Michael,

> I have posted several messages earlier about my DB getting corrupted.
> I use Firebird version 1.0
> It runs on a server running 2000 with all SP installed.
> I have gotten my hardware checked, and found no errors.

And also you have said multiple machines on this site all have the same
problem.

> On my server there runs a program called NTVDM.EXE which at times
> takes up to 100 % of CPU time (and 90 % kernel time)

Have you tried to downgrade the priority?
What does it do?

> Then after my DB gets corrupted, I have these values when running
> GStat:
>
> Oldest Transaction 4848
> Oldest Active 28304
> Oldest Snapshot 28304
> Next Transaction 28306

What happens if you change the sweep interval (e.g. 0, 40000)?

> To solve my problem I have to do a Backup - Restore.
> Then it all works perfect.

Are the errors all index and orphaned back versions or actually data pages
as well.

> Now my new question:
>
> Could it be, that my DB doesn't get enough time (because of NTVDM)
> when the sweep function kick in ?

You never know, you have an unusual situation. You need to remove
possibles:
1) Establish it is happening when the sweep is occuring.
2) If yes to the above reduce the priority of the NTVDM when the sweep is
about to kick in.
3) Remove the NTVDM when the sweep is about to kick in.

> Does sweep ALWAYS sweep when the difference i 20000 exactly or just
> shortly after ?

Don't know.