[comp.os.vms] How to BACKUP a big VAX?

ZSYJKAA@WYOCDC1.BITNET (Jim Kirkpatrick 307 766-5303) (02/18/88)

I'd like to solicit some comments (directly to me; I'll summarize later) on
how fairly large VMS sites back up user data.  By "fairly large" I mean a
site such as ours, with two 8800s in a cluster, and 4 SA482 cabinets
with an HSC70 and 4 Tx79 tape drives.  One argument I have heard is that
there is so much data space that we cannot afford to regularly back up user
files, so they're on their own.  Never mind the fact that if they actually did
regularly back up their data we'd be swamped with tape requests.  Our current
method deals with 10 of the 12 spindles and goes like this:
  1.  Once a week we shut down VMS and do a full volume backup on 2 spindles;
      the other 8 get an incremental save for the previous week.  This
      rotates on a 5-week cycle, so every spindle gets a full backup and 4
      weekly incrementals, interleaved throughout the cycle, due to the time
      it takes to do a full versus 1-week's worth of incremental.
  2.  Once a day we take an incremental for the previous 24 hours.  These
      tapes get re-used once a week since the "big" incremental of step 1
      obviates their utility once the big one is done.
  3.  Out of sheer paranoia (and experience on a Cyber) we are thinking of
      taking an "interim" incremental a couple of times a day, using the same
      tape(s) each day.  Just stopgap to minimize loss in case of disk failure.
I think I understand the tradeoffs of save time versus load time versus number
of tapes needed.  Is there some other effective way to back up a VMS file base
on a general-purpose University type system?  Does anybody get away with simply
not backing up user data?  Any horror stories?  (though we do not intend our
disk backups to be used as an "archive" system, it happens, which means we
cannot do a physical backup, unless somebody knows how to reload a single
file from such a tape).  One thought involves doing 4 BACKUPs at a time so that
they end up competing for the tape controller, almost ensuring a full 120 ips
tape throughput, where one at a time would spend lots of time diddling around
on the disk and not utilizing the tape full time.