On Fri, 2008-07-11 at 15:25 -0600, Scott Marlowe wrote:
> On Fri, Jul 11, 2008 at 3:24 PM, Simon Riggs <simon@2ndquadrant.com> wrote:
> >
> > On Fri, 2008-07-11 at 14:51 -0600, Scott Marlowe wrote:
> >
> >> I would kindly disagree. I'm looking at a project where HOT updates
> >> are going to be a real performance enhancement, but I'll have to
> >> create a hundred or so tables ALL with fillfactor tacked on the end.
> >
> > You clearly think that adjusting fillfactor helps in all cases with HOT.
> > I disagree with that, else would have pushed earlier for exactly what
> > you suggest. In fact, I've has this exact discussion previously.
>
> How odd, because that's clearly NOT what I said. In fact I used the
> single "a" to describe the project I was looking at where having a
> default table fill factor of < 100 would be very useful. OTOH, I have
> stats databases that have only insert and drop child tables that would
> not benefit from < 100 fill factor. For a heavily updated database,
> where most of the updates will NOT be on indexed columns, as the ONE
> project I'm looking at, a default fill factor would be quite a time
> saver.
I apologise if my phrasing sounded confrontational.
For specific workloads, tuning of particular tables can be effective,
I have not heard of evidence that setting fillfactor < 100 helps as an
across-the-board tuning measure on longer-term tests of performance.
Theoretically, it makes little sense, but current theory is not always
right. Until we have even hear-say evidence of benefit, introducing a
parameter would be inadvisable, IMHO. I will change that view in an
instant, with reasonable evidence.
--
Simon Riggs www.2ndQuadrant.com
PostgreSQL Training, Services and Support
