On Mon, 17 Nov 2008, Scott Marlowe wrote:
> Just wondering if you compared it to how fast it runs if you've got
> lots of shared_buffers and everything fits into memory. That would be
> an interesting comparison.
With a large increase in work_mem as well to speed up sorting. If the
bottleneck is write speed on the intermediate results mentioned, then it
may very well be the case that the best way to accelerate this workload is
with a RAM-based tablespace. But in some cases tuning shared_buffers and
work_mem way upwards is all it takes.
--
* Greg Smith gsmith@gregsmith.com http://www.gregsmith.com Baltimore, MD