[Gllug] Production system - Linux 2.4.24, LVM and cciss

[Simon A. Boggis]

Hi Dale,

just a comment on the tape angle - noted that you're sticking to a
"copy" backup on disk for now, but I though this might be useful for
further down the line.

> These are hot-swap drives, so one on-site, one off-site.  Will do the
> DLT/DAT thing later.  This setup is enough for now, as the DAT solution
> is more expensive, slower to restore, more susceptible to data
> corruption... basically, sticking to what's been ordered for now.

Note that DAT != DLT. I think DLT has many superior features, not least
write speed and tape size (last series of DLT IV is 40 Gb native). These
days I wouldn't buy DLT but rather LTO, since it comes in native sizes
of 100 or 200 Gb with high write-to-tape-speeds (c.a. 35 Mbs-1?). Sony
have AIT, but I steered clear due to some propriatoryness in some
on-tape flash storage.

When we costed LTO last year, the price per Gb of storage wasn't so far
from storing on disk when considering the same size of storage - about 4
grand for a single tape LTO drive (or 8 grand for a small changer) with
three years on site warranty; about 100 UKP per 200Gb native tape, and
the reliability was much higher, so we went for tape (again).

Agreed that it is slower to restore than a live copy on disk.

But, DLT or LTO have properties that you can't buy easily with disk
backups:

  * good for a million tape head passes
  * storage life of 30 years
  * your backups spread across N such media, rather than (above) 2
    disks: this gives the possibilty of back tracking by up to N days of
    backups in case of corruption or user error in files on the
    filesystem, with more possibility to recover.
  * possibility to store many (small and not too delicate c.f. disk)
    media offsite.
  * verification of data on media after it is written, on-the-fly, with
    on-the-fly error recovery - most drive systems read the tape just 
    after it is written to check that the write worked OK.

I've had DLT for several years, with many tapes, and my own experience
bears out the makers reliability claims - I've not had a tape failure
(aside from a couple of tapes getting temporarily stuck in a dying tape
drive, recovered when drive repaired or replaced). It's possible for the
pickup ends to come off the tapes (not happened to me personally), but
this doesn't damage data (plenty of leader) and this is repairable with
a simple kit.

You're considering RAID with high redundancy for your filesystem so I
presumably don't need to tell you how likely and often disks fail - your
proposed system is twice as likely to fail as a single disk, and if it
does you loose 50% of your backups instantly. 

I would suggest that the proposed system is only good for disaster
recovery, because it gives little protection against user error
destroying files (or contents) or software or hardware error corrupting
files - you only have two backups, which isn't much history if you
backup every day, as you should (and if you only backup once a week I
hope you're sure that your users don't mind loosing up to a weeks worth
of work/mail/whatever!).

So I would disagree with your assertion that tape is more susceptible to
data corruption than disk for the above reasons - what is the your
reasoning?

My 2p is that under current conditions, I'd buy DLT or LTO for backup of
production systems.

Simon

-- 
Gllug mailing list  -  Gllug at gllug.org.uk
http://lists.gllug.org.uk/mailman/listinfo/gllug