Joe Martinez wrote:

> I am in the process of making my application multi-site right now with
> replication (I'm writing my own replication server). I thought about doing
> 2-part primary keys, but a much more elegant solution is to use
> GUID's. They are guaranteed to be unique, and don't require worrying about
> colliding values, plus it's a single field.
>

Umn, that kind of depends on what you want. There are applications
where knowing the origin of a record is a good thing. And, of course,
generators are a bit faster than calling a UDF to get a GUID. If you
are going to use GUID's, you can save a lot by inverting the bytes so
the heavily duplicated parts (indicating universe, solar system, planet,
computer manufacturer, and MAC address(? <- brain fade, what is that
part called?) come first so they disappear in index prefix compression.


Regards,


Ann