AdminSysGNULinux

# ls -l /dev/mapper/ | grep part | awk '{print $9}'
3624a937043be47c12334399b00016d73-part1
3624a937043be47c12334399b00016d74-part1
3624a937043be47c12334399b00016d75-part1
# parts=`ls -l /dev/mapper/ | grep part | awk '{print $9}'`
# for i in $parts
> do
> mkfs.xfs -d su=131072,sw=8 -i size=1024 $i
> done

You can verify that the underlying filesystem has the correct values for stripe unit and stripe width by using the xfs_info command:

mount -t xfs -o nobarrier,noatime,nodiratime,inode64 /dev/mapper/3624a937043be47c12334399b00016d73-part1 /disk1
# xfs_info /disk1
meta-data=/dev/mapper/3624a937043be47c12334399b00016d73-part1 isize=1024   agcount=4, agsize=83623808 blks
         =                       sectsz=512   attr=2
data     =                       bsize=4096   blocks=334495232, imaxpct=5
         =                       sunit=32     swidth=256 blks
naming   =version 2              bsize=4096   ascii-ci=0
log      =internal               bsize=4096   blocks=163328, version=2
         =                       sectsz=512   sunit=32 blks, lazy-count=1
realtime =none                   extsz=4096   blocks=0, rtextents=0

sunit and swidth are shown in bsize (block size) units in the xfs_info command output.

stripe unit= 32 sunits * 4096 bsize (block size)= 131072 bytes = 128K
stripe width= 256 blocks * 4096 bsize = 1M = 128K * 8 drives

The sysctl fs.xfs.rotorstep can be used to change how many files are put into an XFS allocation group. Increasing the default number from 1 to 255 reduces seeks to multiple allocation groups. Improved performance has been observed in some cases by increasing this number. You can put the following line in /etc/sysctl.conf to ensure this change is affected on each boot of the system:

fs.xfs.rotorstep = 255

When mounting the XFS filesystem that resides on the LUNs offered from the FlashArray, be sure to use the following mount options:

mount –t xfs –o “discard,nobarrier,noatime,nodiratime,inode64” \
/dev/mapper/nodeX /srv/node/sdb1

Source : https://pure-storage-openstack-docs.readthedocs.io/en/wallaby/swift/section_swift-partitioning.html

The overlay2 driver is supported on xfs backing filesystems, but only with d_type=true enabled.

Autres /mkfs.xfs -m reflink=1 /dev/sdc

mkfs.xfs -n ftype=1 -m reflink=1 /dev/mapper/vg_data-data

Lazy pulling of container images can run more efficiently when the file system has reflink support. The file systems XFS and BTRFS have reflink support. Source : podman