|
Archives of the TeradataForumMessage Posted: Tue, 06 Nov 2012 @ 11:14:14 GMT
Hi Sravan, I'm afraid you will have to test case by case to verify if dropping the statistics change the plan or not. Generally, it is recommended to collect stats on predicates, joining columns & indexes, more details in the following, very useful article: Having too much stats collected has no other impact apart of higher CPU consumption during recollection. But that impact can be significant. In rare cases, you can confuse the optimizer with some multicolumn stats and, of course, stale stats. Also be aware, that prior TD14, there was a 16-byte limit on the value size of the histogram. So if you collect on large columns or multicolumns, only first 16 bytes are considered. This can be tricky, especially when those first 16 bytes are not distinct enough. Cheers, Tomas
| ||||||||||||||||||||||||||||||||||||||||||||||||
| ||||||||||||||||||||||||||||||||||||||||||||||||
Copyright 2016 - All Rights Reserved | ||||||||||||||||||||||||||||||||||||||||||||||||
Last Modified: 15 Jun 2023 | ||||||||||||||||||||||||||||||||||||||||||||||||