> As I wrote, I did the test with isql with direct connect to the DB file, not through fb_inet_server for FB1.53 as well as
> for FB2.0.

Nope, thats not the point, you built classic, even though you are using the local access mechanism for
classic is still classic, that means co-operative garbage collection where users clean up after each other
automatically when garbage is detected assuming that they can as defined by the current state of transactions.

Only SuperServer has threads and a GC colelction thread that runs and cleans up garbage in the background as
and when it gets noticed and can do so.

> I still think that something was changed between FB1.5 and FB2.0 to explain such a difference : why does it take just a
> few seconds for FB2 where it takes 2430 s for FB1.5 ?

A lot of work was put in by Vlad Horsun re. how garbage collection operates, and the issue re. duplicate FKs
may have been worked on too. Have a look at the 2.0 release notes for more info.

http://www.ibphoenix.com/main.nfs?a=ibphoenix&page=ibp_download_20

> BTW, I've just testing it and setting the sweep interval to 0 does not change anything for both FB1.5 and FB2.0.

It wouldn't :-), not in your scenario. A sweep takes place across the whole database when it needs to
happen. Setting the value to 0, just means that the database doesn't get swept. In your tests I doubt
that a sweep would ever run.

Please read: http://www.ibphoenix.com/main.nfs?a=ibphoenix&page=ibp_expert4 it will help.

Regards
Paul