Re: OPTIMIZER-Problem

Ralf Korell <ralf_at_hydrogen.emi.de> wrote:
>This resulted into the first problem, because the init.ora parameter
>OPTIMIZER_MODE won't run with 'COST' (ORA Server Concepts ,
>13-29), but it runs with 'FIRST_ROWS'.
>
>Q 1a : is this the same ?

The COST keyword has been changed to CHOOSE.

FIRST_ROWS tries to give you some data as quickly as possible. ALL_ROWS tries to give you ALL data as quickly as possible. I dont know for sure but I expect that CHOOSE would be the same as (or very close to) ALL_ROWS.

Anybody know how CHOOSE and ALL_ROWS compare?

>Then the surprise : as we put the tuned application on our
>customers's machine (an HP 9000/887, HP-UX 9.05 ) and force the
>optimizer mode to FIRST_ROWS, the application SLOWS DOWN
>(analyze IS performed) !
>
>Q 1b: has anybody an idea about this ?

Is your development system using a snapshot of production data or just a sampling? Try increasing your sampling rate I have had problems in the past with development data not containing the real (Wierd) data distributions that existed in production. In order to get around this it was necessary to anlayze one table in particular with a sample rate of 35% and all others with 5%

-- 
Regards, 

Lee E. Parsons                  		lparsons_at_world.std.com