Bumped to amanda-users since no response on amanda-hackers.

On 2/15/07, Terry Burton <[EMAIL PROTECTED]> wrote:
Hi,

I wouldn't like to guarantee that I have completely understood the
intended tape selection algorithm so please correct me if this
suggestion does not make sense.

I am running a multi-tape random access tape changer with a barcode
scanner. My tapelist contains a number of entries that at least for
the moment contain an ordered sequence of labels, some of which have
been used for backup (date != 0) and others are new (date == 0).

After a dump/flush, amadmin.c reports "The next %d new tapes already
labelled are: %s"... which successfully picks the next few tapes from
the bottom of the tapelist using lasttp =
lookup_tapepos(lookup_nb_tape()). This is a sensible selection of
tapes based on the current tapelist.

However, the amtape_taper_scan function of amtape.c invokes
lookup_last_reusable_tape(0) which will in this case return NULL to
signify that any new tape is required. Hence, amtape Daily taper will
settle on whatever new tape is in or ahead of the current slot, rather
make the better choice of the "next tape already labelled" (as derived
in amadmin.c).

Consequently, in order to maintain our labeling scheme (necessary
since the tape are collected from offsite), our backup operators have
to load the tapes in ascending label order and perform amtape Daily
reset/first to reset the current slot, or alternatively they could
load the tape in ascending order from the position of the current slot
without running invoking amtape.

I suggest making it an option to (possibly) trade in fast tape
selection for the sake of more intelligent tape selection so that when
this option is enabled amtape Daily taper should first attempt to
select the "next tape already labelled" (tapelist-based) and if this
tape is not available then fall back to "the any new tape"
(slot-based) algorithm, assuming of course you have a random access
changer with a barcode unit. In the case that the changer is inventory
aware (via a barcode scanner), there would be no performance penalty
and this could become the default selection algorithm.

The benefit would be that people running random access, inventory
aware changers would be able to load their tapes in random slots and
change tapes without resorting to the amtape utility.

I look forward to hearing if this is possible.


Many thanks,

Tez

Reply via email to