>>One more question; on one server the Vacuum Analyze before the insert takes
>>approx. 2min after that the same command takes 15min.
>
>
> You might try a VACUUM FULL sometime when you can deal with 15min of downtime
> or so. Actually it would probably be longer. Perhaps the table that's taking
> 15min has a ton of extra dead tuples left over from the fsm settings being too
> low and/or vacuum being too infrequent.
Does VACUUM FULL just lock entire tables and thus cause you to
essentially have downtime on that database because it doesn't respond
quickly or do you actually have to shut down postgres to safely do a
vacuum full?