The run with explain analyze is done. The result page is:
The analyze output is under 'Run log data' *.result.
I will try your suggestion today.
On Wed, 2004-03-03 at 20:52, Tom Lane wrote:
> > One thing that just occurred to me is that we renamed sort_mem to
> > work_mem. If you were trying to set a non-default value of sort_mem
> > then you need to adjust your script to call it work_mem.
> Actually, forget that --- there is backwards-compatibility code in there
> so that setting "sort_mem" still has the intended effect, even though
> it's not the preferred name anymore.
> > That's what the (lack of) I/O seems to suggest. But I don't understand
> > why we see no I/O *and* idle CPU. One or the other ought to be maxed out.
> I'm still confused about that one. On the power test it makes sense
> that we couldn't get better than 12.5% usage, since there's only one
> thread going. I don't see why that happens on the throughput test
> The only possibly relevant change that I can see is that spinlocks now
> have a delay instruction that is alleged to do good things on Xeon-class
> processors. Could it be that that hack is a tremendous loss on your
> setup? Try diking out the SPIN_DELAY() call at line 90 in
> regards, tom lane
Get latest updates about Open Source Projects, Conferences and News.