Il 18/03/2011 00:57, Phil Stracchino ha scritto: > On 03/17/11 18:46, Marcello Romani wrote: >> Il 16/03/2011 18:38, Phil Stracchino ha scritto: >>> On 03/16/11 13:08, Mike Hobbs wrote: >>>> Hello, I'm currently testing bacula v5.0.3 and so far so good. One >>>> of my issues though, I have a 16 bay Promise Technologies VessJBOD. How >>>> do I get bacula to use all the disks for writing volumes to? >>>> >>>> I guess the way I envision it working would be, 50gb volumes would be >>>> used and when disk1 fills up, bacula switches over to disk2 and starts >>>> writing out volumes until that disk is filled, then on to disk3, etc.. >>>> eventually coming back around and recycling the volumes on disk 1. >>>> >>>> I'm not sure the above scenario is the best way to go about this, I've >>>> read that some people create a "pool" for each drive. What is the most >>>> common practice when setting up a JBOD unit with bacula? Any >>>> suggestions or advice would be appropriated. >>> >>> That scheme sounds like a bad and overly complex idea, honestly. >>> Depending on your data load, I'd use software RAID to make them into a >>> single RAID5 or RAID10 volume. RAID10 would be faster and, if set up >>> correctly[1], more redundant; RAID5 is more space-efficient, but slower. >>> >>> >>> [1] There's a right and a wrong way to set up RAID10. The wrong way is >>> to set up two five-disk stripes, then mirror them; lose one disk from >>> each stripe, and you're dead in the water. The right way is to set up >>> five mirrored pairs, then stripe the pairs; this will survive multiple >>> disk failures as long as you don't lose both disks of any single pair. >>> >>> >> >> Hi Phil, >> that last sentence sounds a little scary to me: "this will survive >> multiple disk failures *as long as you don't lose both disks of any >> single pair*". >> Isn't RAID6 a safer bet ? > > That depends. > > With RAID6, you can survive any one or two disk failures, in degraded > mode. You'll have a larger working set than RAID10, but performance > will be slower because of the overhead of parity calculations. A third > failure will bring the array down and you will lose the data. > > With RAID10 with sixteen drives, you can survive any one drive failure > with minimal performance degradation. There is a 1 in 15 chance that a > second failure will be the other drive of that pair, and bring the array > down. If not, then there is a 1 in 7 chance that a third drive failure > will be on the same pair as one of the two drives already failed. If > not, the array will still continue to operate, with some read > performance degradation, and there is now a just less than 1 in 4 chance > (3/13) that if a fourth drive fails, it will be on the same pair as one > of the three already failed. ... And so on. There is a cumulative 39% > chance that four random failures will fail the entire array, which rises > to 59% with five failures, and 78% with six. (91% at seven, 98% at > eight, and no matter how many leprechauns live in your back yard, at > nine failures you're screwed of course. It's like the joke about the > two men in the airliner.) > > But if the array was RAID6, it already went down for the count when the > third drive failed. > > > > Now, granted, multiple failures like that are rare. But ... I had a > cascade failure of three drives out of a twelve-drive RAIDZ2 array > between 4am and 8am one morning. Each drive that failed pushed the load > on the remaining drives higher, and after a couple of hours of that, the > next weakest drive failed, which pushed the load still higher. And when > the third drive failed, the entire array went down. It can happen. > > But ... I'm running RAIDZ3 right now, and as soon as I can replace the > rest of the drives with new drives, I'll be going back to RAIDZ2. > Because RAIDZ3 is a bit too much of a performance hit on my server, and > - with drives that aren't dying of old age - RAIDZ2 is redundant > *enough* for me. There is no data on the array that is crucial *AND* > irreplaceable *AND* not also stored somewhere else. > > What it comes down to is, you have to decide for yourself what your > priorities are - redundancy, performance, space efficiency - and how > much of each you're willing to give up to get as much as you want of the > others. > >
Phil, that was an interesting read. Thanks for your detailed response. (Your last paragraph is of course the definitive word on the subject.) Now that I think about it, I realize I didn't fully take into account the high number of drives we're talking about. Probably if using RAID6 a spare drive is to be considered. Or, better yet, a mirror machine... But then we're back to "it depends", I guess :-) Oh, and BTW, maybe it's time for me to move past these old limited raid levels and investigate ZFS and those intriguing RAIDZx arrays... Marcello ------------------------------------------------------------------------------ Colocation vs. Managed Hosting A question and answer guide to determining the best fit for your organization - today and in the future. http://p.sf.net/sfu/internap-sfd2d _______________________________________________ Bacula-users mailing list Bacula-users@lists.sourceforge.net https://lists.sourceforge.net/lists/listinfo/bacula-users