Re: Amanda backup strategy thoughts, full and incremental backups to tape once a week.
2004-04-06 15:14:19
Here's a strategy that I implemented about a month ago that is working
pretty well so far:
1. run amdump every night to large RAID w/o tape, mix of full and incr
2. run script to separate images to amanda-incr and amanda-full
3. when amanda-full exceeds size of tape, run amflush
4. when RAID approaches capacity, flush some incrs to tape
Since it's a RAID, there is less chance it will go to crap than a
regular disk partition, but it is still possible. And if it does, I
still have the full dumps on tape to fall back on.
To give you some gauge of capacities, I'm backing up about 300 GB of
disks, have ~600 GB RAID5 (on a Snap 4500) and am using 100/200 GB LTO
(Ultrium-1) tapes. So far, it's mostly web servers, but I haven't moved
everything to the new backup cycle yet. I have flushed 2 tapes worth of
full dumps, and accumulated about 167 GB of incrementals that are still
on the RAID.
The goal and idea is that > 90% of the time, I will only need to load
one tape to do a restore, and the rest will come out of holding files.
Even better, there is a good chance that the tape that I need will be
the last one that I flushed, so I won't have to change the tape.
/snap/amanda-hd is the directory that amanda.conf looks for, which is
actually a symbolic link to /snap/amanda-full. Therefore, I just move
the incremental dumps from amanda-full to amanda-incr, both actually on
the same filesystem (a simple directory rename in the filesystem inode
table). If I need to flush incrementals, I just change the link
amanda-hd -> amanda-incr and run the flush (this could be done more
elegantly with a patch, but this is so easy, so why bother?) In the
cron job, I make sure amanda-hd -> amanda-full before amdump runs.
I'm using NFSv3 to XFS filesystem through a dedicated 100baseTX FD
connection, so I'm using a chunksize of ~600 GB, and that makes the
rename script very simple since I don't have to worry about any backup
getting split into chunks. AFAIK NFSv3 and XFS both support 64-bit
addressing, so I shouldn't run into any filesize limits.
--jonathan
|
|
|