[PLUG] Re: Large Data Back-Ups to CD-R(W)

[Alan]

On Mon, 2004-05-03 at 09:38, Keith Lofstrom wrote:
> Robert McIntosh writes:
> 
> > I'm working with a large volume of data that's closing in on 20 GB that
> > I'd like to back-up bi-annually to multiple CD-R(W)s.  I'm familiar with
> > Hugo Rabson's powerful Mondo Rescue (http://www.microwerks.net/~hugo/)
> > but in my particular environment (FreeBSD) it seems to be a real push to
> > configure the prerequisite packages into FreeBSD.
> 
> <soapbox=1>
> 
> At the risk of missing an important reason to use CD-R disks, may I 
> suggest you look at  http://www.keithl.com/linuxbackup.html ?  The
> disk-to-disk RSYNC backup techniques are fully portable to BSD and
> other forms of Unix. 
> 
> Bi-annual backups are risky - chances are, when the source disk fails,
> you will have an average of 3 months of data entry lost.  Scary.
> 
> More importantly, writable CDs fail frequently and are expensive; I
> see about 5% failure-to-verify on the "good" brands of media (I like
> Fuji and Memorex), and 10-15% on "bad" brands (Imation, Dysan, Verbatim,
> TDK, store-brand-du-jour).  So 20GB -> 30 media, thus you can expect to
> see about 2 fails per backup pass (and your back software had better
> handle this gracefully).  Given that CD-R media can also fail on the
> shelf (it is not an archival process)  you will need to make at least
> two interchangeable sets of disks if you want some chance of ever
> reading back those data sets at a later time.   At 5 minutes per disk,
> including verify and labelling and replacement of fails, that sounds
> like a whole day of work.

The other big issue in doing backups to cd-r are the problems of
preserving filenames and permissions.

If you have filenames that are longer than about 131 characters, they
will get truncated.  

If you have filenames that contain non-english character sets that have
not been translated to utf-8, they will not always go on the disc.

If you have file attributes beyond the basic permissions, then they will
probably forgotten.  (Such as extended attributes or posix security
permissions like what SELinux uses.)

I am not certain how much of the above gets preserved by tar or cpio. 
(Especially extended permission bits and odd acls.)

I once thought dvd backups were all I needed.  Now I have my doubts.

-- 
We are living in the "interesting times" the fortune cookies warned us
about.