Being one of the maintainers of Amanda (www.amanda.org), I'd always been of the opinion that tape backups were the only way to do backups seriously.
The recent explosion in disk capacities and decrease in prices got me to rethink this, just when it came the time for me to set up a home office. When I compared the cost of a reasonably-good tape drive and a number of tapes large enough for me to get at least a month of backups in rotation, and computed how many 60GB disks I could buy with that money, the solution was clear.
I ended up setting up 3 machines with 4x60GB each. They're all on RAID 5, such that if any single disk fails, the machine keeps running (actually, I have/boot on RAID 1 over the 4 disks and / on RAID 1 over 2 of the disks and an alternate root to test upgrades over the other 2, but you get the point). This got me blazingly fast disk access, that tapes would never help me get:-)
I get all my backup-worthy data rsynced over to the other machines daily or so. I plan to start playing with Inter-Mezzo soon, so that I don't have to remember to run these backups, and so that I don't run these backups on the wrong direction.
But that's not all. With the mind-boggling amount of disk space I could afford, I could (actually, I will, but you get the idea) set up Amanda to backup interesting portions of my home directory to disk, and also replicate this to at least another of my local machines. Such backups can use software compression, such that they don't take as much space as live data. Also, I intend to use another form of compression: instead of backing up CVS trees (I've got loads of check outs), I'm going to back up only local changes to files, so that, in case of disaster, I can still download the original CVS tree and re-apply patches. But this is still a plan, not something I've got running.
Finally, I've got yet another disk on a remote site, to which I rsync not only the interesting portions of my data, but also my backups. I could convince someone else to run this remote backup site for me by offering this person the speed up of RAID 0 over two disks (one of those mine). As for keeping the secrecy of the data on this remote backup site, I'd just get the backup files encrypted, no big deal.
I can strongly recommend this solution: I got pretty much as much data safety as could be expected from a tape-based backup, without any of the hassle of having to switch tapes and moving them off-site and back on-site, and with the bonus of very fast access to local data, unlikely donw-time and fast recovery except in case of total disaster (i.e., having all of my local machines failing, in which case I'd have to either download my backups from the remote site over the net or, more likely, take a replacement machine over to the remote backup site and copy files over a fast local network connection, or from disk to disk.
As for getting 4 IDE disks into a single machine, don't even think of using only the 2 IDE controllers that come on most motherboards these days (for RAID set-ups, you really want one IDE disk per controller). There are a few good motherboards that come with 4 IDE controllers, so that you can even have a CD-ROM and/or a CD-RW in addition to the 4 disks. If you can't find such a motherboard that suits your needs, you can always get one of those PCI cards that adds 2 IDE controllers to your machine.
As for the problem of fitting so many disks in a standard ATX chassis, it can be done. Cooling may be a problem, but a good cooler has been good enough.
All in all, I'm very happy with this arrangement. It was not cheap, but it was not as expensive as a tape-based solution, and it's far more flexible, way faster and it doesn't require any baby-sitting after you get it going. And I can keep far more backup history than I thought it was going to be possible.
I wrote about this stuff a while ago. See `The Competitive advantages of Free Software', on my home page. And then, perhaps I wasn't original myself, but I certainly hadn't read about it before.
I've got direct access to 15 Deskstar 60GXP 60GB drives, 2 of them since May, 12 more since August and 1 more since last month, and none of them have failed, made odd noises, etc. The only time I had a bad surprise was when I only had the oldest 2 disks, and the machine on which they were installed starting making horrible noises. I thought it was something wrong with the disks, but it turned out that the second flat cable I had installed myself was touching the processor fan. Oops:-)
> What we need now is to create an environment, where users can easily create customized tools for the way they work, and developers can easily add new functionality.
Slashdot Mirror
Being one of the maintainers of Amanda (www.amanda.org), I'd always been of the opinion that tape backups were the only way to do backups seriously.
/boot on RAID 1 over the 4 disks and / on RAID 1 over 2 of the disks and an alternate root to test upgrades over the other 2, but you get the point). This got me blazingly fast disk access, that tapes would never help me get :-)
The recent explosion in disk capacities and decrease in prices got me to rethink this, just when it came the time for me to set up a home office. When I compared the cost of a reasonably-good tape drive and a number of tapes large enough for me to get at least a month of backups in rotation, and computed how many 60GB disks I could buy with that money, the solution was clear.
I ended up setting up 3 machines with 4x60GB each. They're all on RAID 5, such that if any single disk fails, the machine keeps running (actually, I have
I get all my backup-worthy data rsynced over to the other machines daily or so. I plan to start playing with Inter-Mezzo soon, so that I don't have to remember to run these backups, and so that I don't run these backups on the wrong direction.
But that's not all. With the mind-boggling amount of disk space I could afford, I could (actually, I will, but you get the idea) set up Amanda to backup interesting portions of my home directory to disk, and also replicate this to at least another of my local machines. Such backups can use software compression, such that they don't take as much space as live data. Also, I intend to use another form of compression: instead of backing up CVS trees (I've got loads of check outs), I'm going to back up only local changes to files, so that, in case of disaster, I can still download the original CVS tree and re-apply patches. But this is still a plan, not something I've got running.
Finally, I've got yet another disk on a remote site, to which I rsync not only the interesting portions of my data, but also my backups. I could convince someone else to run this remote backup site for me by offering this person the speed up of RAID 0 over two disks (one of those mine). As for keeping the secrecy of the data on this remote backup site, I'd just get the backup files encrypted, no big deal.
I can strongly recommend this solution: I got pretty much as much data safety as could be expected from a tape-based backup, without any of the hassle of having to switch tapes and moving them off-site and back on-site, and with the bonus of very fast access to local data, unlikely donw-time and fast recovery except in case of total disaster (i.e., having all of my local machines failing, in which case I'd have to either download my backups from the remote site over the net or, more likely, take a replacement machine over to the remote backup site and copy files over a fast local network connection, or from disk to disk.
As for getting 4 IDE disks into a single machine, don't even think of using only the 2 IDE controllers that come on most motherboards these days (for RAID set-ups, you really want one IDE disk per controller). There are a few good motherboards that come with 4 IDE controllers, so that you can even have a CD-ROM and/or a CD-RW in addition to the 4 disks. If you can't find such a motherboard that suits your needs, you can always get one of those PCI cards that adds 2 IDE controllers to your machine.
As for the problem of fitting so many disks in a standard ATX chassis, it can be done. Cooling may be a problem, but a good cooler has been good enough.
All in all, I'm very happy with this arrangement. It was not cheap, but it was not as expensive as a tape-based solution, and it's far more flexible, way faster and it doesn't require any baby-sitting after you get it going. And I can keep far more backup history than I thought it was going to be possible.
I wrote about this stuff a while ago. See `The Competitive advantages of Free Software', on my home page. And then, perhaps I wasn't original myself, but I certainly hadn't read about it before.
I've got direct access to 15 Deskstar 60GXP 60GB drives, 2 of them since May, 12 more since August and 1 more since last month, and none of them have failed, made odd noises, etc. The only time I had a bad surprise was when I only had the oldest 2 disks, and the machine on which they were installed starting making horrible noises. I thought it was something wrong with the disks, but it turned out that the second flat cable I had installed myself was touching the processor fan. Oops :-)
> What we need now is to create an environment, where users can easily create customized tools for the way they work, and developers can easily add new functionality.
Err... Haven't we had Emacs for a while already?