Archives of the TeradataForum
Message Posted: Thu, 06 Mar 2008 @ 14:08:46 GMT
Yes, I've seen the case where current stats gives a worse access path than no stats. It is a bug. Good, current information should never cause the optimizer to make a less optimal access path choice than with no stats.
Please open an incident. It could be that the extra stats cause different code to referenced that could be improved. It'll never get fixed if it isn't identified.
You do refer to it as a single column stat so it is not the case, but if you have multi-column stats collected and one of the columns has nulls, unfortunately the optimizer will see this as 1 value and potentially produce a less than optimal access path. Example two column multi- stats.....one column has 1 million unique values, 1 million nulls, other column has 1 million values and 1 million nulls as in
Col1 col2 Abcd abcd 1234 null 233434 null Null 1234 Nul 14545 Etc (you get the idea)
The optimizer will think there is only 2 distinct values for the column combination. If there is the potential for many nulls in a column grouping, DO NOT collect multi-column stats. (this may already be corrected in TD 12 - but I don't remember)
|Copyright 2016 - All Rights Reserved|
|Last Modified: 27 Dec 2016|