| From: | Josh Berkus <josh(at)agliodbs(dot)com> |
|---|---|
| To: | Mark Kirkwood <mark(dot)kirkwood(at)catalyst(dot)net(dot)nz>, Heikki Linnakangas <hlinnakangas(at)vmware(dot)com>, Simon Riggs <simon(at)2ndQuadrant(dot)com> |
| Cc: | PostgreSQL-development <pgsql-hackers(at)postgresql(dot)org> |
| Subject: | Re: ANALYZE sampling is too good |
| Date: | 2013-12-10 21:45:29 |
| Message-ID: | 52A78B79.6060904@agliodbs.com |
| Views: | Raw Message | Whole Thread | Download mbox | Resend email |
| Thread: | |
| Lists: | pgsql-hackers |
On 12/10/2013 01:33 PM, Mark Kirkwood wrote:
> Yeah - and we seem to be back to Josh's point about needing 'some math'
> to cope with the rows within a block not being a purely random selection.
Well, sometimes they are effectively random. But sometimes they are
not. The Chaudri et al paper had a formula for estimating randomness
based on the grouping of rows in each block, assuming that the sampled
blocks were widely spaced (if they aren't there's not much you can do).
This is where you get up to needing a 5% sample; you need to take
enough blocks that you're confident that the blocks you sampled are
representative of the population.
--
Josh Berkus
PostgreSQL Experts Inc.
http://pgexperts.com
| From | Date | Subject | |
|---|---|---|---|
| Next Message | Greg Stark | 2013-12-10 21:48:10 | Re: ANALYZE sampling is too good |
| Previous Message | Andres Freund | 2013-12-10 21:41:45 | Re: pg_stat_statements fingerprinting logic and ArrayExpr |