ZFS Space Inflation

[LnxBil]

Exactly! I wasn't aware of that order of magnitude.

You really have to check the ashift and the corresponding recordsize setting before send/receive. I run into serious space problems with a recordsize of 4K and synching that from ashift 8 to ashift 12. I needed almost twice the space due to the space waste of compressed blocks.

root@proxmox4 ~ > zpool list
NAME                 SIZE  ALLOC   FREE  EXPANDSZ   FRAG    CAP  DEDUP  HEALTH  ALTROOT
local-zfs           2,48T   260G  2,23T         -     9%    10%  1.00x  ONLINE  -
proxmox4-sicherung   508G   441G  67,0G         -    65%    86%  1.00x  ONLINE  -

I think I should increase the recordsize and volblocksize on the target if it has a higher ashift value and if I do not want to waste a lot of space.

 

[LnxBil]

Yes, that makes a difference again (Note for other readers: you can only change the volblocksize by 'dd'ing onto a new volume with that blocksize):

NAME                                       USED  REFER  LUSED  LREFER  RATIO  COMPRESS VOLBLOCKS
ashift-08/test-volblocksize-8K             685M   685M  1,72G   1,72G  2.60x        on        8K
ashift-08/test-volblocksize-128K           516M   516M  1,71G   1,71G  3.39x        on      128K
ashift-08/test-volblocksize-128K-gzip-9    361M   361M  1,71G   1,71G  4.86x    gzip-9      128K
ashift-08/test-volblocksize-512K-gzip-9    352M   352M  1,71G   1,71G  4.97x    gzip-9      512K
ashift-12/test-volblocksize-8k             939M   939M  1,72G   1,72G  1.90x      gzip        8K
ashift-12/test-volblocksize-4K            1,65G  1,65G  1,73G   1,73G  1.06x      gzip        4K
ashift-12/test-volblocksize-128k           389M   389M  1,71G   1,71G  4.51x      gzip      128K