Mr. Troxel,
The system we have in place here brings the offsite storage people to our site once per week. So, every week, the previous week's tapes are sent offsite for six weeks, and the oldest set in the rotation (a week's worth of tapes that was cut 7 weeks before) is brought back. This way, we always have 6 weeks of tapes offsite. So, effectively, there are 6 dumpcycle's worth of tapes offsite at any given time. Usually, I am pretty sure that I have a level 0 of every machine offsite, somewhere in those 6 weeks worth of tapes. This of course is using our old backup software. I am at the point now with amanda that I am assured that the system is reliable enough to run with, and I want to expand my test system and migrate away from the old software. My biggest question here was HOW I should expand my tape list. I currently have 10 tapes in the amanda cycle. Should I expand the tape-list such that I have one set of tapes (ie. Tape00 through Tape99) or should I make several sets of 1 week dumpcycle tapes (several sets of Tape00 - Tape9)? I am thinking that I should go with the former idea, because if I were to make several 10-tape sets and use those in my rotations, I would run into a problem where it would be difficult to find which tape has the image of any given backup that I might need (this problem arising from the amanda scheduling - one can never be sure which night of the week that the level 0 of any given partition was done on!). My other main question was concerning our yearly backups. We send the last week's backups of every month offsite for 1 year. This is where it gets hairy, because a failure during that last week would mean that my long term backup storage would have a 2 month gap between backups of the failed partition. The other problem with that is, how would the tape scheduler work with having tapes from the middle of the tapecycle offsite that long? Would amanda scream if Tape12 - Tape 24 is offsite for a year, and I try to use Tape25 when it is expecting Tape12 in the tapecycle? The only way I can think of to avoid those issues would be to make a second backup run (config = daily2), in which a level 0 is forced for every machine, and which I give a dumpcycle of say 2 days. That way I *should* be assured of getting at least one level 0 dump of every machine during the last week of every month. I'd probably have to run such a backup config over the weekends so as not to insense my users over slow network performance, but its probably do-able. Any comments suggestions on this idea (that is, am I totally missing something?) :-) Thanks again! Brandon Moro Systems Administration, Unify Corporation [EMAIL PROTECTED] -----Original Message----- From: Greg Troxel [mailto:[EMAIL PROTECTED]] Sent: 2002?5?6? 5:07 To: Brandon Moro Cc: [EMAIL PROTECTED] Subject: Re: Expanding tapecycle Be careful - this is more complicated than it seems. If Amanda always put exactly one full of each filesystem on every week, and nothing were ever delayed (because no nightly dumps were ever missed due to tape problems, and no clients were ever down), then this might work. You might want to consider taking 2 dumpcyle's worth of tapes offsite, or write some code to see if your candidate 5 tapes actually do have a full dump of everything. So it might make sense to have a regular rotation and then 6 sets of offsite tapes, with different sequence numbers, and interpose the offsite tapes during 'offsite snapshot weeks'. Another approach is to run separate offsite dumps that skip incrementals. Greg Troxel <[EMAIL PROTECTED]> --------------------- >From meanness first this Portsmouth Yankey rose, And still to meanness all his conduct flows. --Oppression, A poem by an American (Boston, 1765).
