On Fri, Feb 28, 2014 at 1:07 AM, Jon LaBadie <[email protected]> wrote:

> On Thu, Feb 27, 2014 at 07:19:24PM -0500, Michael Stauffer wrote:
> > Amanda 3.3.4
> >
> > Hi,
> >
> > Another long post from me - thanks to anyone who has time to read it.
> >
> > I've been reading various docs and posts about dumpcycle, runspercycle,
> > tapecycle, runtapes online, but still can't figure out how I should set
> > things up for my needs.
> >
> > I've got:
> > - ca. 30TB of data to backup
> > - I'll try to split this into 500GB or 1TB DLE's, but some will be much
> > easier to leave as 2-4TB DLE's
> > - I'm fine with a level 0 only every 30 days, especially considering a
> full
> > backup will take 8-10 days.
> > - I've got 1.5TB tapes that hold about 2TB of data with hardware
> > compression from testing
> > - my changer holds 35 tapes
>
> First question, will you be using more than 35 tapes.  I.e. will you
> periodically pull some recently used and replace with less recently
> used tapes?  If not, I think you are short on tapes.  A full dump will
> take 15 tapes.  You really want a MINIMUM of two full dumps.  I prefer
> more.  So 2x15 is two full dumps leaving only 5 tapes for incrementals.
> I don't think that will be enough.
>

I was thinking of doing a periodic (at the start and then every 3 months)
level 0 archive dump to a different set of tapes (probably four sets to
retain 1 year's worth of dumps). Then I thought the library in the changer
would be fine if it held less than two level 0 dumps at any time. I'd
rather just go switch tapes once every 3 months than more often, and have
offset archive too. Does that seem reasonable?


>
> > I'd like to have the changer always hold a level 0 dump and then the set
> of
> > subseuqent incrementals. So it should take about 15-20 tapes for a level
> 0
> > of all DLE's, and then the incrementals over a month should easily fit
> > within 2-4 tapes, judging from experience here. When the next level 0
> dump
> > starts, I'd like amanda to use as many of the remaining 10 or so tapes
> > before overwriting tapes from the old level 0 dump (overwriting only
> tapes
> > of DLE's that have just had a new level 0, of course). (I will
> periodically
> > do a level 0 dump to a different tape set for offset archiving)
> >
> Do you plan to let amanda do the scheduling?  Or are you going to force
> her kicking and screaming into the traditional schedule of one monster
> full dump followed by all incrementals.  Then another monster.  Blech.


I'm fine with Amanda's scheduling. When I do my first round of amdump's
though, it will be effectively a monster dump until all DLE's have a level
0. Then I presume amanda can even things out using her own scheduling?


> > How do I setup dumpcycle, runspercycle, tapecycle, runtapes to achieve
> this?
> >
> > >From the docs, it seems I'd want:
> > dumpcycle   30 days
> > runspercycle 15  #15, for running amdump every other day
> > runtapes       2   # to allow for DLE's that can get up to 4TB
> > tapecycle     34  # at least (runspercycle + 1) * runtapes - per docs
> > suggestion
> >                         # and leave one extra as a spare
> >
> > Is this right?
> >
> > Does this mean that when I run amdump, it will at most write two
> > tapes-worth of DLE's, and then stop? Then the next run will pick up from
> > there? I think so, but would like to make sure. I'm used to the manual
> > paradigm of "run a full backup" and then do incrementals. But this seems
> > that it will level out to be the same in the end as that?
> >
> > HOWEVER, I'd rather have runtapes at 3 or 4 to minimize tape waste and
> make
> > it less critical to have evenly-sized DLE's, which will be difficult to
> > maintain. But if runtapes is 3, the recommended value of tapecycle would
> be
> > >= 48, more than my # of tapes. But in practice, 35 should still plenty
> of
> > tapes to do what I want without overwriting level 0's prematurely. It
> seems
> > like tapecycle minimum should be more like '# of tapes per full backup +
> #
> > of tapes for incrementals over dumpcycle + 2 * runtapes', plus one or two
> > as a buffer.
>
> The following sentence shows you are still thinking the "non-amanda way.
>
> >               The formula in the docs of "(runspercycle + 1) * runtapes"
> > plays it very safe when you consider many incremental dumps will go to
> > holding disk and be collected onto one tape periodically.
>
> There are not going to be a bunch of incrementls to collect onto one
> tape.  Each amdump run will be a mix of level 0's for some DLEs and
> incrementals for the others.
>
> With 15 runspercycle you will AVERAGE 2TB of level 0 plus ??GB of
> incrementals.  But remember you said some DLEs will be 4 or more TB.
> When they get level 0's you'll need more than 2 tapes.


Can I be sure that there will always be some level 0's in a run of amdump?
I guess because there's ~30TB of data and 15 runs, so that pretty much
guarantees a level 0 every run? If not, or if they're << tape-size, I'd
like them to go to holding disk so as not to greatly underutilize a tape.
I'm going to try to have my largest DLE < 2TB - there will be 3 or 4 of
those probably - for simplicity's sake.

However it's not clear, but it seems like when using a holding disk the
whole DLE is written to holding disk before streaming to tape, which seems
inefficient time-wise when the DLE is large enough that's going to get
streamed to tape anyway. I guess I have to work params something like this:

  # to have amanda start writing when 10% of volume capacity is written to
holding disk, and
  #  flush-threshold-scheduled is satisfied
  flush-threshold-dumped 10

  # have amanda write to tape if the holding-disk storage plus planned data
exceeds this
  #  % of tape capacity
  flush-threshold-scheduled 90

  # taperflush param not needed if above two params are used
  #taperflush 90

Do I understand right that these above settings will tell amanda to start
streaming to tape relatively early in a dump process if it knows it's going
to end up with >= 90% of volume capacity on the holding disk? I'm thinking
this should make the process more time efficient since it won't write
everything to disk before streaming to tape?

Thanks again

-M

Reply via email to