While we are still working out the bugs in an unexpected change of management (the original author got promoted by the Guy Upstairs), folks looking for backup solutions should give serious thought to rsync-based disk-to-disk backups. I am acting manager for one package (see http://www.dirvish.org, and http://www.keithl.com/linuxbackup.html ), and there are others. rsync/dirvish is fast, inexpensive, and restoring single files or subsystems is random access and dead easy. With some advanced preparation, I can restore an exact image of my server hard drive with a reboot and about a minute of typing, followed by the time required to copy a disk image - 2 hours for 80gb in my case. I can very easily combine restores, using (for example) a two-week-old base, then upgrades, then yesterday's user files, to recover from a system crack. All from one script, off the same backup drive. Backup media cost about $0.50/GB US, including swap cages for the drives; this wildly exaggerates the operating cost, though. Using reiserfs and Unix hard links on the target drive, I can get hundreds of full nightly images into a 2X sized drive, so I can store 100X more images per dollar using rsync. While I can expire older images, I do not bother. When a triplet of backup disks (I do a 3 way rotation) finally fill up, I just retire them to offsite storage. As a consultant, I am required by some contracts to destroy my copies of client data after the end of a contract. It is easy enough to just scribble over the particular client's data on a hard drive, leaving the rest intact, or else erase the files and all their links. I can't do that with tapes. The entire process for my systems (5 machines, including one off-site) takes about an hour. It really thrashes the network, but rsync is checking every byte on every machine, and moving all the bytes that change, every night. The biggest problem is to remember to rotate the drives daily into the fire-resistant safe, away from power supply failures and prying script kiddies. When I forget, I do not lose data, but my survivability suffers. I have an Exabyt tape jukebox sitting next to the server, and I suppose I could be making occasional full image copies to that, but it is so time consuming, and the tapes are so unreliable and slow to verify, that it really doesn't seem worth the effort. I have done tapes for 20 years, and the extreme pain of pulling files back off a tape is never welcome. Single file restore from tape can take hours, depending on how much effort you put into file-level cataloging. The longest it takes with the hard-drive backup is about 5 minutes, assuming I have the right backup drive on-site. Of course, the drives take up a small fraction of the space, too. Hard drives are a little more fragile, but a dropped tape is easily damaged, too. I have padded boxes for drive transport. My next improvement will probably be to run a second ethernet over to a neighboring building, and blow a copy of the nightly backups over there, too. That will take my Mean Time To Catastrophic Failure, currently dominated by a total gutting fire or room-emptying theft, down from 1/50y to 1/200y. I've mentioned some of this stuff before, and I repeat here occasionally until more folks use rsync. This is selfish; when more of you use D2D backups, more of you will be pushing for datex logrotate, more robust reiserfs, and faster drive-to-drive copy tools. Again, check the sites above. Tapes are obsolete. Keith -- Keith Lofstrom keithl@xxxxxxxxxx Voice (503)-520-1993 KLIC --- Keith Lofstrom Integrated Circuits --- "Your Ideas in Silicon" Design Contracting in Bipolar and CMOS - Analog, Digital, and Scan ICs
