Archives of the TeradataForum

Message Posted: Fri, 06 Sep 2002 @ 14:42:25 GMT

	<Prev	Next>		<<First	<Prev	Next>	Last>>

The TD optimizer is not 'skew aware', explains don't indicate skew. So the explain always shows all-amps for a full table scan, even though in reality only 1 amp is doing all the work. Skew is often a performance problem and it sure would be nice if explains indicated skew. Skew is bad (period - in any PI or join condition). If you have 128 availabe amps and all your data is on just one , then your query is 128X slower (and required spool 'limit' is 128X greater). Often joins on other (e.g. none PI) skewed columns are unaviodable - it would be nice if TD provided some nifty 'internal equi-redistribution' of skewed data for these skewed joins (in worst case scenarios you may find yourself redistributing skewed data via concatentation on random(1, number_amps) or somesuch - or a model change).

FYI: Doing a select from dbc.tablesize for the table will confirm all the data (currentperm) is on a single amp (skewed).

	<Prev	Next>		<<First	<Prev	Next>	Last>>