On Tue, Aug 08, 2006 at 10:27:00AM -0500, Frank Smith wrote:
> Jon LaBadie wrote:
> > On Mon, Aug 07, 2006 at 02:37:17PM -0400, Ian Turner wrote:
> >> Marilyn,
> >>
> >> Amanda's tape selection policy is as follows.
> >>
> >> Consider the set of tapes T. We can partition the set into two disjoint 
> >> subsets A (the set of active tapes) and I (the set of inactive tapes). 
> >> Assuming I is nonempty, there exists a subset P of I, called the set of 
> >> preferred tapes. Note that T = A + I, and P is a weak subset of I.
> >>
> >> Amanda will only use tapes from I; active tapes are not considered for 
> >> overwriting. Also, tapes from P are preferred to other tapes in I; a tape 
> >> not 
> >> in P (but in I) will be used only if no tapes in P are available. If no 
> >> tapes 
> >> from I are available, then no tapes are used and Amanda will go into 
> >> degraded 
> >> mode.
> >>
> >> Tapes are assigned to each of the two sets as follows:
> >> -- Any labeled but unused tapes are in I and P. This includes unlabeled 
> >> tapes 
> >> if the label_new_tapes option is set.
> >> -- The most recently used "tapecycle" number of tapes is in A.
> >> -- Any remaining tapes are in I. The single least recently used of these 
> >> is 
> >> also in P.
> >>
> >> This algorithm is applied from scratch any time a new tape is needed 
> >> during a 
> >> backup run. You can run the algorithm without running Amanda by doing 
> >> 'amtape 
> >> taper'.
> >>
> >> What all of this means from a slightly less mathematical perspective is 
> >> that 
> >> Amanda will not consider overwriting the tapecycle most recent tapes. If 
> >> you 
> >> want to relax this restruction, just reduce tapecycle, and Amanda will 
> >> countenance the use of newer (more recently used) tapes.
> >>
> >> Alternatively, if you have a specific tape that you want Amanda to reuse, 
> >> just 
> >> relabel it, and it will be treated as a new tape.
> >>
> >> --Ian
> >>
> >> On Monday 07 August 2006 13:10, HUGHES Marilyn F wrote:
> >>> We have a situation where the next  Amanda tapes that it is asking for
> >>> are currently offsite.  It costs $75 for them to be retrieved so we
> >>> don't want to do that.
> >>>
> >>>
> >>> Besides we have available tapes here onsite.  Is there a way (a command
> >>> or other way?) to force Amanda to select one of the tapes that we have
> >>> onsite?  Does Amanda select from the top of the tapelist on down?
> > 
> > 
> > Ian,
> > 
> > Thanks for your description.  I was thinking of trying to put together
> > a description of the tape selection algorithm myself.  But I didn't
> > know some(most?) of the detail.  Certainly not in mathematical sets.
> > 
> > One thing still up in the air (to me anyway) is final tape selection
> > from within the tapelist and physical tape changer.  Your description
> > gets to which tapes are "eligible" to be selected, but not which tape
> > (or runtape number of tapes) among that set is ultimately chosen.
> > 
> > Let me try to use my own terminology to descibe your algorithm so
> > that I'm sure I understand it.  I'd appreciate corrections to any
> > mis-statements, in fact or in timing.  And, perhaps you could extend
> > it by describing the final tape selection.
> > 
> > From the entire tapelist, those marked "no-reuse" are eliminated
> > from further consideration.  Only those marked "reuse" are considered.
> > (is there any other tapelist classification?)
> > 
> > The reuse"able" list is divided into "previously used" and "never used"
> > (have a valid date stamp or have a 0 date stamp respectively).
> > 
> > The previously used tapes are date-sorted and the most recently used
> > tapecycle-1 of those are eliminated from further consideration.  This
> > would be your "active" set that are reusable but can't be overwritten
> > at this time as they fall within a tapecycle's number of tapes.
> > 
> > Any remaining, previously used tapes, plus the labeled but never used
> > tapes (if any), constitute the set of tapes eligible to be used for
> > the next run.
> > 
> > If what I've described is reasonably accurate, none of it is dependent
> > on tapelist order nor availability in the tape changer.  So the physical
> > device must be accessed before the final selection of a tape to use.
> > 
> > Does amanda at this point look specifically for the next tape based on
> > the order in the tapelist file?  (i.e. the last tape in the tapelist
> > file that is on the eligible list)  Will it scan the entire changer
> > looking for that specific tape?  Or will it start to scan the changer
> > looking for any tape from the eligible list?  Or something else?
> > 
> > Thanks.
> > 
> > jl
> If Amanda's behavior hasn't changed (I'm still using 2.4.5), it will
> use the first non-active tape it finds in the changer.  It appears
> to just load the next tape until it finds a non-active one, not look
> for an unused one or find the oldest.
> 

>From this discussion, your and my observations also, it appears the
order of tapes listed in the tapelist file is immaterial.

Anyone disagree?

-- 
Jon H. LaBadie                  jon AT jgcomp DOT com
 JG Computing
 4455 Province Line Road        (609) 252-0159
 Princeton, NJ  08540-4322      (609) 683-7220 (fax)