Archives of the TeradataForum

Message Posted: Mon, 28 Feb 2005 @ 20:35:45 GMT

	<Prev	Next>		<<First	<Prev	Next>	Last>>

Hello,

as most of you probably know one can't collect statistics on volatile tables. I have recently seen a situation (V2R5) where the optimizer obviously believes that a volatile table is empty. Therefore it chooses to product join the VT right in the beginning (after having estimated that the duplicated to all amps spool # records = # amps).

I have seen this happening with between 0.1 and 8 % of the AMPs having records (did not test with > 8 %). Changing the VT to real tables with collected statistics changed the explains to be as expected.

Now, I'm wondering: Is there a rule of thumb indicating when a table is too small or too skewed to use volatile tables (as the random amp sample is too unrepresentative)? Or is there anything else I can do to improve the plans when using small VTs? Is there maybe a 'secret way' or trick how I could get the statistics?

Thanks for your help!

Mirjam Berger

	<Prev	Next>		<<First	<Prev	Next>	Last>>