Differential backup recommendation?

[Ed Wilts]

On Wed, Jun 09, 2004 at 09:02:39AM -0400, Reuben D. Budiardja wrote:
> 
> I am trying to come up with a modest, cheap, and simple backup system
> for some of our servers & workstations here. This will be basically a
> kind of home-grown thing, rather than spend the money for commercial
> backup system.
> 
> We have one server that's going to have a rather big storage space for
> this purpose (about 400 GB) with RAID. So here's my plan: For every
> remote server / workstation (or a directory in the remote machine, eg
> /home), I want to create a :
> 
> 1. Full backup on day 1 to the backup server.
> 
> 2. Write the full backup to media (either CD-R or CD-RW or DVD-RW,
> haven't really decided yet), and leave the data in the drives in the
> backup server

mondo might be good for this.

> 3. Differential backup on day 2 to day 7
> 
> 4. Remove lask week backup from Harddrive, then Goto 1.
> 
> For step 1 and 2, I can use something like rsync, or rsnapshot, or
> even mondo-rescue which will nicely create the ISO files for me
> directly for the remote machines that I backup. So the first two steps
> is not terribly hard.
> 
> Now for step 3, I don't know what to do, and can use some
> recommendation. Is there an open source utility out there that will do
> this differential backup? 

I use rsnapshot for it all at home.  With very limited disk space, I can
create backups of 2 systems and keep monthlies, weeklies, dailies, and
hourlies.  Every copy past the original is just a link so the
incrementals don't really take up any space and everything looks like a
full backup.  If I want to restore /home/ewilts, I just grab the backup
from someplace like /snapshots/hourly.3/server/home/ewilts and it looks
like it's complete.  My "hourlies" are actually done every 2 hours
during the day.

> I am thinking about storing this differential backup in different directory 
> everyday, like day0/, day1/, day2/ etc. I've googled, and only come up with:  
> http://rdiff-backup.stanford.edu/
> I'll try that, but I'm wondering if people here knows some other, or can 
> recommend some other. 

rsnapshot will keep hourly.0, hourly.1, etc. with hourly.0 always being
the most recent copy.  Ditto for daily.0, daily.1, etc.

> The advantage of doing this is so that I can have one week versions of
> changes on the machine, and can go back to specific day during that
> week to restory. 

You can keep as many copies as you want with rsnapshot and they don't
take much space at all if the data isn't too dynamic.

My total snapshots (20 "copies") for 3 months take 13G.  My hourly.0
snapshots are 6.4GB so clearly I'm saving a *lot* of space doing my
backups this way.

> I know the disadvantage is to restore could be more difficult, and after a 
> week, data cannot be recovered to specific day, but we're willing to live 
> with that (what I have now is even worse because I only have mirror for 
> backup, with incremental backup every day). So  our requirement is pretty 
> simple.

You tell rsnapshot how long to keep each daily so if you wanted a month
of dailies, that's easy to set up.

The real reason for differential incrementals is to keep your backup
space utilization down and speed the backups down.  rsnapshot addresses
both of these issues with rsync.

-- 
Ed Wilts, RHCE
Mounds View, MN, USA
mailto:ewilts at ewilts.org
Member #1, Red Hat Community Ambassador Program