FYI,
I just patched the fill-factor calculation and re-ran my test.
The index size dropped from 513M to 43M which is the same disk
footprint as the corresponding btree index.
Have a nice weekend.
Ken
On Fri, Jul 18, 2008 at 12:23:14PM -0500, Kenneth Marshall wrote:
> I just ran my original 16M word test case against the patched
> version, and like Tom noted below, the tuples per bucket
> calculation is wrong which results in identical index sizes
> for both the original version and the hash-value-only version.
>