>> I once run a test with insertion performance in Firebird 2.5 with a)
>> sequential PK values coming from a generator and b) via the built-in
>> function giving me a UUID.
>
> reversed UUIDs, by putting the fixed part at the front, should perform
> better in this regard than plai UUID.
> I don't know if the function you have used generates that kind of UUID.
> I know that I once used a UDF that did that and performance was better
> than with (at the time, client-generated) UUIDs.
> High/Low is also good, as huge batch of inserts would inherently use
> sequential IDs, but this requires a little more setup.

Sorry for the confusion.

My test case was a bit different than I first thought. I measured the
penality of additional indexes beside a PK constraint when inserting
bulk data and how an increased page size has influenced the test result.

Regards,
Thomas