[Gllug] Tape backups

[Nix]

On 16 Feb 2009, Martin A. Brooks uttered the following:
> -----------------------------------------------------------------------
> A copy on disk is a copy.
>
> A backup is an indexed archive that guarantees a point-in-time recovery 
> of data to a time resolution decided by local policy. A backup is 
> distributed, is stored off or near line, and is non-trivial to corrupt 
> or delete.  A backup is on non-fragile media - a hard disk drive will 
> almost certainly not survive dropping 1 meter onto concrete, a DLT tape 
> almost certainly will.  A backup lets you see what changed, when (within 
> the above set time resolution), and preferably whom it was changed by.
>
> A backup is what you use when your data centre has burned down, all the 
> hardware is beyond recovery and you somehow have to get your production 
> systems operational again.
>
> In this scenario if I turned round and said "Oh sorry, I copied all that 
> important stuff onto a RAID5 set, well, that's sort of a backup, isn't 
> it? And would you know it, it's corrupted on me!"  I'd expect to be 
> stared at in wild-eyed disbelief followed by laughter as they realise 
> that no-one, no-one who has ever dealt with mission critical data, ever, 
> could possibly confuse a backup system for copying stuff onto a bunch of 
> hard disks.
>
> And then, as I have done mercifully few times before, I call up Hayes, 
> who's number has been in my mobile phone for a long time, and say 
> "Please send our last full backup set, and all the dailies since, on a 
> bike as soon as possible please."
> -----------------------------------------------------------------------

By that definition there is no real point individual users ever taking
'backups' at all: all backup media that satisfy those constraints are
intolerably expensive.

It's sufficient to take an additional copy that requires conscious
effort to change thenceforward, to some other medium which is not likely
to be damaged *at the same time* as your primary copy. Hard drives are a
perfectly good example: while they fail, drives from different batches
or of different makes don't fail simultaneously (if you leave it
unplugged, not even if your house is struck by lightning).

If my house has burned down and all my hardware is beyond recovery, what
I care about is *not* my data. What I care about is that, well, my house
has burned down. Not even an ubergeek is going to care *that* much (and
in any case the critical stuff is backed up on CD-Rs at work every six
months, but that doesn't count as a backup by your criteria because
CD-Rs are notably non-robust).

And as for hard drives 'corrupting on you', well, unless the OS goes
insane it simply doesn't happen often enough to worry about. They have
mechanical failures, but the last time I can recall a hard drive having
electronic failure leading to disk corruption was in 1986. One instance
in thirty years (and that in a major storm) is an acceptable failure
rate in my book (and given that the box was running DOS, it was just as
likely that the excuse-for-an-OS shagged itself and damaged the drive).

If you're worrying about this sort of thing, you should probably worry
about meteor impact recovery too. In the long run, it's much more
probable.

-- 
`We must stand together and fight for our shared cultural heritage as a
 group of people who cannot stand together to fight for our shared
 cultural heritage.' --- jspaleta on Balkan balkanization
-- 
Gllug mailing list  -  Gllug at gllug.org.uk
http://lists.gllug.org.uk/mailman/listinfo/gllug