Jon LaBadie <jon AT jgcomp DOT com> writes:

> The strange observations.
>
> - first dump it picked to tape was not the largest,
>   but the 3rd largest at 5500MB
> - second dump it picked was the 28th largest at 8MB
> - then back to a big one, 8th largest at 2500MB
> - two more files were taped bringing the total taped to
>   8900MB, leaving about 2700MB of claimed capacity

As I observed this: Even with largestfit amanda flushes in time order,
so all holdingdisk files of the oldest run are flushed first, then the
second-oldest DLEs on holdingdisk and so on.

There are pros and cons, but this could explain your observations.

        Sven