[Ilugc] Question about RedHat clustering (GFS vs NFS) - Documentation
- From: rajasuperman@xxxxxxxxx (Raja Subramanian)
- Date: Sun Nov 28 12:38:01 2010
On Sun, Nov 28, 2010 at 6:17 AM, Ashish Verma <ilugid@xxxxxxxxx> wrote:
They have shown GFS servers over iScsi servers and mentioned the benefits of
single file-system over the multiple block device paths thereby providing
redundancy. However cant the same be achieved with NFS by placing nfs over
the multiple iscsi servers and still you would get the same results.
On the NFS servers, how will you mount the shared iSCSI volume?
Which FS will you use to format the iSCSI volume? Ext3/etc are not
cluster aware, ie, you cannot mount the same block device from
multiple servers at the same time.
NFS is not a true file system -- it's only the means of exporting a
backend FS on to the network. Unless you have a cluster aware
FS on the backend which is shared across multiple servers, you
will end up with independent storage islands for each NFS server.
It is also not best practice to take a backend cluster FS and re-export
it using NFS/CIFS. This typically breaks Unix file locking semantics.
Best option is to use cluster aware file systems like GFS, Luster,
Coda, StorNext (proprietary) end to end.
NFS has addressed this lack of parallelism in version 4.1 -- see pNFS.
pNFS provides multiple data paths for the block access to increase
performance, while meta data is on a single server as before. You still
need a shared backend storage and FS anyway.
- Raja
Other related posts:
