For license reasons (god damn GPL), ZFS can't be integrated into the Linux kernel, but there is a userland fuse version called ZFS-FUSE, so I tried to measure the fuse overhead vs others filesystems presents on my server.

This tests are not benchmarks, but just show that ZFS-FUSE is fast enough and seems to be stable (at least for my personal server)

  • ZFS with lzjb compression
    • #  dd if=/dev/zero of=/testzfs/toto bs=2048 count=200000 
409600000 bytes (410 MB) copied, 9.93444 s, 41.2 MB/s

#  dd if=/dev/zero of=/testzfs/toto bs=2048 count=200000 
409600000 bytes (410 MB) copied, 9.88071 s, 41.5 MB/s

#  dd if=/dev/zero of=/testzfs/toto bs=2048 count=200000 
409600000 bytes (410 MB) copied, 10.2585 s, 39.9 MB/s
  • ZFS with compression off
    • #  dd if=/dev/zero of=/testzfs/toto bs=2048 count=200000 
409600000 bytes (410 MB) copied, 17.2505 s, 23.7 MB/s

#  dd if=/dev/zero of=/testzfs/toto bs=2048 count=200000 
409600000 bytes (410 MB) copied, 17.0921 s, 24.0 MB/s

#  dd if=/dev/zero of=/testzfs/toto bs=2048 count=200000 
409600000 bytes (410 MB) copied, 16.8956 s, 24.2 MB/s
  • EXT3
    • #  dd if=/dev/zero of=/toto bs=2048 count=200000 
409600000 bytes (410 MB) copied, 9.92158 s, 41.3 MB/s

#  dd if=/dev/zero of=/toto bs=2048 count=200000 
409600000 bytes (410 MB) copied, 8.50674 s, 48.2 MB/s

#  dd if=/dev/zero of=/toto bs=2048 count=200000 
409600000 bytes (410 MB) copied, 9.02419 s, 45.4 MB/s
  • REISERFS (the way reiserfs works makes this big file creation faster)
    • #  dd if=/dev/zero of=/opt/toto bs=2048 count=200000 
09600000 bytes (410 MB) copied, 2.29901 s, 178 MB/s

#  dd if=/dev/zero of=/opt/toto bs=2048 count=200000 
409600000 bytes (410 MB) copied, 2.1869 s, 187 MB/s

#  dd if=/dev/zero of=/opt/toto bs=2048 count=200000 
409600000 bytes (410 MB) copied, 1.9057 s, 215 MB/s

    Also remember that FreeBSD 8.1 now includes ZFS version 14, which is a good (native) alternative