[zfs-discuss] RAIDZ-2 disk space

Alexander I Kulyavtsev aik at fnal.gov
Thu Apr 5 17:48:40 EDT 2018


You have to keep about 8% free space rather than 20% recommended for older versions of zfs.

The metaslab allocator changes free block algorithm search when 4% of space left and that can be slow;
http://open-zfs.org/wiki/Performance_tuning#Metaslab_Allocator
I do not recall right away where the other 4% come from.

Also, you may try different settings for metaslab_lba_weighting_enabled for SSD pool described at link above. 
Disabling lba_weighting will distribute data more evenly between slabs. It may also affect fragmentation in zfs and garbage collection time on SSD.
This is global module setting so it shall not be used when there is other HDD zpool. 

Alex.

On 4/5/18, 4:08 PM, "zfs-discuss on behalf of Gandalf Corvotempesta via zfs-discuss" <zfs-discuss-bounces at list.zfsonlinux.org on behalf of zfs-discuss at list.zfsonlinux.org> wrote:

    The 647GB available, can be used fully or should I reserve about 20%
    for ZFS? (i've read somewhere that is not suggested to fill ZFS
    totally)





More information about the zfs-discuss mailing list