RE: weird behaviour for mreadtim < sreadtim in 10g

Amit

> My question is what calculation 10g is using for calculating FTS
> cost when mreadtim < sreadtim.

When mreadtim <= sreadtim the query optimizer fall back to the computation used with no workload system stats. Once I tried to find out how the computation is done... The best I found was:

I/O cost = blocks / ( 1 + A*ln(MBRC) + B*ln(MBRC)^2 + ... )

Notes:

- A and B depends on IOSEEKTIM and IOTFRSPEED
- MBRC is used, not DB_FILE_MULTIBLOCK_READ_COUNT
- In any case with system stats Wolfgang's formula is no more valid
- In 9i neither IOSEEKTIM nor IOTFRSPEED exist...

HTH
Chris

--
http://www.freelists.org/webpage/oracle-l