[Lustre-discuss] number of inodes in zfs MDT

[Isaac Huang]

On Wed, Jun 18, 2014 at 06:11:33AM -0400, Anjana Kar wrote:
> ......
> Instead we have moved to ldiskfs MDT and zfs OSTs, with the same lustre/zfs
> versions, and have a lot more inodes available.
> 
> Filesystem            Inodes   IUsed   IFree IUse% Mounted on
> x.x.x.x at o2ib:/iconfs
>                      39049920 7455386 31594534   20% /iconfs

Since ZFS doesn't create [iz]nodes statically, in statfs it simply
estimates the free inodes as availbytes >> 9, see zfs_statvfs().

So I'd guess that the difference in reported free inodes was caused by
differences in space efficiency.

I don't know about ldiskfs but two things come to mind about ZFS:
1. By default ZFS stores two copies (a.k.a. ditto blocks) of file
system metadata, in addition to whatever replication the pool
already has. As a result, unless ldiskfs does something similar, ZFS
would be only 50% space efficient as ldiskfs, since the work load was
mostly metadata. An easy way to verify this would be to compare the
reported free space of ZFS and ldiskfs.

2. ZFS might be using a bigger sector size for the disks. Some drives
report 512 bytes for compatibility while they are truly 4K, but ZFS
has a built-in list of devices that lie to override the answer from
the drives. This may also contribute to space efficiency of ZFS.


-Isaac