>I looked primary key generation routine from Craig website and
>found that it uses flock() instead on rlock()
>
>This causes several perfomance slowdown if there are many
>different primary keys required concurrently.
>
>In code comment there is a note that there are problems using
>rlock().
>
>Can anybody give me information against using rlock()
>to lock record in a table used for sequential
>primary key generation ?

I found that there was contention at times on the keys table when using RLOCK, so I changed it and my problems went away.