I would like to avoid setting up a file server as back-end to my web server cluster, as this would give a single point of failure. Give we are running a iSCSI array of 8 nodes with redundancy and replication etc, I would like to take advantage of this.

It's the first time I'll be working with iSCSI from within Linux. A) is it possible and b) if it is, how do I best configure multiple machine to use the same iSCSI LUN / data store, as a "common share " . I assume I need a more advance file system that ext3.

iSCSI doesn't generally [1] work that way. You can't mount a volume in more than one place at once. That's the other difference between a "SAN" and a "NAS" [2]. If you have a storage appliance like a NetApp filer, there's a chance you can say "these disks are an iSCSI target", "these disks are NFS". A solution is to mount the LUN on one machine, and cluster that machine with a second using Heartbeat and a floating IP. Craig [1] Without black-magic clustering such as http://blogs.msdn.com/b/clustering/archive/2009/02/19/9433146.aspx. [2] Other than the letters being backwards, of course.

Hi On 10/12/2010, at 10:58 AM, Gregory Machin <clubbing80s(a)gmail.com> wrote:

Hi.

I would like to avoid setting up a file server as back-end to my web server cluster, as this would give a single point of failure. Give we are running a iSCSI array of 8 nodes with redundancy and replication etc, I would like to take advantage of this.

It's the first time I'll be working with iSCSI from within Linux. A) is it possible and b) if it is, how do I best configure multiple machine to use the same iSCSI LUN / data store, as a "common share " . I assume I need a more advance file system that ext3.

If you share a block device with more than one computer you need a cluster file system. I suggest you look at GFS, which is part of Redhat and also in fedora among others. You may also find multipath useful if the backend block devices are highly available.