Re: Slow count(*) again...

Kevin Grittner <kevin.grittner@wicourts.gov>

From: "Kevin Grittner" <Kevin.Grittner@wicourts.gov>

To: "Greg Smith" <greg@2ndquadrant.com>, "Jon Nelson" <jnelson+pgsql@jamponi.net>

Cc: "Vitalii Tymchyshyn" <tivv00@gmail.com>, "david@lang.hm" <david@lang.hm>, "pgsql-performance@postgresql.org" <pgsql-performance@postgresql.org>, "Craig Ringer" <craig@postnewspapers.com.au>, <mladen.gogala@vmsinfo.com>

Date: 2010-10-12T15:29:22Z

Lists: pgsql-hackers, pgsql-performance

Jon Nelson <jnelson+pgsql@jamponi.net> wrote:
> Greg Smith <greg@2ndquadrant.com> wrote:

>> Usually the sequence used to remove all cached data from RAM
>> before a benchmark is:
> 
> All cached data (as cached in postgresql - *not* the Linux system
> caches)..., right?

No.  The stop and start of PostgreSQL causes empty PostgreSQL
caches.  These lines, in between the stop and start, force the Linux
cache to be empty (on recent kernel versions):

sync
echo 3 > /proc/sys/vm/drop_caches

-Kevin