>...  But on an active
>filesystem, won't you get to level 9 early in a long dumpcycle, increasing
>the total size of your dumps?  ...

Yes.  However, note that an incremental is everything changed since
the previous incremental level.  Put another way, a file changed on the
first day after a full dump and not again will only be on the level 1
incrementals.  When incrementals advance to level 2, it won't be on those.

Now, if all the changes are to "new" files (truely new or constantly
modified), then yes, the incremental size does ramp up.  Amanda slows down
the increment number increases via the "bump" paramenters.  In general,
going from level N to N+1 requires twice as much savings as going from
N-1 to N did.

Playing with the bump* parameters to keep the incrementals from jumping
up too quickly often goes hand in hand with messing with dumpcycle.

>I can see that on some partitions you may
>never even need an incremental, even on a long dumpcycle ...

And Amanda has configuration parameters to skip doing incrementals on
such completely idle file systems.

>but depending on
>your mix of stable and churning filesystems, there has to be a maximum
>length of dumpcycle before the total size actually starts increasing instead
>of decreasing as the cycle is lengthened.  Or is Amanda smart enough to
>decide to schedule an early level 0 (or drop back to 8 or 7 or ..) instead
>of yet another large level 9?

Remember that dumpcycle is the *maximum* amount of time between doing
full dumps.  It does not say full backups will only happen every
dumpcycle runs.

The guiding principle in Amanda is balance (*).  It should (although
I'm not an expert on this area of the code) "do what it needs to do" to
adjust the backup levels to achieve balance, even if that means running
full dumps more often than the dumpcycle value.

As to your last point, Amanda doesn't base a decision on going to a
full dump early on the size of the incremental, although I've sometimes
wished it would.  If a full is going to be very close to the size of
an incremental, it might as well do the full just to make any potential
restore that much easier.

>Frank

John R. Jackson, Technical Software Specialist, ITaP/RCS, jrj AT purdue DOT edu

(*) Yes, I know some of you see "odd" balancing going on.  No, I don't
know why it does that (without getting my PhD in Amanda balancing
algorithm analysis :-).