[rhelv6-list] Distributed parallel fault-tolerant file systems

[Jan-Frode Myklebust]

On Mon, Mar 05, 2012 at 02:11:28PM -0500, Prentice Bisbal wrote:
> 
> I specialize in HPC,

I'm an ex-HPC admin, now more a server/HA-admin, and have brought GPFS
with me from the HPC days. It's a great tool for creating highly
available scale out clusters, and has never failed on me.

The OP was asking for redundant storage via nfs and cifs, and this is
something GPFS has created a solution for (and they package it into IBM
SONAS boxes):

	http://www.samba.org/~tridge/ctdb.pdf
	http://www.samba.org/~tridge/ctdb_movies/ (videos showing snapshots used in windows explorer)

Clustered NFS configuration:

	http://publib.boulder.ibm.com/infocenter/clresctr/vxrx/index.jsp?topic=%2Fcom.ibm.cluster.gpfs.v3r4.gpfs200.doc%2Fbl1adv_cnfs.html

I couldn't find the CIFS configuration in the admin guide, but think it
should be there.. otherwise the wiki.samba.org has configuration
examples for both GPFS, lustre, GFS, ocfs2 and glusterfs:

	http://wiki.samba.org/index.php/CTDB_Setup


> GPFS is a commercial product, and an expensive one at that,
> but I know plenty of people who've decided
> it's worth the cost (or really, really want someone else to blame when
> things go south. 

Yes, the licensing model for GPFS seems tuned for HPC-clusters with few
"server" nodes. But I believe they'll give large discounts to stay
competitive.. 


  -jf