On 17 Oct 2014 13:41, "Isaac Huang" <he.huang@intel.com> wrote:

On Thu, Oct 16, 2014 at 09:26:59PM +0000, Dilger, Andreas wrote:
> On 2014/10/16, 9:10 AM, "Kurt Strosahl" <strosahl@jlab.org> wrote:
>
> >Good Morning,
> >
> >    Has anyone out there been using zfs as the back end for lustre?  I am
> >curious as to how you set up raid-z when setting up an ost.
>
> Two options exist:
> - specify raidz/raidz2 on mkfs.lustre command line (see man page):
>
>  mkfs.lustre {other opts} --ost --backfstype=zfs mypool/testfs-OST0000 \
>    raidz2 /dev/sd[a-j] raidz2 /dev/sd[k-t] raidz2 /dev/sd[u-z]
> /dev/sda[a-d]
>
> would, for example, create three RAID-Z2 VDEVs with 10 disks each.  Of
> course the use of /dev/sd* device names is frowned upon in real life ZFS
> usage, but are shorter for this example.
>
> - create RAID-Z/Z2 datasets normally, then just point mkfs.lustre at them:
>
>  mkfs.lustre {other opts} --ost --backfstype=zfs mypool/testfs-OST0000
>
> IIRC you have to disable automatic mounting on the pool (legacy mode).

The xattr=sa option is also needed, for performance at least, if the
dataset is not created by mkfs.lustre.

-Isaac
_______________________________________________
HPDD-discuss mailing list
HPDD-discuss@lists.01.org
https://lists.01.org/mailman/listinfo/hpdd-discuss