[SAC] Server Planning - Backups

[Martin Spott]

On Fri, Feb 26, 2010 at 05:13:58PM -0800, Alex Mandel wrote:
> On 02/26/2010 03:40 PM, Martin Spott wrote:
> > On Fri, Feb 26, 2010 at 12:23:27PM -0800, Alex Mandel wrote:

> >> 1. We could mount the BackupVM in each of the virtual machines so that
> >> when a cron script is run that dumps databases or hot copies sites it
> >> writes to the backup server.
> > 
> > Do we have access to a regular "backup server", like in 'professional'
> > IT, as a distinct machine equipped with some sort of mass storage
> > device ?
> > 
> I don't believe so, but can check with OSL and what they have to offer.

Ah, ok. I was just asking because the term "backup server" the way I'm
used to it typically relates to the above mentioned sort of device or
hardware and I didn't spot such thing in any of our plans.

> This was part of the decision to go with a RAID 6 on the machine that
> will have the backupVM (2 disks can die without data loss) and the
> proposal to rsync/backup this virtual machine to a Telescience blade at
> least weekly. Do we really need tape backup if we have good redundancy?

Most of the cases I've seen (experienced ....) in professional
environments which required a restore from backup were _not_ caused by
a disk failure  :-)
Actually the biggest data losses I had to deal with during the past
five years or so had been caused by a) well, 'unexpected' behaviour of
the RAID controller (after performing a power cycle the controller
didn't accept one of the disks of a degraded array) and b) someone
trying to 'e2fsck' an ext3 filesystem after a system crash (not my
install, I never use ext3 for critical setups).

Certainly, these are rare cases, but they _do_ happen occasionally. So,
at least to my firm opinion it's always a matter of cost-vs-belief
wether to use a distinct, separate backup storage or not.

> That's good to see someone already has some experience with one of these
> tools. Do you have any particular hints, tips, gotchas etc - I notice
> that bacula uses a database backend to track what it's done, do you have
> a recommendation for which db to use?

SQlite(3) with Bacula scales surprisingly !! well up to writing over
1.5 million files (approx. 400 GByte) a day to tape backup and dumping
the resulting backup catalogue (the dump is approx. 1 GByte large,
including history) takes just a couple of seconds. Therefore I'd expect
it to scale far beyond this point but I don't have any experience with
using SQLite on bigger setups.
Therefore I'd tend to use a PostgreSQL database as a backend if we were
to have our own Bacula instance.

Cheers,
	Martin.
-- 
 Unix _IS_ user friendly - it's just selective about who its friends are !
--------------------------------------------------------------------------