>Here is a simple scenario for headache:
>On day one 10 mobile workers download data on their laptops. One of the procedures have a composite/aggregate data (like employee-customer-order-invoice). They work hard for 3 days and each acquire 10-20 new customers which with their detail lines that may sum up to thousands rows each. They come back and upload the data. You know that there is no collision and it's a simple insert into but since identity was PK it is no more that simple and an ID fixup should take place. When you cahnge data structures you need to remember updating your fixup handling.

Cetin,
You have a point here but using this argument to generalise the use of identities as PK as a bad idea is a bridge too far ...

In this particular scenario, what do you propose ? UniqueIdentifiers ?