Re: OPTIMIZER_INDEX_COST_ADJ, Conceptual Idea

basis_consultant_at_hotmail.com (SAP BASIS Consultant) wrote in news:dd2036f3.0301192023.5e1cad59_at_posting.google.com and I quote:

>
> I am still not sure why an average physical I/O read time for an indexed
> data is different from average physical I/O read time for an FTS.

What is being measured is the *average* time. So, if there are many index entries in a single block, won't that provide a different *average* read time than a few rows/block?

Also, don't forget that an "index entry read" may require reading more than one block. Depends on number of levels in the b-tree and how well the higher levels are cached.

> It is
> not a matter of the data being buffered as we are measuring physical
> I/O.

It very much can be. Completely dependent on how the averages are calculated and what part of the I/O is truly sampled for said average. That's why it is too vague a starting point for setting the parameter, IMO: it depends highly on what is being measured and where. Wide open, since the exact description of how those stats are arrived at by Oracle is not well documented.

-- 
Cheers
Nuno Souto
nsouto_at_optusnet.com.au.nospam