home *** CD-ROM | disk | FTP | other *** search
- Newsgroups: comp.unix.admin
- Path: sparky!uunet!destroyer!cs.ubc.ca!newsserver.sfu.ca!sfu.ca!vanepp
- From: vanepp@fraser.sfu.ca (Peter Van Epp)
- Subject: Re: Dumping a Gigabyte
- Message-ID: <vanepp.727719072@sfu.ca>
- Sender: news@sfu.ca
- Organization: Simon Fraser University, Burnaby, B.C., Canada
- References: <1j55eiINNk3d@BSDI.COM> <id.DZQW._25@ferranti.com> <1993Jan17.210451.14999@siemens.com> <761@sadis01.sa.aflc.af.mil>
- Date: Fri, 22 Jan 1993 16:11:12 GMT
- Lines: 49
-
- donr@sadis05.kelly.af.mil (DONALD G. RUEBENSON II - SCST) writes:
-
- >Some very good information in this thread. Sorry I missed the original
- >posting. One issue that has not been addressed: what about mirroring
- >filesystems and backing up from a offline mirror when necessary? Yes,
- >this is expensive (disk space $ * 2) but cheap compared to corrupt
- >backups when you need them, and would reduce the likelyhood of needing
- >the backups in the first place significantly. performance is usually
- >better on the read side as well, depending on the mirroring
- >configuration (full spindles preferred), by eliminating head contention
- >during the backup operation. However, few if any tape subsystems I have
- >had the pleasure of working with could take advantage of the
- >improvement. NOTE: if the mirrored disk is being used as storage for a
- >database (ala ORACLE, UNIFY) care should be taken to shut down the
- >database engine prior to taking the mirror partition offline for backup
- >or the above claims of data integrity are null and void. Once the
- >mirrored piece is offline the datbase can be restarted immediately and
- >can run during the backup operation. There is also no reason to halt
- >the database to reattach the mirrored piece - it will 'catch up'
- >automatically.
-
- >We run a Pyramid MIS-4/4, but the above behavior has been noted from
- >other platforms as well.
-
- > -- donr
-
- We run a (home grown) system like this on our Auspex NFS file server. I am
- curious about what you mean about the tape subsystem not taking advantage
- of it. How would a tape drive take advantage of it (ours dumps the mirror
- volume)?
- There is a possible additional restore advantage here too, you can
- use dd to dump the mirror instead of backup and avoid the time it takes
- restore to create directories on a full restore (you just dd the tape back
- to a new disk and fsck). Since (at least on Exabyte 8200s and an Auspex)
- you are limited by tape speed, the is no time cost to single file restores
- either, you just dd the tape to the mirror drive, fsck it and mount it
- read only somewhere, and all the files on the backup volume are availible
- for the users to pick and choose (and they can remain there until the
- next full backup). I have code that both locks /etc/dumpdates (so multiple
- dump streams work correctly) and lies through its teeth about what partition
- just got dumped (since it is always the mirror volume not the real volume)
- so that dump/restore incrementals without mirrors work as well. While I was
- at it, I also added labels to the tapes (and the code to check them in the
- dump scripts) so that we are sure that the correct partition got dumped to
- the correct tape (and it is an ascii file so that it can be dumped with
- dd from the tape if the tape label falls of).
-
- Peter Van Epp / Operations and Technical Support
- Simon Fraser University, Burnaby, B.C. Canada
-