When changing to bigger tapes, label the new tapes with different
values.  If you have FOO_000 to FOO_019 now, use something like
FOO_020 to FOO_039.  Making 2 tapes with the same label just causes
trouble, and there is no reason to do it.  Once you have new tapes,
just swap them in.  Make sure you have a second drive around that can
read them; you'll want that when your main drive fails.

With tapetype, if you underestimate the tape size, nothing bad will
happen except that amanda will not pack as many dumps on when behind.
If you overestimate, you'll get end-of-tape errors.  We got an HP DDS3
drive (which died in a very bad way, silently writing bad data, and
has now been replaced by a seagate), and guessed at the size based on
the (uncompressed) capacity.  This worked fine when amanda was not
behind due to balancing which resulting in ~9GB being written.  But,
when dumps were missed and amanda was behind, the tape overflowed a
number of times.  When it does overflow, you can just flush.

I went over the logs for 4.5 months and looked at the sizes that were
written successfully and not successfully, and picked the following
length.  (At the time I was unaware of the 'once written with
compression, always compressed' possible tape issue, so it may be that
this is for gzipped data that then sees hardware compression.)

define tapetype HP-DAT3 {       # first guess at HP DDS-3 drive
    comment "DAT tape drives"
#   length 12000 mbytes         # may not be accurate (this is what box says)
# gdt 2002-04-18 from analysis 2002-01-01 to 2002-04-18
    length 10000000 kbytes
    filemark 100 kbytes         # but you get the idea
    speed 1000 kbytes           # from HP web page drive spec for DDS-3
}


It would be nice to have a script to go over the logs and produce a
plot of amounts that were written successfully and not in different
colors and perhaps estimate some 95% or so limit.
It would be really spiffy to have an option to write a file of random
bits to the tape until EOT if > 90% of the declared size had already
been used, in order to get better data, which could perhaps be
maintained per-tape, not just per-drive.  But it is probably not
worth the extra wear.

Reply via email to