Comment 3 for bug 686353

Revision history for this message
Arjen Lentz (arjen-lentz) wrote : Re: [Bug 686353] [NEW] 10x performance regression with correlated indexes and sort_intersection in maria-5.1-wl21

Hi Philip, Igor, all

On 07/12/2010, at 5:45 PM, Philip Stoev wrote:
> The following query against the DBT-3 dataset
>
> SELECT count(*) FROM lineitem WHERE ( l_receiptDATE BETWEEN
> '1992-11-01'
> AND '1992-11-29' ) AND l_shipdate < '1992-11-15';
>
> is 10x times slower when executed with sort_intersection.
>
> Igor says that this is " not interesting because lineitem is naturally
> ordered by l_receiptDATE and l_receiptDATE is strongly correlated
> with
> l_shipdate" , " in the case when indexes are correlated intersection
> causes degradation: this is a well know fact.".

I don't care very much for the specific scenario, I think the main
issue is performance regression.
New features or changes should not make existing/old features slower -
that has been one of the fundamental rules that Monty has worked with
since the start as a guideline for development.

Regards,
Arjen.
--
Arjen Lentz, Exec.Director @ Open Query (http://openquery.com)
Remote expertise & maintenance for MySQL/MariaDB server environments.

Follow us at http://openquery.com/blog/ & http://twitter.com/openquery