On Thu, 2003-02-13 at 00:16, Christopher Kings-Lynne wrote:
> Even if you look at the attached charts and you think that 128 buffers are
> better than 8, think again - there's nothing in it. Next time I run that
> benchmark it could be the same, lower or higher. And the difference between
> the worst and best results is less than 3 TPS - ie. nothing.
One could conclude that this a result of the irrelevancy of wal_buffers;
another possible conclusion is that the testing tool (pgbench) is not a
particularly good database benchmark, as it tends to be very difficult
to use it to reproduceable results. Alternatively, it's possible that
the limited set of test-cases you've used doesn't happen to include any
circumstances in which wal_buffers is useful.
We definitely need some better benchmarking tools for PostgreSQL (and
no, OSDB does not cut it, IMHO). I've been thinking of taking a look at
improving this, but I can't promise I'll get the time or inclination to
actually do anything about it :-)
Cheers,
Neil
--
Neil Conway <neilc@samurai.com> || PGP Key ID: DB3C29FC