George Georgalis wrote: > What are people doing for off site backup? I know there is a lot > of options, my only requirement is to mitigate risk, which is > kinda vague. There is about 60Gb which may double in a year. It > would be desirable to have on line access to the older data. > Typically I do frequent rsync with hardlinks into dated folders on > HD, so local snapshots are available at the site.
This is always one of my favorite topics. I am always concerned about losing data. I have all of my email going back to 1996 and have not lost any significant amount of data since then! And nowadays since I have 1.5G of digital photos which are iireplaceable I am especially careful. "If it wasn't backed up, it wasn't important!" It's funny...my parents and grandparents have these big 3-ring binders full of fading pictures which could only be viewed if you were physically in their living room susceptable to fire and flood and not easily backed up...I have a website full of pics that will never fade and are viewable from anywhere in the world with a net connection and although any one instance of it is easily deleted it is trivially easy to ensure that half a dozen copies exist in various parts of the world. I used to have a DDS-4 tape drive but I never really quite trusted it. I've never been a big fan of tape in general, especially the consumer/small business tape stuff I can easily afford. Plus tape is just so incredibly slow. For the last year and a half I have been backing up to DVD. I can get 4.7G native per disk. I have around 20G of data that I back up after compression so that needs 6 disks. Quality Memorex DVD+RW disks (I first bought generic DVD+RW disks that gave me all kinds of trouble so now it is only name-brand. I'm not going to sacrifice the reliability of my backups to save a few bucks) cost $49.98 for a 50 pack so right around $1/disk so each backup requires $6 in media and the media is theoretically reusable up to 1000 times. I just recently restored something from a backup over a year old so it seems to be working well for me. Only problem is it takes a number of hours and 6 disk changes to do a full backup this way. The actual disk burning time is only 17 minutes per disk or 102 minutes total burn time. I only do a full once a month though so it's not too bad. But it takes 5 times longer than this to do the backup. I think it is because of hard disk contention since it has to read the data from one part of the disk then build an ISO image in another part. I bet if I had another spindle to write the ISO onto it would be a lot faster. The incremental is no problem time-wise though since not much changes and all fits on one disk. I just ordered a pair of 80G USB2/Firewire drives to augment my DVD backup. Should be a lot faster. I think I will keep a set of DVD's off-site and use the hard drives for my day to day backups and one-off archives. Between the drives and the 50 DVD+RW's I have 395G of backup capacity so I think I am pretty well set. As of this week I am using bacula to do my backups. I used to use mondo. Bacula seems more popular, more flexible, and I like that it can do network backups. But it is complicated software and it has a number of rough edges. It took me probably 3 full working days (24 hours) to read the manual, install it, debug a number of problems, and get all of the kinks worked out of it. At least I hope I have all of the kinks worked out. The command line interface is pretty sucky. I accidentally purged all of my backups from the database once. You have to be very careful with it. I am using SQLite3 as my database. I'm a big fan of MySQL but SQLite3 is just so much easier to set up. If my whole machine gets toasted I don't want to have to configure mysql before I can restore my backups. Aside from user-interface rough edges it definitely has a lot of the features of an enterprise backup system. > This data is highly private, so that must be part of the formula. > I was thinking of a system similar to the local backups, doing a > pull from a colo machine; and at the same time making cd/dvd or > tapes at the main site which would go into a (bank) safe deposit > box or some other facility. I have piped backup images though gpg with a symmetric encryption key which worked wonderfully. You can even automate this by keeping the key in a file on the server being backed up (if they can access the key they can access your private data anyhow) and tell gpg to read the key in from the file and encrypt before it goes off site. I was piping a tar -zcvf through gpg then through ssh into a file on disk storage on the other end. I don't think bacula does encryption yet but it has hooks to link to scripts which I bet you could get to do the encryption for you. > My tendency is to think an off site raid mirror functioning as a > copy of the local raid mirror snapshots is pretty comprehensive. > It is difficult for me to focus on the benefits of dated tape, hd, > or cd/dvd in a storage facility. What is the real benefit of that? > Very quick disaster recovery is not necessary, but 3 days would > defiantly be better than a week to recover. How do I find the > sweet spot of medium, interval and rotation? Is off line medium > really necessary here? I think off-site is definitely necessary. That may imply off-line or it may just imply some disk in another location with broadband internet connecting them. I would avoid tape unless you want to spend thousands and thousands on DLT or LTO or something. DVD is working great for me. I think the USB2/Firewire HD's will work great also and I think this might be my preferred way to go in the future. DVD media is pretty cheap at around 20 cents/G but the HD's are much faster (faster throughput and no disk changing) although they cost $1/G. The HD's are also probably more reliable since optical media has to be taken care of and kept clean (I always keep my disks covered) and I have to worry about the DVD drive working properly whereas with the HD I just plug it into any computer with USB or Firewire, no special reader hardware necessary. I would definitely do an incremental every night. Maybe a full backup once a week or once a month depending. Definitely send a copy off-site once a month. Andy's suggestion of hot swap drive bays and a mirror, half of which gets rotated regularly seems like a good plan also. I would do that myself but I don't really want to invest in the hot swap chassis and all of the drives etc. Actually, that is why I started looking into USB/Firewire drives. They are inherently hot swap themselves without a fancy hot swap chassis but probably not really fast enough to be pared with the internal SATA disk as a mirror. -- Tracy R Reed http://copilotconsulting.com -- [email protected] http://www.kernel-panic.org/cgi-bin/mailman/listinfo/kplug-list
