Thomas,

> > Wow, that's a hugh difference. I suspected there would be some
> > penalty, but never thought it would be that great. Nice going on the
> benchmark.
>
> The basic outcome with additional tests was, that using a numeric pk value
> pretty much behaves as a GUID, if the numeric pk value isn't feed by a
> generator/sequence providing an increasing number, but with a somehow
> random value. It's possibly related to keeping the index tree in-sync with
> newly inserted records.

What was the size of the page cache?

Does performance change if you increase the page cache?


Sean