On Fri, 2007-10-26 at 16:37 -0400, Pablo Alcaraz wrote:
> I executed 2 equivalents queries. The first one uses a union structure.
> The second uses a partitioned table. The tables are the same with 30
> millions of rows each one and the returned rows are the same.
>
> But the union query perform faster than the partitioned query.
>
> My question is: why? :)
The two queries are equivalent but they have different execution plans.
The UNION query has explicit GROUP BY operations within it. We do not
currently perform a push-down operation onto the individual partitions.
This results in more data copying as well as requiring a single very
large sort, rather than lots of small ones. That is probably enough to
allow it to perform the sort in memory rather than on-disk, thus
allowing a considerable speed-up.
This is on my list of requirements for further partitioning improvements
in 8.4 or beyond.
--
Simon Riggs
2ndQuadrant http://www.2ndQuadrant.com