> I agree - there should never be a PK collision (or threat of
> one) in a good
> replication system.
> You are suggesting the use of two fields to do the work -
> what's wrong with
> each database having an installation number. A table with one value,
> installation one to 100 or whatever.
> One generator per table for one field with full integer scope
> (even 64bit
> integer)
> And the other field being installation number which is
> inserted by trigger
> on every record insert. You can use the single PH field for
> local work and
> the replation service uses the twin field PK for replication.
> As I remember, there was a long discussion about this quite a
> while ago and
> this seemed to me to be the most respected solution.
> Alan

That is exactly what I'm planning on! I'm glad we're thinking alike!
The timestamp is a 'just in case' measure, and not used for any referential
integrity, or identification purposes.

Thanks all, and sorry for going over something that was discussed
earlire(very red faced!)

N.