Re: Some performance numbers, with thoughts

Brian Hurt <bhurt@janestcapital.com> writes:
> For long involved reasons I'm hanging out late at work today, and rather
> than doing real, productive work, I thought I'd run some benchmarks
> against our development PostgreSQL database server.  My conclusions are
> at the end.

Ummm ... you forgot to mention Postgres version?  Also, which client and
server encodings did you use (that starts to get to be a noticeable
issue for high COPY rates)?

> 1) Transaction time is a huge hit on the small block sizes.

Right.  For small transactions with a drive honoring fsync, you should
expect to get a max of about one commit per platter revolution.  Your
numbers work out to a shade under 5000 commits/minute, from which I
speculate a 7200 RPM drive ... do you know what it really is?

            regards, tom lane