OK, I'll take a swing at this :)

On Wed, Apr 1, 2009 at 3:50 PM, Stefan G. Weichinger <sgw AT amanda DOT org> 
wrote:
> I want to avoid to copy ~60 GB from disk to holdingdisk internally at
> every amdump, so I decided to set them to "holdingdisk no" --> write the
> to tape directly.
>
> taperalgo first, btw

The taperalgo decides which dump, among those available in holding,
the taper will choose.  I think the one you are thinking of is
dumporder.

> Now they sit there and "wait for dumping", I would like them to be taped
> asap and the taper shouldn't get idle waiting for other dumpers until
> everything local is taped, correct?

However, the driver always schedules its PORT_WRITE (direct to tape)
dumps *after* its holding-based runs.  The idea is that dumping
directly to tape basically monopolizes the entire server
infrastructure, so it's better to get the cooperating dumps out of the
way first.

So the behavior you're seeing with this particular config is intended,
although it's not what you expected.

Dustin

-- 
Storage Software Engineer
http://www.zmanda.com