Jon, thanks for the whole overview. THAT’s what was needed!!
> On Nov 18, 2015, at 10:35 AM, David Simpson <david.simp...@eng.ox.ac.uk>
> wrote:
>
> An important point. Perhaps it would be prudent to change to this (?) to
> reduce the nightly volume:
>
> dumpcycle 2 weeks
> runspercycle 10
> tapecycle 8
>
>
It’s not locked in stone, once you start. You can make adjustments later.
Still — if you have 10 runspercycle, you DO need to have at least 10 tapes
in the tapecycle. So you’ll have to change the tapes
once a week — but you were planning to do that anyway, back when your cycle
was going to be one week long. So maybe allow
tapecycle to be 16, and plan to change out the tapes each week, but that 2
weeks worth of runs constitute “one cycle”.
Or something along those lines. As you find out how well your data
fits onto the tapes, you can change the numbers.
Amanda will spend some time each night analyzing the data, to see what the
sizes of dumps are (incremental and
fulls) and how it needs to split things (PLANNER). Then it’ll start doing
backups, to the holding disk (DUMPER). Then
when some DLE (or partial DLE, if you allow splits) is done, on the holding
disk, TAPER will start writing things to the tape.
After a day or two, you will learn whether there is time overnight to write to
more than one tape.
Plus, is amanda running on a dedicated machine? If the backups have been
done overnight, it doesn’t matter if the
taping continues on into the day, does it? Would that bother other users,
if the holding disk and the tape unit are on the
same machine and not loading down the network?
So you might want to have runtapes 2 so she can write to 2 tapes each
night, iff that won’t bother your users. It would
certainly help get the data backed up faster.
Lots of parameters you can plan with!
Deb Baddorf
Fermilab
