Gene,

> I'm puzzled, 4 days, but 5 tapes?  Whats the 5th tape?

I thought that's what you're supposed to do: (tapecycle >=
runspercycle + 1). This will keep the 5th run from writing over the
first tape.

> Bear in mind that a full backup isn't any one tape, but every tape for 
> the most recent dumpcycle days will be required to do a full bare 
> metal recovery.

True. I didn't think of that. But this point to what I believe is a
weakness of Amanda:

Say I have a 1 week dumpcycle, runspercycle = 7. And for simplicity,
say I have 7 files: file_0 .. file_6. Also assume that every file is
changed everyday so that a level 1 is needed everyday.

Also assume that we only do levels 0 and 1.

On Sunday, Amanda does a level 0 for file_0, and level 1 for all other
files; on Monday, Amanda does a level 0 for file_1, and level 1 for
all other files, and so on.

At the end of the week, I will have tapes from Sunday through
Saturday. Let's call them "tape zero" .. "tape six."

Then my customer calls and asks me to restore a file as it was on
Wednesday. I can do that easily for file_0, since the level 0 for that
file is on "tape zero," backed up on Sunday; and I also have all the
necessary level 1's for it. So I grab tape zero (Sunday's level 0) and
tape three (Wednesday's level 1), run amrecover, and voila.

But if I were to need to recover file_5 "as it was on Wednesday," I
would be out of luck. This is because among these 7 tapes, the only
level 0 for file_5 is on tape five, backed up on Friday. All level 1's
from tape zero through tape 4 for file_5 are useless unless I happen
to have preserved the level 0 from the previous cycle.

What this means is: unless I preserve tapes from the previous cycle,
tapes for one cycle can only guarantee that a file can be restored to
its state _sometime_ in that cycle.

The more traditional full/incremental backup scheme allows you to,
in the case of a 7-day cycle, have the confidence that you can restore
a file to its state on any of the last 7 days. But that isn't true
with Amanda. :(

Regards,

Ivan