[snip]

> Yeah, something similar. My reason is not the inefficient usage of a
> tape (i'm using the virtual tape setup). I only want to achieve this,
> so I can be sure that the largest dump is finished first. In other
> words, when *production*  starts in the morning the backup on that
> specific DLE (venus) has been done (roughly 120 GB in that directory
> alone) and the gzip/tar processes won't be active anymore. As some
> people might say that it could slow down production time. I start the
> dump on 9:00 PM, the estimate for this host (venus) is finished at
> around 9:35 PM. But there is another DLE (mars) that roughly has 80 GB
> in a similar directory. That host (mars) is finished a little later
> with the estimate (around 21:40) but it still gets dumped before the
> *largest* DLE (venus)

Dumping starts when all the estimates are done.
Amanda needs the estimates to decide what to dump.

I thought that there *might* be a rule in the planner that could decide:
if dump is too large for holdingdisk, then give fast estimate time a
higher priority in the order of DLE's to be dumped. I guess that's not
true :-)

>
> I have added:
>
> dumporder "SSSS"
> taperalgo largest
>
> to the amanda.conf file (using 4 dumpers as a maximum).


When you are not using a holdingdisk, there is effectively only
one dumper running.  If you are using a holdingdisk, but some of the
DLE's don't fit, then those DLE's are always scheduled at the end of
the run.
Actually, I'm not exactly sure of that last statement (and don't have
the time to read the source code), but it is what I observed when I
had that situation a few years back.  Maybe it is a side effect of
the fact that as long as there are dumps to flush in the holdingdisk,
then the dumps bypassing holdingdisk have to wait, because the tape
is still busy.  Maybe with one dumper looking for the largest dump to
start first, this does not happen.
I would like to know too.

I have removed the taperalgo line. Unfortunately we're not yet running
a similar setup in a test environment, but I'm planning on setting up
a very realistic test environment. Hopefully I can speed up the
answers :-)

[snip]

You could specify a starttime in the dumptype for all DLE's but the
largest, that is just a little larger than the estimate time:
     starttime 2200
Once the large dump is started, the others have to wait (or dump to
holdingdisk in the meantime) until the large one is finished.

I could try that too, although I'd prefer to try without taperalgo
first. I've got quite an extensive list with dumptypes. Instead of
expanding the dumptypes, I was hoping to get rid of the unused ones.

Thanks,

Dennis