The only difference between noworkload stats and workload stats (for mbrc) is that, in noworkload “_db_file_optimizer_read_count” can be altered at session level which will be only useful for testing and won’t work at system level.

So it means the “cost is dictated by a fixed MBRC and therefore does not
change as we modify the multiblock read size” in both.

Could you please crosscheck and correct me if I’m missing something.

Sorry about the delay responding.

When you gather NOWORKLOAD statistics, I believe Oracle uses the _db_file_optimizer_read_count in place of the MBRC to derive a value for mreadtim. But I haven’t done enough experiments with using alter system and alter session to confirm all the details of how settting db_file_multiblock_read_count can affect the value of _db_file_optimizer_read_count, and whether this will then echo on to the calculation of mreadtim.

Could you please explain above differences.

Also while I was finding Adjusted dbf_mbrc for different mbrc it is found that when using “alter system” for mbrc it is not effecting and have to bounce the db in order to use changed value.
Though using “alter session” works as expected.
So that means the value taken by db_file_mbrc at startup is being applied unless db is bounced after change. (10.2.0.4)

So when we collect NOWORKLOAD it will use _db_file_optimizer_read_count for costing and when we collect WORKLOAD stats then it will use db_file_mutliblock_read_count for costing.

So could you please throw some light as what is the difference and how it will calculate the cost in both cases?

You’re absolutely right – I should have checked it sooner.

I’ve just run up a 10.2.0.4, and if I set db_file_multiblock_read_count to zero in the pfile and restart then the optimizer read count follows the db_file_multiblock_read_count, which sets itself the the largest it can. (In fact in my case it went to 79 because of my settings for db_cache_size and sessions, rather than 128).

SQL> alter system set db_file_multiblock_read_count=0 scope=spfile;

System altered.

SQL> shutdown immediate;
Database closed.
Database dismounted.
ORACLE instance shut down.
SQL> conn / as sysdba
Connected to an idle instance.
SQL> startup
ORACLE instance started.

Total System Global Area 1207959552 bytes
Fixed Size 2083528 bytes
Variable Size 771753272 bytes
Database Buffers 419430400 bytes
Redo Buffers 14692352 bytes
Database mounted.
Database opened.

NAME VALUE
—— —–
db_file_multiblock_read_count 128
_db_file_exec_read_count 128
_db_file_optimizer_read_count 128

The important point, though, is that if you set db_file_multiblock_read_count to zero (in 10.2.0.4+) in the spfile and restart the database then the _db_file_optimizer_read_count will default to 8, and not follow. i.e. “alter system” and “bounce database” have different effects.

1)On first db block_size=16k after setting dbmbrc to 0 following is the result

NAME VALUE
—— —–
db_file_multiblock_read_count 64
_db_file_exec_read_count 64
_db_file_optimizer_read_count 64

2) On second db block_size=8k after setting dbmbrc to 0 following is the result

NAME VALUE
—— —–
db_file_multiblock_read_count 128
_db_file_exec_read_count 128
_db_file_optimizer_read_count 128

Then on 2nd db
SQL> alter system set db_file_multiblock_read_count=64 scope=memory;

NAME VALUE
—— —–
db_file_multiblock_read_count 64
_db_file_exec_read_count 64
_db_file_optimizer_read_count 64

So it seems it works on 10.2.0.4 without bouncing the db.