On Mon, Feb 03, 2020 at 04:08:16PM -0800, Alan Latteri wrote: > You are correct. There is an LVM layer on top made with the following commands: > > pvcreate /dev/sda > vgcreate chwumbo /dev/sda > lvcreate -l 100%FREE -n data chwumbo > > Running mkfs.xfs against raw block device produces the following result. > > mkfs.xfs -f -d su=2560k,sw=5 /dev/sda > mkfs.xfs: Specified data stripe unit 5120 is not the same as the volume stripe unit 512 > mkfs.xfs: Specified data stripe width 25600 is not the same as the volume stripe width 512 Ok, so the device firmware is ... lacking in functionality. It's not telling the OS it's layout information through the VPD inquiry page that specified to provide this info to the OS. That's a vendor issue, nothing we can do about it. What it means, though, is that LVM doesn't know it's suppose to align itself to the underlying device, and neither does mkfs.xfs. So if you are going to use LVM, you need to configure it with the correct physical layout information, and then it should export that to the filesystem automatically. If you can't configure the LVM layout exactly, then you'll have to ensure that the LVM devices are correctly aligned yourself and then manually configure the filesystem (as per above). You may end up being better off exporting 5 RAID-6 luns from the hardware RAID and then RAID-0 striping them in the OS using md, dm or lvm... Cheers, Dave. -- Dave Chinner david@xxxxxxxxxxxxx
