From Linux NFS

File Layout

Accessing a storage system with pNFS

Step 0: Get pNFS kernel from the PNFS Development Git tree and install on all involved servers.

Step 1: Setup NFSv4 servers

• Create /etc/exports file on all DS's and the MDS.

  /export  *(rw,sync,fsid=0,insecure,no_subtree_check)
  

  Note: starting with 2.6.32-rc1, the "pnfs" export option is required.

  /export  *(rw,sync,fsid=0,insecure,no_subtree_check,pnfs)
  

  Until support for the "pnfs" export option is publicly released, please build and install exportfs, rpc.mountd, rpc.nfsd, and optionally, nfsstat from:

  git://linux-nfs.org/~bhalevy/pnfs-nfs-utils.git
  
  To install just the required binaries:
  cp utils/exportfs/exportfs /usr/sbin/exportfs
  cp utils/mountd/mountd /usr/sbin/rpc.mountd
  cp utils/mountd/nfsd /usr/sbin/rpc.nfsd
  cp utils/nfsstat/nfsstat /usr/sbin/nfsstat
  

• Tell the metadata server the IP addresses of the data servers:

  echo "/dev/sdc:192.168.0.1,192.168.0.2"
  

  (replacing /dev/sdc by the name of the device hosting the exported gfs2 filesystem, and the ip addresses by the ip addresses of your data servers.)

• If the nfs service requires starting then run the following commands on the metadata server

  /etc/init.d/nfs restart
  
  or
  
  rpc.mountd
  rpc.nfsd 8
  exportfs -r
  

Step 2: Load Layout driver on clients

modprobe nfs_layout_nfsv41_files

Step 3: Mount pNFS file system.

• On pNFS client:

  mount -t nfs4 -o minorversion=1 <mds_server>:/ /mnt/pnfs
  
  Note: Each file system has its own way of picking the MDS.  Ensure you mount only the MDS and not a DS.
  

Debugging Help

• NFS debugging:

  echo 32767 > /proc/sys/sunrpc/nfsd_debug
  echo 32767 > /proc/sys/sunrpc/nfs_debug