On Thu, May 31, 2012 at 9:17 AM, Robert Haas <robertmhaas@gmail.com> wrote:
>
> Oh, ho. So from this we can see that the problem is that we're
> getting huge amounts of spinlock contention when pinning and unpinning
> index pages.
>
> It would be nice to have a self-contained reproducible test case for
> this, so that we could experiment with it on other systems.
I just posted a patch under subject "pgbench--new transaction type"
that introduces a pgbench -P option.
I think that that would do a good job of simulating unique-key
look-ups on the inner side of a nested loop (which is basically what
we have here) and so creating contention on index pages. Right now I
don't have anything with more than 2 CPUs and 2 is not high enough to
get much contention so I can't post any meaningful numbers. (pgbench
-P might also be of interest in hash index investigation)
Cheers,
Jeff