 |
 |
Archives of the TeradataForum
Message Posted: Fri, 05 Nov 2010 @ 14:14:18 GMT
| Subj: | | Re: Multi-column statistics |
| |
| From: | | John_Wight |
Ruth Wrote:
| | You need another Teradata Expert. | |
| | Multi-column stats are not always necessary, but they often help. A good example would be two columns that are not hugely unique
individually, but together, their combination are more unique and you are constraining on both those columns. The optimizer may choose a
different access path. If both columns are very unique on their own, you may not benefit from multi-column stats. | |
Agree with Ruth's assessment. I've found that Multi-column stats are the 'exception' to the rule but as she intimates, when you discover those
conditions and queries are being qualified with those columns they do and will help! However, before you commit any stats to go forward to
production, you need to create them, test and if they improve your plan - keep them, else throw them away. Collecting stats is an intensive
process, specially when there are lots of 'unnecessary' stats.
Hope this helps;-)
JK
| |
