Archives of the TeradataForum
Message Posted: Mon, 19 Jan 2004 @ 17:16:00 GMT
These statistics will serve different queries. Theoretically, you could take the largest case (6 columns?) and calculate 6-dim statistics and all possible projections (from 1- to 5-dim), if users often define 1 to 5 WHERE clauses (or columns in joins) in their ad hoc analysis ot batch jobs etc. It will be quite a task to collect them all for a large table:
1-dim: 6 statistics;
You can collect as many as you can afford time-wise (up to 63 statistics listed above); all of them might prove useful in busy (OLAP, data mining ...) applications. If you don't have time for all of them (most people won't) - just collect the multi-column stats for as many columns as the largest number of where clauses (or columns in joins) in the query you care about. You could also add a few most frequently used single-column stats (as in R4). Perhaps you will find your optimum somewhere between these two extreme cases.
|Copyright 2016 - All Rights Reserved|
|Last Modified: 15 Jun 2023|