 |
 |
Archives of the TeradataForum
Message Posted: Wed, 25 Jun 2003 @ 19:34:48 GMT
| Subj: | | Re: Partial & Early Group By |
| |
| From: | | Ballinger, Carrie |
The terms "Partial Group By" and "Early Group By" are not mutually exclusive. Both result in aggregation being performed "earlier" in
the plan. In some cases, only a subset of the aggregation is performed earlier, so in that case it would be "Partial (yet still early)
Group By".
There is a Teradata Orange Book authored by Ahmad Ghazal entitled "Block Optimization in Teradata" which has a section describing Partial
Group By optimization, pages 4-6.
What Ahmad describes as Partial Group By is very much in line with Clay's description below. Partial group by is a new technique in V2R5
for performing an aggregation. In Ahmad's example, if you have two large tables that need to be joined and then aggregated on state
(geography), it is sometimes more efficient to aggregate each table separately on state first, then perform the join. Prior to V2R5 the
optimizer always chose to aggregate after the join. Whether or not partial group by will be used in a plan is a cost-based decision, as it
may sometimes be a more expensive operation. Having statistics collected on the group by columns will enable the optimizer to determine the
cost-effectiveness of partial group by.
Thanks, --Carrie
| |
