On 23 January 2016 at 22:41, Vik Fearing <vik@2ndquadrant.fr> wrote:
> On 01/23/2016 10:28 AM, John R Pierce wrote:
>> anyways, yeah, BRIN sounds very cool for very large tables with
>> relatively stable data.
>
> I have found it particularly efficient when the BRIN index contains all
> columns of the table. Just using it on one or two columns is not a win
> over btree.
I've found it to be very useful for very large INSERT only tables with
a column which increments with each insert, e.g. a timestamp. This
allows large portions on the table to be skipped during a scan, and
also maintains sequential read speeds which I don't think would work
quite as efficiently with btree index performing heap lookups.
--
David Rowley http://www.2ndQuadrant.com/
PostgreSQL Development, 24x7 Support, Training & Services