Archives of the TeradataForum
Message Posted: Tue, 21 Jun 2011 @ 11:34:24 GMT
Subj: | | Re: Distinct vs. group by in TD 13 onwards |
|
From: | | Walter, Todd |
100% is the intent. By the time it gets to plan generation there is only one representation internally so the decisions are all made from that
one form. Then Opt decides which algorithm to use based on what it knows about demographics, especially information about how unique the source
relation already is.
Side effect is that while you could force one algorithm or another before, you no longer can. So it becomes much more important to have Stats
on the columns that go into the list that is being grouped/distinct-ed.
Also, there is a run-time component that detects cases where the estimates did not accurately portray the demographics and adjust the algorithm
on the fly - eg turn off local aggregate cache if it is not doing any good.
|