emv@math.lsa.umich.edu (Edward Vielmetti) (12/21/89)
Hi. I'm trying to arrange it so that I can back up an entire network of Ultrix systems onto one Exabyte tape overnight, with no operator intervention. Slip in a tape at 5 pm, take it out at 9 am, no worry, no muss, no fuss. It would appear that the following problems need to be addressed before this is going to go, namely: - dumping live filesystems is a problem. Fixed with the "Purdue-dump-mods" reposted to alt.sources not too long ago. If a file changes between passes, you don't dump it; some files will be missing from the resulting dump tapes but the tape will be readable. - sequencing the dumps so that one doesn't start until the previous is finished. It seems like it would be best to have the system with the tape drive control the whole operation. The 'rrdump' program that just showed up recently would seem to handle that fairly well. - convincing dump to put more than one dump on a dump tape. This I don't have an answer to; any clues? From all I can tell the first thing it does is to rewind the tape. I think I've heard of an 'mdump' but I don't have any pointers. Obviously the best situation would be a mdump program that could induce other systems start up dump in order, i.e. rmdump 9uf /dev/exabyte elvis:/dev/rz1g orpheus:/dev/rz1g .... but that's too much to ask. Free or for-pay solutions welcome, in theory at least I have access to sources. --Ed
emv@math.lsa.umich.edu (Edward Vielmetti) (12/28/89)
Thanks to everyone who sent me mail with information and shell scripts about how to dump multiple filesystems to one Exabyte tape overnight. I have in my hand a tape which contains the whole local network, about a 4.5 hour job, neatly done while I was out. In regards to the questions I posed, here are the reasonable answers that came about. - dumping live filesystems. If your dump has been trained not to dump files that have changed between passes, you're OK; this is one of the fixes that Purdue did. Rich Kulawiec <rsk@boulder.colorado.edu>, a partial author of some of the Purdue changes, says the supplied Ultrix dump works OK. - sequencing the dumps in order. The obvious solution, one script running from the dump host and generating commands like 'rsh foo -n rdump 0ouf dumphost:/dev/tape /filesystem' is what I have now. I don't like keeping a huge /.rhosts file on the tape host, but a script sent in suggested copying in a jumbo /.rhosts file just for the duration of the dumps and then moving the real one back. - convincing dump to put more than one dump on a tape. Duh to me on this one: use the non-rewinding entry to the device, e.g. /dev/nrmt1h instead of /dev/rmt1h. The dumps go one after the other on the tape just as you expect. At least on Ultrix, there's an 's' parameter to restore which lets you decide which of the files on the tape to restore from (I see it in SunOS 4.x too), so that should be no problem to restore from. Thanks to all who sent mail. I have a solution which worked at least once, and it wasn't as hard or necessarily as complicated as I thought it might be. As one person said "it is not terrible hard". In among the stuff that I haven't tried were the 'rrdump' thing that I mentioned which removes some of the overhead of 'rsh', and a nice perl script which parses a 'backup.conf' file and acts accordingly. Thanks again, --Ed