On Fri, Sep 29, 2006 at 08:10:23 -0700,
Richard Broersma Jr <rabroersma@yahoo.com> wrote:
> > The most difficult part of this question is justifying WHY we would
> > want to use random primary keys! There is a very strong reason for
> > doing so, although not quite compelling.
>
> One problem with using random generated primary keys that I've
> recently read about deal with insert failing do to primary key
> duplication.
>
> If the size of your dataset grows to become a significant percentage
> of the size of the integer type used for your random primary key,
> the probability of inserting a duplicated number dramatically
> increases. I imagine that this problem could contribute to poor
> preformance for large bulk inserts that have to add logic for
> dealing with re-trying a insert if a duplicate number is created.
They are using 128 bit keys! If their random number generator actually
works, they shouldn't have a problem until they have generated on the order
of 2^64 keys. That isn't likely to happen any time soon.