I took a look at the documents and they explain some of the questions I had.
Seems like coordination occurs via the LAN and not the disk/SAN interface.
I'm wondering why? From what I remember, the GFS people kept the locks on
the drives themselves.
[mailto:linux-xfs-bounce@xxxxxxxxxxx]On Behalf Of Mike Gigante
Sent: Thursday, December 05, 2002 1:57 PM
To: Christoph Hellwig
Cc: N. Richard Solis; Craig Tierney; 'Ray Muno '; 'Steve Lord ';
Subject: Re: Clustered XFS?
On Thu, Dec 05, 2002 at 12:54:30PM -0500, N. Richard Solis wrote:
> Sounds more like a Clustered NFS solution to me? Would I be on target
> that one?
No, In NFS data travels over the wire, in CXFS data is direct access via the
SAN. NFS is not posix compliant, CXFS is.
Think of CXFS as a local filesystem for all intensive puposes - both
performance and semantics.
The client/server distinction is only with respect to maintaining a
coherent cluster-wide view of filesystem by providing consistency of
metadata. There are some things that must be coordinated such as extent
allocation, kernel filesystem caches, memory mapped files etc etc. The
CXFS server is the gatekeeper in this respect.
Apart from the consistency management both the CXFS client and server
can be considered local XFS.
Oh, and there is no single point of failure -- clients can become the
server (if they are capable).
There are some docs on the SGI website that describe CXFS. Check that out
if you want more info.