Re: How to estimate TEMP usage for DBMS_STATS?
Date: Wed, 17 Feb 2010 09:05:42 -0600 (CST)
> Temp is used for the NDV calculation because it does a sort distinct
> so the larger the number of distinct values the more space it needs.
> This changes in 11.1 with the new "fast NDV" that was introduced for
> dbms_stats.auto_sample_size (no temp is used - a hash based algo makes
> this possible). So worst case scenario I'd say, temp would need to be
> as large as your largest segment.
Great stuff! But I'm not sure about TEMP needing to be as large as the largest segment. I'm off to test that theory as soon as I clean up another mess...
There's a proposal for us to get to 11g this year, but I'm not optimistic.