Hi,
I am trying to get the most out of my 40TB xfs file system and I have noticed that the inode64 mount option gives me a roughly 30% performance increase (besides the other useful things). The problem is that I have to export the filesystem via NFS and I cannot seem to get this working with the current version of nfs-utils (1.2.3).
The export option fsid=uuid cannot be used (the standard nfs-utils does accept this option). Also I am not exporting the root of the filesystem, but a couple of subdirs. If I mount the subdirs on the clients it seems to go alright. But when I do e.g. an 'ls' I get a 'Stale NFS handle' error.
Updating nfs-utils seems not so trivial so I would like to skip this option :-)
How can I export this filesystem with inode64 enabled and mount it on my clients? All clients are Centos6 64bits as is the server.
Cheers, Maarten van Ingen
On 2013-06-19, Maarten van Ingen maarten.vaningen@surfsara.nl wrote:
I am trying to get the most out of my 40TB xfs file system and I have noticed that the inode64 mount option gives me a roughly 30% performance increase (besides the other useful things). The problem is that I have to export the filesystem via NFS and I cannot seem to get this working with the current version of nfs-utils (1.2.3).
The export option fsid=uuid cannot be used (the standard nfs-utils does accept this option). Also I am not exporting the root of the filesystem, but a couple of subdirs. If I mount the subdirs on the clients it seems to go alright. But when I do e.g. an 'ls' I get a 'Stale NFS handle' error.
Updating nfs-utils seems not so trivial so I would like to skip this option :-)
How can I export this filesystem with inode64 enabled and mount it on my clients? All clients are Centos6 64bits as is the server.
I've found a few workarounds, but nothing really compelling.
First, you can try exporting the filesystem root instead if you're just using NFSv3. (I don't believe this will work with NFSv4.) That might mean you have to reorganize some mount points, but it might be worth the effort if you're banging your head against the wall.
Second, you can try this pretty crazy strategy suggested on the XFS list:
http://oss.sgi.com/archives/xfs/2009-11/msg00161.html
Third, you might need to verify that your kernel supports the nfs.enable_ino64 option:
http://patchwork.xfs.org/patch/609/
You could try passing it explicitly as a kernel boot option.
--keith