On Mon, Mar 12, 2012 at 2:27 PM, Steve Lindemann steve@marmot.org wrote:
I believe in tape... it's just not a viable option with the large disk sizes we have today unless you have a lot of money for a fast, multi-drive solution. I can backup a bit over 500GB daily in 3 hours to external disk. Using a single tape drive that would (and did) take far too long.
I think both amanda and backula would use disk as temporary holding space. Amanda will compute the best mix of incrementals and fulls to fit on the tape and run the backups to disk in parallel, starting the tape write when it has one complete file.
The big difference with disks is that I just do full backups each time. In our situation there is time for that and it saves a lot of grief when trying to restore something in particular. None of this running back thru the incrementals to get at what you want. Of course, with incremental backups the typical daily time would be much much shorter.
Backuppc will let you do normal fulls and incremental backups, but will transparently merge them when doing a restore.
For backing up multiple production servers I have a backup server and a private GB network to each system. Each server runs a backup script at night (via cron) to backup to the backup server. Then we backup the backup server to the external disk during the day. At least one external disk is off site at any given time.
The one down side of backuppc is that due to the extensive use of hardlinks to de-dup the content, it can be impractical to use normal approaches to copy a large archive for offsite rotation. Some people rotate the whole thing, letting the next incremental run catch up with the differences, some just run another instance over the network from a different location (often practical with rsync as the transfer method), and some use an image-copy scheme to copy the whole filesystem or device.
I'm aware of the fancy tools to do the job for you but I like the simplicity of our home grown solution. And the only thing I absolutely need for a restore is tar. No databases, no extra applications, just tar. The catch is that I'm not sure it would scale up to a huge number of servers gracefully.
Backuppc has a command line tool that will generate a tar image out of its storage format, and a way to do that from the web interface. You can make them periodically for archive copies that could be re-installed without backuppc. But, backuppc itself is just a perl script using files in a linux file system (with some compression/linking tricks to hold about 10x what you would expect).