>>>>> On Wed, 24 Apr 2024 23:40:31 +1000, Gary R Schmidt said: > > On 24/04/2024 22:33, Gary R. Schmidt wrote: > > On 24/04/2024 21:30, Roberto Greiner wrote: > >> > >> Em 24/04/2024 04:30, Radosław Korzeniewski escreveu: > >>> Hello, > >>> > >>> wt., 23 kwi 2024 o 13:33 Roberto Greiner <mrgrei...@gmail.com> > >>> napisał(a): > >>> > >>> > >>> Em 23/04/2024 04:34, Radosław Korzeniewski escreveu: > >>>> Hello, > >>>> > >>>> śr., 17 kwi 2024 o 14:01 Roberto Greiner <mrgrei...@gmail.com> > >>>> napisał(a): > >>>> > >>>> > >>>> The error is at the end of the page, where it says that you > >>>> can see how > >>>> much space is being used using 'df -h', but the problem is > >>>> that df can't > >>>> actually see the space gain from dedup, it shows how much > >>>> would be used > >>>> without dedup. > >>>> > >>>> > >>>> This command (df -h) shows how much allocated and free space is > >>>> available on the filesystem. So when you have a dedup ratio 20:1, > >>>> and you wrote 20TB, then your df command shows 1TB allocated. > >>> > >>> But that is the exact problem I had. df did NOT show 1TB > >>> allocated. It indicated 20TB allocated (yes, in ZFS). > >>> > >>> I have not used ZFS Dedup for a long time (I'm a ZFS user from the > >>> first beta in Solaris), so I'm curious - if your zpool is 2TB in size > >>> and you have a 20:1 dedup ratio with 20TB saved and 1TB allocated > >>> then what df shows for you? > >>> Something like this? > >>> Size: 2TB > >>> Used: 20TB > >>> Avail: 1TB > >>> Use%: 2000% > >>> > >> No, the values are quite different. I wrote 20tb to stay with the > >> example previously given. My actual numbers are: > >> > >> df: 2,9TB used > >> zpool list: 862GB used, 3.4x dedup level. > >> Actual partition size: 7.2TB > >> > > You use zpool list to examine filespace. > > Or zfs list.
On FreeBSD at least, zfs list will show the same as df (i.e. will include all copies of the deduplicated data in the USED column). I think the reason is that deduplication is done at the pool level, so there is no single definition of which dataset owns each deduplicated block. As a result, the duplicates have to be counted multiple times. This is different from a cloned dataset, where the original dataset owns any blocks that are shared. __Martin _______________________________________________ Bacula-users mailing list Bacula-users@lists.sourceforge.net https://lists.sourceforge.net/lists/listinfo/bacula-users
