On Sun, Aug 21, 2005 at 09:50:10PM -0400, Tom Lane wrote:
> Satoshi Nagayasu <nagayasus@nttdata.co.jp> writes:
> > I've done a quick hack to implement PCTFREE on PostgreSQL.
> > ...
> > According to my experiments, pgbench score was improved 10% or more
> > with 1024 bytes free space.
>
> I'm not very enthused about this. Enforcing 12.5% PCTFREE means that
> you pay 12.5% extra I/O costs across the board for INSERT and SELECT
> and then hope you can make it back (plus some more) on UPDATEs.
> pgbench is a completely UPDATE-dominated benchmark and thus it makes
> such a patch look much better than it would on other workloads.
>
> I think the reason Oracle offers this has to do with their
> overwrite-based storage management; it's not obvious that the tradeoff
> is as useful for us. There are some relevant threads in our archives
> here, here, and here:
> http://archives.postgresql.org/pgsql-patches/2005-04/msg00078.php
> http://archives.postgresql.org/pgsql-performance/2004-08/msg00402.php
> http://archives.postgresql.org/pgsql-performance/2003-10/msg00618.php
It should be possible to see what the crossover point is in terms of
benefit using dbt2 and tweaking the transactions that are run, something
I can do if there's interest. But I agree with Satoshi; if there are
people who will benefit from this option (which doesn't hurt those who
choose not to use it), why not put it in?
--
Jim C. Nasby, Sr. Engineering Consultant jnasby@pervasive.com
Pervasive Software http://pervasive.com 512-569-9461