Archives of the TeradataForum

Message Posted: Tue, 25 May 2004 @ 14:29:21 GMT

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

Hello Ole:

This is a very informative forum. There are many people out here with some valuable experience and insight. More importantly, they are will in share it.

The way that global aggregation was described to me in an NCR class is via the use of an acronym: ARSA

A- aggregate locally to eliminate dups on each AMP

R- redistribute these values from all AMPs (using aggregated column(s) value like a PI) to get all the same values on a single AMP

S- Sort

A- aggregate again on each AMP to eliminate all duplicates

Although this is a good memory tool, I have often questioned the actual use of a sort if it is truly doing the bucketing of unique values. Plus, the results of a GROUP BY do not come back sorted.

Your observation is therefore correct, the only way to get local aggregation is via the use of the PI.

Hope this helps,

Michael Larkins
Certified Teradata Master

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