WHen you say you "manually balanced it", what does that mean?  Using 
  "amadmin" to bump things around?  My main goal is to have no operator 
  intervention in this.

I commented out half the total size worth of DLEs, IIRC using amadamin
balance to see the sizes.  Then after one dump I put them all in.

  I've also fully seperated my configs, and set the "norecord" option in the 
  archive config.

<aol>Me too.</aol>

  > Sometimes I am tempted to not disable incrementals to get balancing to
  > work right, and at others to modify no-incr so that it schedules
  > incrementals but does not include them in size or actually do them.

  That sounds interesting as well, could you expand on this?

Normally amanda does a dump of every DLE every run.  I'm a bit fuzzy
on the precise details, but basically she calculates which ones are
'due', and sets those to 0, and sets the rest to 1.  There is some
method to bump incrementals to level 2, but I think that's irrelevant
to the problem at hand.  Then, 0s are changed to 1 if the total size
is bigger than the tape (with a warning, since something that's due
won't get a 0).  The, incrementals are changed to 0s if the new size
would be less than the tape and the total full size will be less
than/near the total of all fulls divded by dumpcycle.

With a new config, this results in all fulls the first run, as many as
will fit, and then fulls of those that haven't had a full, with incrs
of the ones that have, until all get a full.  Then, you have a grossly
unbalanced config, nothing is due, and things get promoted.  After a
few cycles things are balanced.

In my setup (2.4.4ish), this balancing process doesn't work on my
archive config, I think because there are no incrs to get promoted.
So I think it might help to change the no-incr rule to say instead:

  schedule incrementals
  when calculating the total size of the dump, don't count incrementals
  when dumping, don't actually do the incrementals

Then, the whole 0/1 promotion/delay scheme should work, and the fulls
should get balanced across the runs.  I may well be missing something
- I have never read all the planner code.