Re: dbms_stats.auto_sample_size or sample size 20

  • From: Connor McDonald <mcdonald.connor@xxxxxxxxx>
  • To: oracle-l@xxxxxxxxxxxxx
  • Date: Wed, 1 Feb 2006 09:19:36 +0800

On 2/1/06, Juan Carlos Reyes Pacheco <juancarlosreyesp@xxxxxxxxx> wrote:
>
> Thanks Jurijs,
> In my situation when gathering histograms had reduced the process to
> about 30%, we are in test process so we will see if we lost accurracy,
> thank you for your comment.
> --
> //www.freelists.org/webpage/oracle-l
>
>
>
its a trade off between row and block sampling.

if you run 1% sampling, but at row level (the default) then you want to
sample 1% of the rows.  If you got 100 rows per block, then you may well
visit every single block in the table...

If you're going to use sampling (for efficiency reasons), then block
sampling will assist (at the cost of statistical accuracy)

cheers
connor


--
Connor McDonald
===========================
email: connor_mcdonald@xxxxxxxxx
web:   http://www.oracledba.co.uk

"Semper in excremento, sole profundum qui variat"

Other related posts: