs = true

_optimizer_cost_based_transformation = linear

nls_sort = 'BINARY'

optimizer_dynamic_sampling=0

optimizer_index_caching=100 

optimizer_index_cost_adj=1

optimizer_secure_view_merging=FALSE

query_rewrite_enabled=FALSE

optimizer_mode=FIRST_ROWS

session_cached_cursors=0

New environment- (19C, Exadata - X8):-

_buffered_publisher_flow_control_threshold = null

_capture_publisher_flow_control_threshold =null

_enable_query_rewrite_on_remote_objs = NULL

_optimizer_cost_based_transformation = null

nls_sort =NULL

optimizer_dynamic_sampling=2

optimizer_index_caching=0

optimizer_index_cost_adj=100

optimizer_secure_view_merging=TRUE

query_rewrite_enabled=TRUE

optimizer_mode=ALL_ROWS

session_cached_cursors=200

-- 
Mladen Gogala
Database Consultant
Tel: (347) 321-1217
https://dbwhisperer.wordpress.com