Archives of the TeradataForum

Message Posted: Wed, 26 May 2004 @ 13:37:30 GMT

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

<-- Anonymously Posted: Wednesday, May 26, 2004 09:37 -->

Let me take a stab at this..

In my mind GLOBAL aggregation is akin to joining non-PI columns: the values need to be hashed for redistribution to amps where the same values can be joined together. I think the same principle is at work with global aggregation:

Aggregate by PI column = no data movement and LOCAL aggregation only Aggregate by non-PI => data movement and LOCAL + GLOBAL aggregation

Example: If "Nation" were a non-PI column, then initially rows containing Nation=Sweden would be spread across all amps. The LOCAL aggregation takes care of all of the "Swedens" on each individual amp, but then the GLOBAL aggregation has to bring all of them together in one place, i.e., the amp to which "Sweden" is hashed/redistributed. If nation is PI, then all "Sweden" rows are already on the same amp and aggregation is local. Presumably the PI of the aggregate table would be the same as your GROUP BY columns.

Hope that helps!

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